Merge branch 'hn/reftable'

The "reftable" backend for the refs API, without integrating into
the refs subsystem, has been added.

* hn/reftable:
  Add "test-tool dump-reftable" command.
  reftable: add dump utility
  reftable: implement stack, a mutable database of reftable files.
  reftable: implement refname validation
  reftable: add merged table view
  reftable: add a heap-based priority queue for reftable records
  reftable: reftable file level tests
  reftable: read reftable files
  reftable: generic interface to tables
  reftable: write reftable files
  reftable: a generic binary tree implementation
  reftable: reading/writing blocks
  Provide zlib's uncompress2 from compat/zlib-compat.c
  reftable: (de)serialization for the polymorphic record type.
  reftable: add blocksource, an abstraction for random access reads
  reftable: utility functions
  reftable: add error related functionality
  reftable: add LICENSE
  hash.h: provide constants for the hash IDs

69 files changed, 11125 insertions, 12 deletions

diff --git a/Makefile b/Makefile
index ed75ed4..9c00a79 100644
--- a/Makefile
+++ b/Makefile
@@ -256,6 +256,8 @@ all::
 #
 # Define NO_DEFLATE_BOUND if your zlib does not have deflateBound.
 #
+# Define NO_UNCOMPRESS2 if your zlib does not have uncompress2.
+#
 # Define NO_NORETURN if using buggy versions of gcc 4.6+ and profile feedback,
 # as the compiler can crash (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49299)
 #
@@ -732,6 +734,7 @@ TEST_BUILTINS_OBJS += test-read-cache.o
 TEST_BUILTINS_OBJS += test-read-graph.o
 TEST_BUILTINS_OBJS += test-read-midx.o
 TEST_BUILTINS_OBJS += test-ref-store.o
+TEST_BUILTINS_OBJS += test-reftable.o
 TEST_BUILTINS_OBJS += test-regex.o
 TEST_BUILTINS_OBJS += test-repository.o
 TEST_BUILTINS_OBJS += test-revision-walking.o
@@ -810,6 +813,8 @@ TEST_SHELL_PATH = $(SHELL_PATH)
 
 LIB_FILE = libgit.a
 XDIFF_LIB = xdiff/lib.a
+REFTABLE_LIB = reftable/libreftable.a
+REFTABLE_TEST_LIB = reftable/libreftable_test.a
 
 GENERATED_H += command-list.h
 GENERATED_H += config-list.h
@@ -1189,7 +1194,7 @@ THIRD_PARTY_SOURCES += compat/regex/%
 THIRD_PARTY_SOURCES += sha1collisiondetection/%
 THIRD_PARTY_SOURCES += sha1dc/%
 
-GITLIBS = common-main.o $(LIB_FILE) $(XDIFF_LIB)
+GITLIBS = common-main.o $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB)
 EXTLIBS =
 
 GIT_USER_AGENT = git/$(GIT_VERSION)
@@ -1720,6 +1725,11 @@ ifdef NO_DEFLATE_BOUND
 	BASIC_CFLAGS += -DNO_DEFLATE_BOUND
 endif
 
+ifdef NO_UNCOMPRESS2
+	BASIC_CFLAGS += -DNO_UNCOMPRESS2
+	REFTABLE_OBJS += compat/zlib-uncompress2.o
+endif
+
 ifdef NO_POSIX_GOODIES
 	BASIC_CFLAGS += -DNO_POSIX_GOODIES
 endif
@@ -2431,7 +2441,36 @@ XDIFF_OBJS += xdiff/xutils.o
 .PHONY: xdiff-objs
 xdiff-objs: $(XDIFF_OBJS)
 
+REFTABLE_OBJS += reftable/basics.o
+REFTABLE_OBJS += reftable/error.o
+REFTABLE_OBJS += reftable/block.o
+REFTABLE_OBJS += reftable/blocksource.o
+REFTABLE_OBJS += reftable/iter.o
+REFTABLE_OBJS += reftable/publicbasics.o
+REFTABLE_OBJS += reftable/merged.o
+REFTABLE_OBJS += reftable/pq.o
+REFTABLE_OBJS += reftable/reader.o
+REFTABLE_OBJS += reftable/record.o
+REFTABLE_OBJS += reftable/refname.o
+REFTABLE_OBJS += reftable/generic.o
+REFTABLE_OBJS += reftable/stack.o
+REFTABLE_OBJS += reftable/tree.o
+REFTABLE_OBJS += reftable/writer.o
+
+REFTABLE_TEST_OBJS += reftable/basics_test.o
+REFTABLE_TEST_OBJS += reftable/block_test.o
+REFTABLE_TEST_OBJS += reftable/dump.o
+REFTABLE_TEST_OBJS += reftable/merged_test.o
+REFTABLE_TEST_OBJS += reftable/pq_test.o
+REFTABLE_TEST_OBJS += reftable/record_test.o
+REFTABLE_TEST_OBJS += reftable/readwrite_test.o
+REFTABLE_TEST_OBJS += reftable/refname_test.o
+REFTABLE_TEST_OBJS += reftable/stack_test.o
+REFTABLE_TEST_OBJS += reftable/test_framework.o
+REFTABLE_TEST_OBJS += reftable/tree_test.o
+
 TEST_OBJS := $(patsubst %$X,%.o,$(TEST_PROGRAMS)) $(patsubst %,t/helper/%,$(TEST_BUILTINS_OBJS))
+
 .PHONY: test-objs
 test-objs: $(TEST_OBJS)
 
@@ -2447,6 +2486,8 @@ OBJECTS += $(PROGRAM_OBJS)
 OBJECTS += $(TEST_OBJS)
 OBJECTS += $(XDIFF_OBJS)
 OBJECTS += $(FUZZ_OBJS)
+OBJECTS += $(REFTABLE_OBJS) $(REFTABLE_TEST_OBJS)
+
 ifndef NO_CURL
 	OBJECTS += http.o http-walker.o remote-curl.o
 endif
@@ -2589,6 +2630,12 @@ $(LIB_FILE): $(LIB_OBJS)
 $(XDIFF_LIB): $(XDIFF_OBJS)
 	$(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^
 
+$(REFTABLE_LIB): $(REFTABLE_OBJS)
+	$(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^
+
+$(REFTABLE_TEST_LIB): $(REFTABLE_TEST_OBJS)
+	$(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^
+
 export DEFAULT_EDITOR DEFAULT_PAGER
 
 Documentation/GIT-EXCLUDED-PROGRAMS: FORCE
@@ -2887,7 +2934,7 @@ perf: all
 
 t/helper/test-tool$X: $(patsubst %,t/helper/%,$(TEST_BUILTINS_OBJS))
 
-t/helper/test-%$X: t/helper/test-%.o GIT-LDFLAGS $(GITLIBS)
+t/helper/test-%$X: t/helper/test-%.o GIT-LDFLAGS $(GITLIBS) $(REFTABLE_TEST_LIB)
 	$(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) $(filter %.a,$^) $(LIBS)
 
 check-sha1:: t/helper/test-tool$X
@@ -3225,7 +3272,7 @@ cocciclean:
 clean: profile-clean coverage-clean cocciclean
 	$(RM) *.res
 	$(RM) $(OBJECTS)
-	$(RM) $(LIB_FILE) $(XDIFF_LIB)
+	$(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB)
 	$(RM) $(ALL_PROGRAMS) $(SCRIPT_LIB) $(BUILT_INS) git$X
 	$(RM) $(TEST_PROGRAMS)
 	$(RM) $(FUZZ_PROGRAMS)
diff --git a/ci/lib.sh b/ci/lib.sh
index 82cb17f..994050f 100755
--- a/ci/lib.sh
+++ b/ci/lib.sh
@@ -224,6 +224,7 @@ linux-gcc-default)
 	;;
 Linux32)
 	CC=gcc
+	MAKEFLAGS="$MAKEFLAGS NO_UNCOMPRESS2=1"
 	;;
 linux-musl)
 	CC=gcc
diff --git a/compat/.gitattributes b/compat/.gitattributes
new file mode 100644
index 0000000..40dbfb1
--- /dev/null
+++ b/compat/.gitattributes
@@ -0,0 +1 @@
+/zlib-uncompress2.c	whitespace=-indent-with-non-tab,-trailing-space
diff --git a/compat/zlib-uncompress2.c b/compat/zlib-uncompress2.c
new file mode 100644
index 0000000..722610b
--- /dev/null
+++ b/compat/zlib-uncompress2.c
@@ -0,0 +1,95 @@
+/* taken from zlib's uncompr.c
+
+   commit cacf7f1d4e3d44d871b605da3b647f07d718623f
+   Author: Mark Adler <madler@alumni.caltech.edu>
+   Date:   Sun Jan 15 09:18:46 2017 -0800
+
+       zlib 1.2.11
+
+*/
+
+#include "../reftable/system.h"
+#define z_const
+
+/*
+ * Copyright (C) 1995-2003, 2010, 2014, 2016 Jean-loup Gailly, Mark Adler
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include <zlib.h>
+
+/* clang-format off */
+
+/* ===========================================================================
+     Decompresses the source buffer into the destination buffer.  *sourceLen is
+   the byte length of the source buffer. Upon entry, *destLen is the total size
+   of the destination buffer, which must be large enough to hold the entire
+   uncompressed data. (The size of the uncompressed data must have been saved
+   previously by the compressor and transmitted to the decompressor by some
+   mechanism outside the scope of this compression library.) Upon exit,
+   *destLen is the size of the decompressed data and *sourceLen is the number
+   of source bytes consumed. Upon return, source + *sourceLen points to the
+   first unused input byte.
+
+     uncompress returns Z_OK if success, Z_MEM_ERROR if there was not enough
+   memory, Z_BUF_ERROR if there was not enough room in the output buffer, or
+   Z_DATA_ERROR if the input data was corrupted, including if the input data is
+   an incomplete zlib stream.
+*/
+int ZEXPORT uncompress2 (
+    Bytef *dest,
+    uLongf *destLen,
+    const Bytef *source,
+    uLong *sourceLen) {
+    z_stream stream;
+    int err;
+    const uInt max = (uInt)-1;
+    uLong len, left;
+    Byte buf[1];    /* for detection of incomplete stream when *destLen == 0 */
+
+    len = *sourceLen;
+    if (*destLen) {
+	left = *destLen;
+	*destLen = 0;
+    }
+    else {
+	left = 1;
+	dest = buf;
+    }
+
+    stream.next_in = (z_const Bytef *)source;
+    stream.avail_in = 0;
+    stream.zalloc = (alloc_func)0;
+    stream.zfree = (free_func)0;
+    stream.opaque = (voidpf)0;
+
+    err = inflateInit(&stream);
+    if (err != Z_OK) return err;
+
+    stream.next_out = dest;
+    stream.avail_out = 0;
+
+    do {
+	if (stream.avail_out == 0) {
+	    stream.avail_out = left > (uLong)max ? max : (uInt)left;
+	    left -= stream.avail_out;
+	}
+	if (stream.avail_in == 0) {
+	    stream.avail_in = len > (uLong)max ? max : (uInt)len;
+	    len -= stream.avail_in;
+	}
+	err = inflate(&stream, Z_NO_FLUSH);
+    } while (err == Z_OK);
+
+    *sourceLen -= len + stream.avail_in;
+    if (dest != buf)
+	*destLen = stream.total_out;
+    else if (stream.total_out && err == Z_BUF_ERROR)
+	left = 1;
+
+    inflateEnd(&stream);
+    return err == Z_STREAM_END ? Z_OK :
+	   err == Z_NEED_DICT ? Z_DATA_ERROR  :
+	   err == Z_BUF_ERROR && left + stream.avail_out ? Z_DATA_ERROR :
+	   err;
+}
diff --git a/config.mak.uname b/config.mak.uname
index d0701f9..a3a7793 100644
--- a/config.mak.uname
+++ b/config.mak.uname
@@ -261,6 +261,10 @@ ifeq ($(uname_S),FreeBSD)
 	FILENO_IS_A_MACRO = UnfortunatelyYes
 endif
 ifeq ($(uname_S),OpenBSD)
+	# Versions < 7.0 need compatibility layer
+	ifeq ($(shell expr "$(uname_R)" : "[1-6]\."),2)
+		NO_UNCOMPRESS2 = UnfortunatelyYes
+	endif
 	NO_STRCASESTR = YesPlease
 	NO_MEMMEM = YesPlease
 	USE_ST_TIMESPEC = YesPlease
@@ -516,6 +520,7 @@ ifeq ($(uname_S),Interix)
 	endif
 endif
 ifeq ($(uname_S),Minix)
+	NO_UNCOMPRESS2 = YesPlease
 	NO_IPV6 = YesPlease
 	NO_ST_BLOCKS_IN_STRUCT_STAT = YesPlease
 	NO_NSEC = YesPlease
diff --git a/configure.ac b/configure.ac
index 5ee25ec..d60d494 100644
--- a/configure.ac
+++ b/configure.ac
@@ -664,9 +664,22 @@ AC_LINK_IFELSE([ZLIBTEST_SRC],
 	NO_DEFLATE_BOUND=yes])
 LIBS="$old_LIBS"
 
+AC_DEFUN([ZLIBTEST_UNCOMPRESS2_SRC], [
+AC_LANG_PROGRAM([#include <zlib.h>],
+ [uncompress2(NULL,NULL,NULL,NULL);])])
+AC_MSG_CHECKING([for uncompress2 in -lz])
+old_LIBS="$LIBS"
+LIBS="$LIBS -lz"
+AC_LINK_IFELSE([ZLIBTEST_UNCOMPRESS2_SRC],
+	[AC_MSG_RESULT([yes])],
+	[AC_MSG_RESULT([no])
+	NO_UNCOMPRESS2=yes])
+LIBS="$old_LIBS"
+
 GIT_UNSTASH_FLAGS($ZLIB_PATH)
 
 GIT_CONF_SUBST([NO_DEFLATE_BOUND])
+GIT_CONF_SUBST([NO_UNCOMPRESS2])
 
 #
 # Define NEEDS_SOCKET if linking with libc is not enough (SunOS,
diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt
index be67b4d..5100f56 100644
--- a/contrib/buildsystems/CMakeLists.txt
+++ b/contrib/buildsystems/CMakeLists.txt
@@ -647,6 +647,12 @@ parse_makefile_for_sources(libxdiff_SOURCES "XDIFF_OBJS")
 list(TRANSFORM libxdiff_SOURCES PREPEND "${CMAKE_SOURCE_DIR}/")
 add_library(xdiff STATIC ${libxdiff_SOURCES})
 
+#reftable
+parse_makefile_for_sources(reftable_SOURCES "REFTABLE_OBJS")
+
+list(TRANSFORM reftable_SOURCES PREPEND "${CMAKE_SOURCE_DIR}/")
+add_library(reftable STATIC ${reftable_SOURCES})
+
 if(WIN32)
 	if(NOT MSVC)#use windres when compiling with gcc and clang
 		add_custom_command(OUTPUT ${CMAKE_BINARY_DIR}/git.res
@@ -669,7 +675,7 @@ endif()
 #link all required libraries to common-main
 add_library(common-main OBJECT ${CMAKE_SOURCE_DIR}/common-main.c)
 
-target_link_libraries(common-main libgit xdiff ${ZLIB_LIBRARIES})
+target_link_libraries(common-main libgit xdiff reftable ${ZLIB_LIBRARIES})
 if(Intl_FOUND)
 	target_link_libraries(common-main ${Intl_LIBRARIES})
 endif()
@@ -908,11 +914,15 @@ if(BUILD_TESTING)
 add_executable(test-fake-ssh ${CMAKE_SOURCE_DIR}/t/helper/test-fake-ssh.c)
 target_link_libraries(test-fake-ssh common-main)
 
+#reftable-tests
+parse_makefile_for_sources(test-reftable_SOURCES "REFTABLE_TEST_OBJS")
+list(TRANSFORM test-reftable_SOURCES PREPEND "${CMAKE_SOURCE_DIR}/")
+
 #test-tool
 parse_makefile_for_sources(test-tool_SOURCES "TEST_BUILTINS_OBJS")
 
 list(TRANSFORM test-tool_SOURCES PREPEND "${CMAKE_SOURCE_DIR}/t/helper/")
-add_executable(test-tool ${CMAKE_SOURCE_DIR}/t/helper/test-tool.c ${test-tool_SOURCES})
+add_executable(test-tool ${CMAKE_SOURCE_DIR}/t/helper/test-tool.c ${test-tool_SOURCES} ${test-reftable_SOURCES})
 target_link_libraries(test-tool common-main)
 
 set_target_properties(test-fake-ssh test-tool
diff --git a/contrib/buildsystems/Generators/Vcxproj.pm b/contrib/buildsystems/Generators/Vcxproj.pm
index d258445..1a25789 100644
--- a/contrib/buildsystems/Generators/Vcxproj.pm
+++ b/contrib/buildsystems/Generators/Vcxproj.pm
@@ -77,7 +77,7 @@ sub createProject {
     my $libs_release = "\n    ";
     my $libs_debug = "\n    ";
     if (!$static_library) {
-      $libs_release = join(";", sort(grep /^(?!libgit\.lib|xdiff\/lib\.lib|vcs-svn\/lib\.lib)/, @{$$build_structure{"$prefix${name}_LIBS"}}));
+      $libs_release = join(";", sort(grep /^(?!libgit\.lib|xdiff\/lib\.lib|vcs-svn\/lib\.lib|reftable\/libreftable\.lib)/, @{$$build_structure{"$prefix${name}_LIBS"}}));
       $libs_debug = $libs_release;
       $libs_debug =~ s/zlib\.lib/zlibd\.lib/g;
       $libs_debug =~ s/libexpat\.lib/libexpatd\.lib/g;
@@ -232,6 +232,7 @@ EOM
 EOM
     if (!$static_library || $target =~ 'vcs-svn' || $target =~ 'xdiff') {
       my $uuid_libgit = $$build_structure{"LIBS_libgit_GUID"};
+      my $uuid_libreftable = $$build_structure{"LIBS_reftable/libreftable_GUID"};
       my $uuid_xdiff_lib = $$build_structure{"LIBS_xdiff/lib_GUID"};
 
       print F << "EOM";
@@ -241,6 +242,14 @@ EOM
       <ReferenceOutputAssembly>false</ReferenceOutputAssembly>
     </ProjectReference>
 EOM
+      if (!($name =~ /xdiff|libreftable/)) {
+        print F << "EOM";
+    <ProjectReference Include="$cdup\\reftable\\libreftable\\libreftable.vcxproj">
+      <Project>$uuid_libreftable</Project>
+      <ReferenceOutputAssembly>false</ReferenceOutputAssembly>
+    </ProjectReference>
+EOM
+      }
       if (!($name =~ 'xdiff')) {
         print F << "EOM";
     <ProjectReference Include="$cdup\\xdiff\\lib\\xdiff_lib.vcxproj">
diff --git a/hash.h b/hash.h
index 9e25c40..5d40368 100644
--- a/hash.h
+++ b/hash.h
@@ -95,12 +95,18 @@ static inline void git_SHA256_Clone(git_SHA256_CTX *dst, const git_SHA256_CTX *s
 /* Number of algorithms supported (including unknown). */
 #define GIT_HASH_NALGOS (GIT_HASH_SHA256 + 1)
 
+/* "sha1", big-endian */
+#define GIT_SHA1_FORMAT_ID 0x73686131
+
 /* The length in bytes and in hex digits of an object name (SHA-1 value). */
 #define GIT_SHA1_RAWSZ 20
 #define GIT_SHA1_HEXSZ (2 * GIT_SHA1_RAWSZ)
 /* The block size of SHA-1. */
 #define GIT_SHA1_BLKSZ 64
 
+/* "s256", big-endian */
+#define GIT_SHA256_FORMAT_ID 0x73323536
+
 /* The length in bytes and in hex digits of an object name (SHA-256 value). */
 #define GIT_SHA256_RAWSZ 32
 #define GIT_SHA256_HEXSZ (2 * GIT_SHA256_RAWSZ)
diff --git a/object-file.c b/object-file.c
index eac67f6..7ab1446 100644
--- a/object-file.c
+++ b/object-file.c
@@ -165,7 +165,6 @@ static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
 	BUG("trying to finalize unknown hash");
 }
 
-
 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
 	{
 		NULL,
@@ -184,8 +183,7 @@ const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
 	},
 	{
 		"sha1",
-		/* "sha1", big-endian */
-		0x73686131,
+		GIT_SHA1_FORMAT_ID,
 		GIT_SHA1_RAWSZ,
 		GIT_SHA1_HEXSZ,
 		GIT_SHA1_BLKSZ,
@@ -200,8 +198,7 @@ const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
 	},
 	{
 		"sha256",
-		/* "s256", big-endian */
-		0x73323536,
+		GIT_SHA256_FORMAT_ID,
 		GIT_SHA256_RAWSZ,
 		GIT_SHA256_HEXSZ,
 		GIT_SHA256_BLKSZ,
diff --git a/reftable/LICENSE b/reftable/LICENSE
new file mode 100644
index 0000000..402e0f9
--- /dev/null
+++ b/reftable/LICENSE
@@ -0,0 +1,31 @@
+BSD License
+
+Copyright (c) 2020, Google LLC
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+* Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in the
+documentation and/or other materials provided with the distribution.
+
+* Neither the name of Google LLC nor the names of its contributors may
+be used to endorse or promote products derived from this software
+without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/reftable/basics.c b/reftable/basics.c
new file mode 100644
index 0000000..f761e48
--- /dev/null
+++ b/reftable/basics.c
@@ -0,0 +1,128 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "basics.h"
+
+void put_be24(uint8_t *out, uint32_t i)
+{
+	out[0] = (uint8_t)((i >> 16) & 0xff);
+	out[1] = (uint8_t)((i >> 8) & 0xff);
+	out[2] = (uint8_t)(i & 0xff);
+}
+
+uint32_t get_be24(uint8_t *in)
+{
+	return (uint32_t)(in[0]) << 16 | (uint32_t)(in[1]) << 8 |
+	       (uint32_t)(in[2]);
+}
+
+void put_be16(uint8_t *out, uint16_t i)
+{
+	out[0] = (uint8_t)((i >> 8) & 0xff);
+	out[1] = (uint8_t)(i & 0xff);
+}
+
+int binsearch(size_t sz, int (*f)(size_t k, void *args), void *args)
+{
+	size_t lo = 0;
+	size_t hi = sz;
+
+	/* Invariants:
+	 *
+	 *  (hi == sz) || f(hi) == true
+	 *  (lo == 0 && f(0) == true) || fi(lo) == false
+	 */
+	while (hi - lo > 1) {
+		size_t mid = lo + (hi - lo) / 2;
+
+		if (f(mid, args))
+			hi = mid;
+		else
+			lo = mid;
+	}
+
+	if (lo)
+		return hi;
+
+	return f(0, args) ? 0 : 1;
+}
+
+void free_names(char **a)
+{
+	char **p;
+	if (!a) {
+		return;
+	}
+	for (p = a; *p; p++) {
+		reftable_free(*p);
+	}
+	reftable_free(a);
+}
+
+int names_length(char **names)
+{
+	char **p = names;
+	for (; *p; p++) {
+		/* empty */
+	}
+	return p - names;
+}
+
+void parse_names(char *buf, int size, char ***namesp)
+{
+	char **names = NULL;
+	size_t names_cap = 0;
+	size_t names_len = 0;
+
+	char *p = buf;
+	char *end = buf + size;
+	while (p < end) {
+		char *next = strchr(p, '\n');
+		if (next && next < end) {
+			*next = 0;
+		} else {
+			next = end;
+		}
+		if (p < next) {
+			if (names_len == names_cap) {
+				names_cap = 2 * names_cap + 1;
+				names = reftable_realloc(
+					names, names_cap * sizeof(*names));
+			}
+			names[names_len++] = xstrdup(p);
+		}
+		p = next + 1;
+	}
+
+	names = reftable_realloc(names, (names_len + 1) * sizeof(*names));
+	names[names_len] = NULL;
+	*namesp = names;
+}
+
+int names_equal(char **a, char **b)
+{
+	int i = 0;
+	for (; a[i] && b[i]; i++) {
+		if (strcmp(a[i], b[i])) {
+			return 0;
+		}
+	}
+
+	return a[i] == b[i];
+}
+
+int common_prefix_size(struct strbuf *a, struct strbuf *b)
+{
+	int p = 0;
+	for (; p < a->len && p < b->len; p++) {
+		if (a->buf[p] != b->buf[p])
+			break;
+	}
+
+	return p;
+}
diff --git a/reftable/basics.h b/reftable/basics.h
new file mode 100644
index 0000000..096b368
--- /dev/null
+++ b/reftable/basics.h
@@ -0,0 +1,60 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef BASICS_H
+#define BASICS_H
+
+/*
+ * miscellaneous utilities that are not provided by Git.
+ */
+
+#include "system.h"
+
+/* Bigendian en/decoding of integers */
+
+void put_be24(uint8_t *out, uint32_t i);
+uint32_t get_be24(uint8_t *in);
+void put_be16(uint8_t *out, uint16_t i);
+
+/*
+ * find smallest index i in [0, sz) at which f(i) is true, assuming
+ * that f is ascending. Return sz if f(i) is false for all indices.
+ *
+ * Contrary to bsearch(3), this returns something useful if the argument is not
+ * found.
+ */
+int binsearch(size_t sz, int (*f)(size_t k, void *args), void *args);
+
+/*
+ * Frees a NULL terminated array of malloced strings. The array itself is also
+ * freed.
+ */
+void free_names(char **a);
+
+/* parse a newline separated list of names. `size` is the length of the buffer,
+ * without terminating '\0'. Empty names are discarded. */
+void parse_names(char *buf, int size, char ***namesp);
+
+/* compares two NULL-terminated arrays of strings. */
+int names_equal(char **a, char **b);
+
+/* returns the array size of a NULL-terminated array of strings. */
+int names_length(char **names);
+
+/* Allocation routines; they invoke the functions set through
+ * reftable_set_alloc() */
+void *reftable_malloc(size_t sz);
+void *reftable_realloc(void *p, size_t sz);
+void reftable_free(void *p);
+void *reftable_calloc(size_t sz);
+
+/* Find the longest shared prefix size of `a` and `b` */
+struct strbuf;
+int common_prefix_size(struct strbuf *a, struct strbuf *b);
+
+#endif
diff --git a/reftable/basics_test.c b/reftable/basics_test.c
new file mode 100644
index 0000000..1fcd229
--- /dev/null
+++ b/reftable/basics_test.c
@@ -0,0 +1,98 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+
+#include "basics.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+struct binsearch_args {
+	int key;
+	int *arr;
+};
+
+static int binsearch_func(size_t i, void *void_args)
+{
+	struct binsearch_args *args = void_args;
+
+	return args->key < args->arr[i];
+}
+
+static void test_binsearch(void)
+{
+	int arr[] = { 2, 4, 6, 8, 10 };
+	size_t sz = ARRAY_SIZE(arr);
+	struct binsearch_args args = {
+		.arr = arr,
+	};
+
+	int i = 0;
+	for (i = 1; i < 11; i++) {
+		int res;
+		args.key = i;
+		res = binsearch(sz, &binsearch_func, &args);
+
+		if (res < sz) {
+			EXPECT(args.key < arr[res]);
+			if (res > 0) {
+				EXPECT(args.key >= arr[res - 1]);
+			}
+		} else {
+			EXPECT(args.key == 10 || args.key == 11);
+		}
+	}
+}
+
+static void test_names_length(void)
+{
+	char *a[] = { "a", "b", NULL };
+	EXPECT(names_length(a) == 2);
+}
+
+static void test_parse_names_normal(void)
+{
+	char in[] = "a\nb\n";
+	char **out = NULL;
+	parse_names(in, strlen(in), &out);
+	EXPECT(!strcmp(out[0], "a"));
+	EXPECT(!strcmp(out[1], "b"));
+	EXPECT(!out[2]);
+	free_names(out);
+}
+
+static void test_parse_names_drop_empty(void)
+{
+	char in[] = "a\n\n";
+	char **out = NULL;
+	parse_names(in, strlen(in), &out);
+	EXPECT(!strcmp(out[0], "a"));
+	EXPECT(!out[1]);
+	free_names(out);
+}
+
+static void test_common_prefix(void)
+{
+	struct strbuf s1 = STRBUF_INIT;
+	struct strbuf s2 = STRBUF_INIT;
+	strbuf_addstr(&s1, "abcdef");
+	strbuf_addstr(&s2, "abc");
+	EXPECT(common_prefix_size(&s1, &s2) == 3);
+	strbuf_release(&s1);
+	strbuf_release(&s2);
+}
+
+int basics_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_common_prefix);
+	RUN_TEST(test_parse_names_normal);
+	RUN_TEST(test_parse_names_drop_empty);
+	RUN_TEST(test_binsearch);
+	RUN_TEST(test_names_length);
+	return 0;
+}
diff --git a/reftable/block.c b/reftable/block.c
new file mode 100644
index 0000000..855e3f5
--- /dev/null
+++ b/reftable/block.c
@@ -0,0 +1,437 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "block.h"
+
+#include "blocksource.h"
+#include "constants.h"
+#include "record.h"
+#include "reftable-error.h"
+#include "system.h"
+#include <zlib.h>
+
+int header_size(int version)
+{
+	switch (version) {
+	case 1:
+		return 24;
+	case 2:
+		return 28;
+	}
+	abort();
+}
+
+int footer_size(int version)
+{
+	switch (version) {
+	case 1:
+		return 68;
+	case 2:
+		return 72;
+	}
+	abort();
+}
+
+static int block_writer_register_restart(struct block_writer *w, int n,
+					 int is_restart, struct strbuf *key)
+{
+	int rlen = w->restart_len;
+	if (rlen >= MAX_RESTARTS) {
+		is_restart = 0;
+	}
+
+	if (is_restart) {
+		rlen++;
+	}
+	if (2 + 3 * rlen + n > w->block_size - w->next)
+		return -1;
+	if (is_restart) {
+		if (w->restart_len == w->restart_cap) {
+			w->restart_cap = w->restart_cap * 2 + 1;
+			w->restarts = reftable_realloc(
+				w->restarts, sizeof(uint32_t) * w->restart_cap);
+		}
+
+		w->restarts[w->restart_len++] = w->next;
+	}
+
+	w->next += n;
+
+	strbuf_reset(&w->last_key);
+	strbuf_addbuf(&w->last_key, key);
+	w->entries++;
+	return 0;
+}
+
+void block_writer_init(struct block_writer *bw, uint8_t typ, uint8_t *buf,
+		       uint32_t block_size, uint32_t header_off, int hash_size)
+{
+	bw->buf = buf;
+	bw->hash_size = hash_size;
+	bw->block_size = block_size;
+	bw->header_off = header_off;
+	bw->buf[header_off] = typ;
+	bw->next = header_off + 4;
+	bw->restart_interval = 16;
+	bw->entries = 0;
+	bw->restart_len = 0;
+	bw->last_key.len = 0;
+}
+
+uint8_t block_writer_type(struct block_writer *bw)
+{
+	return bw->buf[bw->header_off];
+}
+
+/* adds the reftable_record to the block. Returns -1 if it does not fit, 0 on
+   success */
+int block_writer_add(struct block_writer *w, struct reftable_record *rec)
+{
+	struct strbuf empty = STRBUF_INIT;
+	struct strbuf last =
+		w->entries % w->restart_interval == 0 ? empty : w->last_key;
+	struct string_view out = {
+		.buf = w->buf + w->next,
+		.len = w->block_size - w->next,
+	};
+
+	struct string_view start = out;
+
+	int is_restart = 0;
+	struct strbuf key = STRBUF_INIT;
+	int n = 0;
+
+	reftable_record_key(rec, &key);
+	n = reftable_encode_key(&is_restart, out, last, key,
+				reftable_record_val_type(rec));
+	if (n < 0)
+		goto done;
+	string_view_consume(&out, n);
+
+	n = reftable_record_encode(rec, out, w->hash_size);
+	if (n < 0)
+		goto done;
+	string_view_consume(&out, n);
+
+	if (block_writer_register_restart(w, start.len - out.len, is_restart,
+					  &key) < 0)
+		goto done;
+
+	strbuf_release(&key);
+	return 0;
+
+done:
+	strbuf_release(&key);
+	return -1;
+}
+
+int block_writer_finish(struct block_writer *w)
+{
+	int i;
+	for (i = 0; i < w->restart_len; i++) {
+		put_be24(w->buf + w->next, w->restarts[i]);
+		w->next += 3;
+	}
+
+	put_be16(w->buf + w->next, w->restart_len);
+	w->next += 2;
+	put_be24(w->buf + 1 + w->header_off, w->next);
+
+	if (block_writer_type(w) == BLOCK_TYPE_LOG) {
+		int block_header_skip = 4 + w->header_off;
+		uLongf src_len = w->next - block_header_skip;
+		uLongf dest_cap = src_len * 1.001 + 12;
+
+		uint8_t *compressed = reftable_malloc(dest_cap);
+		while (1) {
+			uLongf out_dest_len = dest_cap;
+			int zresult = compress2(compressed, &out_dest_len,
+						w->buf + block_header_skip,
+						src_len, 9);
+			if (zresult == Z_BUF_ERROR && dest_cap < LONG_MAX) {
+				dest_cap *= 2;
+				compressed =
+					reftable_realloc(compressed, dest_cap);
+				if (compressed)
+					continue;
+			}
+
+			if (Z_OK != zresult) {
+				reftable_free(compressed);
+				return REFTABLE_ZLIB_ERROR;
+			}
+
+			memcpy(w->buf + block_header_skip, compressed,
+			       out_dest_len);
+			w->next = out_dest_len + block_header_skip;
+			reftable_free(compressed);
+			break;
+		}
+	}
+	return w->next;
+}
+
+uint8_t block_reader_type(struct block_reader *r)
+{
+	return r->block.data[r->header_off];
+}
+
+int block_reader_init(struct block_reader *br, struct reftable_block *block,
+		      uint32_t header_off, uint32_t table_block_size,
+		      int hash_size)
+{
+	uint32_t full_block_size = table_block_size;
+	uint8_t typ = block->data[header_off];
+	uint32_t sz = get_be24(block->data + header_off + 1);
+
+	uint16_t restart_count = 0;
+	uint32_t restart_start = 0;
+	uint8_t *restart_bytes = NULL;
+
+	if (!reftable_is_block_type(typ))
+		return REFTABLE_FORMAT_ERROR;
+
+	if (typ == BLOCK_TYPE_LOG) {
+		int block_header_skip = 4 + header_off;
+		uLongf dst_len = sz - block_header_skip; /* total size of dest
+							    buffer. */
+		uLongf src_len = block->len - block_header_skip;
+		/* Log blocks specify the *uncompressed* size in their header.
+		 */
+		uint8_t *uncompressed = reftable_malloc(sz);
+
+		/* Copy over the block header verbatim. It's not compressed. */
+		memcpy(uncompressed, block->data, block_header_skip);
+
+		/* Uncompress */
+		if (Z_OK !=
+		    uncompress2(uncompressed + block_header_skip, &dst_len,
+				block->data + block_header_skip, &src_len)) {
+			reftable_free(uncompressed);
+			return REFTABLE_ZLIB_ERROR;
+		}
+
+		if (dst_len + block_header_skip != sz)
+			return REFTABLE_FORMAT_ERROR;
+
+		/* We're done with the input data. */
+		reftable_block_done(block);
+		block->data = uncompressed;
+		block->len = sz;
+		block->source = malloc_block_source();
+		full_block_size = src_len + block_header_skip;
+	} else if (full_block_size == 0) {
+		full_block_size = sz;
+	} else if (sz < full_block_size && sz < block->len &&
+		   block->data[sz] != 0) {
+		/* If the block is smaller than the full block size, it is
+		   padded (data followed by '\0') or the next block is
+		   unaligned. */
+		full_block_size = sz;
+	}
+
+	restart_count = get_be16(block->data + sz - 2);
+	restart_start = sz - 2 - 3 * restart_count;
+	restart_bytes = block->data + restart_start;
+
+	/* transfer ownership. */
+	br->block = *block;
+	block->data = NULL;
+	block->len = 0;
+
+	br->hash_size = hash_size;
+	br->block_len = restart_start;
+	br->full_block_size = full_block_size;
+	br->header_off = header_off;
+	br->restart_count = restart_count;
+	br->restart_bytes = restart_bytes;
+
+	return 0;
+}
+
+static uint32_t block_reader_restart_offset(struct block_reader *br, int i)
+{
+	return get_be24(br->restart_bytes + 3 * i);
+}
+
+void block_reader_start(struct block_reader *br, struct block_iter *it)
+{
+	it->br = br;
+	strbuf_reset(&it->last_key);
+	it->next_off = br->header_off + 4;
+}
+
+struct restart_find_args {
+	int error;
+	struct strbuf key;
+	struct block_reader *r;
+};
+
+static int restart_key_less(size_t idx, void *args)
+{
+	struct restart_find_args *a = args;
+	uint32_t off = block_reader_restart_offset(a->r, idx);
+	struct string_view in = {
+		.buf = a->r->block.data + off,
+		.len = a->r->block_len - off,
+	};
+
+	/* the restart key is verbatim in the block, so this could avoid the
+	   alloc for decoding the key */
+	struct strbuf rkey = STRBUF_INIT;
+	struct strbuf last_key = STRBUF_INIT;
+	uint8_t unused_extra;
+	int n = reftable_decode_key(&rkey, &unused_extra, last_key, in);
+	int result;
+	if (n < 0) {
+		a->error = 1;
+		return -1;
+	}
+
+	result = strbuf_cmp(&a->key, &rkey);
+	strbuf_release(&rkey);
+	return result;
+}
+
+void block_iter_copy_from(struct block_iter *dest, struct block_iter *src)
+{
+	dest->br = src->br;
+	dest->next_off = src->next_off;
+	strbuf_reset(&dest->last_key);
+	strbuf_addbuf(&dest->last_key, &src->last_key);
+}
+
+int block_iter_next(struct block_iter *it, struct reftable_record *rec)
+{
+	struct string_view in = {
+		.buf = it->br->block.data + it->next_off,
+		.len = it->br->block_len - it->next_off,
+	};
+	struct string_view start = in;
+	struct strbuf key = STRBUF_INIT;
+	uint8_t extra = 0;
+	int n = 0;
+
+	if (it->next_off >= it->br->block_len)
+		return 1;
+
+	n = reftable_decode_key(&key, &extra, it->last_key, in);
+	if (n < 0)
+		return -1;
+
+	string_view_consume(&in, n);
+	n = reftable_record_decode(rec, key, extra, in, it->br->hash_size);
+	if (n < 0)
+		return -1;
+	string_view_consume(&in, n);
+
+	strbuf_reset(&it->last_key);
+	strbuf_addbuf(&it->last_key, &key);
+	it->next_off += start.len - in.len;
+	strbuf_release(&key);
+	return 0;
+}
+
+int block_reader_first_key(struct block_reader *br, struct strbuf *key)
+{
+	struct strbuf empty = STRBUF_INIT;
+	int off = br->header_off + 4;
+	struct string_view in = {
+		.buf = br->block.data + off,
+		.len = br->block_len - off,
+	};
+
+	uint8_t extra = 0;
+	int n = reftable_decode_key(key, &extra, empty, in);
+	if (n < 0)
+		return n;
+
+	return 0;
+}
+
+int block_iter_seek(struct block_iter *it, struct strbuf *want)
+{
+	return block_reader_seek(it->br, it, want);
+}
+
+void block_iter_close(struct block_iter *it)
+{
+	strbuf_release(&it->last_key);
+}
+
+int block_reader_seek(struct block_reader *br, struct block_iter *it,
+		      struct strbuf *want)
+{
+	struct restart_find_args args = {
+		.key = *want,
+		.r = br,
+	};
+	struct reftable_record rec = reftable_new_record(block_reader_type(br));
+	struct strbuf key = STRBUF_INIT;
+	int err = 0;
+	struct block_iter next = {
+		.last_key = STRBUF_INIT,
+	};
+
+	int i = binsearch(br->restart_count, &restart_key_less, &args);
+	if (args.error) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+
+	it->br = br;
+	if (i > 0) {
+		i--;
+		it->next_off = block_reader_restart_offset(br, i);
+	} else {
+		it->next_off = br->header_off + 4;
+	}
+
+	/* We're looking for the last entry less/equal than the wanted key, so
+	   we have to go one entry too far and then back up.
+	*/
+	while (1) {
+		block_iter_copy_from(&next, it);
+		err = block_iter_next(&next, &rec);
+		if (err < 0)
+			goto done;
+
+		reftable_record_key(&rec, &key);
+		if (err > 0 || strbuf_cmp(&key, want) >= 0) {
+			err = 0;
+			goto done;
+		}
+
+		block_iter_copy_from(it, &next);
+	}
+
+done:
+	strbuf_release(&key);
+	strbuf_release(&next.last_key);
+	reftable_record_destroy(&rec);
+
+	return err;
+}
+
+void block_writer_release(struct block_writer *bw)
+{
+	FREE_AND_NULL(bw->restarts);
+	strbuf_release(&bw->last_key);
+	/* the block is not owned. */
+}
+
+void reftable_block_done(struct reftable_block *blockp)
+{
+	struct reftable_block_source source = blockp->source;
+	if (blockp && source.ops)
+		source.ops->return_block(source.arg, blockp);
+	blockp->data = NULL;
+	blockp->len = 0;
+	blockp->source.ops = NULL;
+	blockp->source.arg = NULL;
+}
diff --git a/reftable/block.h b/reftable/block.h
new file mode 100644
index 0000000..e207706
--- /dev/null
+++ b/reftable/block.h
@@ -0,0 +1,127 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef BLOCK_H
+#define BLOCK_H
+
+#include "basics.h"
+#include "record.h"
+#include "reftable-blocksource.h"
+
+/*
+ * Writes reftable blocks. The block_writer is reused across blocks to minimize
+ * allocation overhead.
+ */
+struct block_writer {
+	uint8_t *buf;
+	uint32_t block_size;
+
+	/* Offset ofof the global header. Nonzero in the first block only. */
+	uint32_t header_off;
+
+	/* How often to restart keys. */
+	int restart_interval;
+	int hash_size;
+
+	/* Offset of next uint8_t to write. */
+	uint32_t next;
+	uint32_t *restarts;
+	uint32_t restart_len;
+	uint32_t restart_cap;
+
+	struct strbuf last_key;
+	int entries;
+};
+
+/*
+ * initializes the blockwriter to write `typ` entries, using `buf` as temporary
+ * storage. `buf` is not owned by the block_writer. */
+void block_writer_init(struct block_writer *bw, uint8_t typ, uint8_t *buf,
+		       uint32_t block_size, uint32_t header_off, int hash_size);
+
+/* returns the block type (eg. 'r' for ref records. */
+uint8_t block_writer_type(struct block_writer *bw);
+
+/* appends the record, or -1 if it doesn't fit. */
+int block_writer_add(struct block_writer *w, struct reftable_record *rec);
+
+/* appends the key restarts, and compress the block if necessary. */
+int block_writer_finish(struct block_writer *w);
+
+/* clears out internally allocated block_writer members. */
+void block_writer_release(struct block_writer *bw);
+
+/* Read a block. */
+struct block_reader {
+	/* offset of the block header; nonzero for the first block in a
+	 * reftable. */
+	uint32_t header_off;
+
+	/* the memory block */
+	struct reftable_block block;
+	int hash_size;
+
+	/* size of the data, excluding restart data. */
+	uint32_t block_len;
+	uint8_t *restart_bytes;
+	uint16_t restart_count;
+
+	/* size of the data in the file. For log blocks, this is the compressed
+	 * size. */
+	uint32_t full_block_size;
+};
+
+/* Iterate over entries in a block */
+struct block_iter {
+	/* offset within the block of the next entry to read. */
+	uint32_t next_off;
+	struct block_reader *br;
+
+	/* key for last entry we read. */
+	struct strbuf last_key;
+};
+
+/* initializes a block reader. */
+int block_reader_init(struct block_reader *br, struct reftable_block *bl,
+		      uint32_t header_off, uint32_t table_block_size,
+		      int hash_size);
+
+/* Position `it` at start of the block */
+void block_reader_start(struct block_reader *br, struct block_iter *it);
+
+/* Position `it` to the `want` key in the block */
+int block_reader_seek(struct block_reader *br, struct block_iter *it,
+		      struct strbuf *want);
+
+/* Returns the block type (eg. 'r' for refs) */
+uint8_t block_reader_type(struct block_reader *r);
+
+/* Decodes the first key in the block */
+int block_reader_first_key(struct block_reader *br, struct strbuf *key);
+
+void block_iter_copy_from(struct block_iter *dest, struct block_iter *src);
+
+/* return < 0 for error, 0 for OK, > 0 for EOF. */
+int block_iter_next(struct block_iter *it, struct reftable_record *rec);
+
+/* Seek to `want` with in the block pointed to by `it` */
+int block_iter_seek(struct block_iter *it, struct strbuf *want);
+
+/* deallocate memory for `it`. The block reader and its block is left intact. */
+void block_iter_close(struct block_iter *it);
+
+/* size of file header, depending on format version */
+int header_size(int version);
+
+/* size of file footer, depending on format version */
+int footer_size(int version);
+
+/* returns a block to its source. */
+void reftable_block_done(struct reftable_block *ret);
+
+#endif
diff --git a/reftable/block_test.c b/reftable/block_test.c
new file mode 100644
index 0000000..4b3ea26
--- /dev/null
+++ b/reftable/block_test.c
@@ -0,0 +1,120 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "block.h"
+
+#include "system.h"
+#include "blocksource.h"
+#include "basics.h"
+#include "constants.h"
+#include "record.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+static void test_block_read_write(void)
+{
+	const int header_off = 21; /* random */
+	char *names[30];
+	const int N = ARRAY_SIZE(names);
+	const int block_size = 1024;
+	struct reftable_block block = { NULL };
+	struct block_writer bw = {
+		.last_key = STRBUF_INIT,
+	};
+	struct reftable_ref_record ref = { NULL };
+	struct reftable_record rec = { NULL };
+	int i = 0;
+	int n;
+	struct block_reader br = { 0 };
+	struct block_iter it = { .last_key = STRBUF_INIT };
+	int j = 0;
+	struct strbuf want = STRBUF_INIT;
+
+	block.data = reftable_calloc(block_size);
+	block.len = block_size;
+	block.source = malloc_block_source();
+	block_writer_init(&bw, BLOCK_TYPE_REF, block.data, block_size,
+			  header_off, hash_size(GIT_SHA1_FORMAT_ID));
+	reftable_record_from_ref(&rec, &ref);
+
+	for (i = 0; i < N; i++) {
+		char name[100];
+		uint8_t hash[GIT_SHA1_RAWSZ];
+		snprintf(name, sizeof(name), "branch%02d", i);
+		memset(hash, i, sizeof(hash));
+
+		ref.refname = name;
+		ref.value_type = REFTABLE_REF_VAL1;
+		ref.value.val1 = hash;
+
+		names[i] = xstrdup(name);
+		n = block_writer_add(&bw, &rec);
+		ref.refname = NULL;
+		ref.value_type = REFTABLE_REF_DELETION;
+		EXPECT(n == 0);
+	}
+
+	n = block_writer_finish(&bw);
+	EXPECT(n > 0);
+
+	block_writer_release(&bw);
+
+	block_reader_init(&br, &block, header_off, block_size, GIT_SHA1_RAWSZ);
+
+	block_reader_start(&br, &it);
+
+	while (1) {
+		int r = block_iter_next(&it, &rec);
+		EXPECT(r >= 0);
+		if (r > 0) {
+			break;
+		}
+		EXPECT_STREQ(names[j], ref.refname);
+		j++;
+	}
+
+	reftable_record_release(&rec);
+	block_iter_close(&it);
+
+	for (i = 0; i < N; i++) {
+		struct block_iter it = { .last_key = STRBUF_INIT };
+		strbuf_reset(&want);
+		strbuf_addstr(&want, names[i]);
+
+		n = block_reader_seek(&br, &it, &want);
+		EXPECT(n == 0);
+
+		n = block_iter_next(&it, &rec);
+		EXPECT(n == 0);
+
+		EXPECT_STREQ(names[i], ref.refname);
+
+		want.len--;
+		n = block_reader_seek(&br, &it, &want);
+		EXPECT(n == 0);
+
+		n = block_iter_next(&it, &rec);
+		EXPECT(n == 0);
+		EXPECT_STREQ(names[10 * (i / 10)], ref.refname);
+
+		block_iter_close(&it);
+	}
+
+	reftable_record_release(&rec);
+	reftable_block_done(&br.block);
+	strbuf_release(&want);
+	for (i = 0; i < N; i++) {
+		reftable_free(names[i]);
+	}
+}
+
+int block_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_block_read_write);
+	return 0;
+}
diff --git a/reftable/blocksource.c b/reftable/blocksource.c
new file mode 100644
index 0000000..0044eec
--- /dev/null
+++ b/reftable/blocksource.c
@@ -0,0 +1,148 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+
+#include "basics.h"
+#include "blocksource.h"
+#include "reftable-blocksource.h"
+#include "reftable-error.h"
+
+static void strbuf_return_block(void *b, struct reftable_block *dest)
+{
+	memset(dest->data, 0xff, dest->len);
+	reftable_free(dest->data);
+}
+
+static void strbuf_close(void *b)
+{
+}
+
+static int strbuf_read_block(void *v, struct reftable_block *dest, uint64_t off,
+			     uint32_t size)
+{
+	struct strbuf *b = v;
+	assert(off + size <= b->len);
+	dest->data = reftable_calloc(size);
+	memcpy(dest->data, b->buf + off, size);
+	dest->len = size;
+	return size;
+}
+
+static uint64_t strbuf_size(void *b)
+{
+	return ((struct strbuf *)b)->len;
+}
+
+static struct reftable_block_source_vtable strbuf_vtable = {
+	.size = &strbuf_size,
+	.read_block = &strbuf_read_block,
+	.return_block = &strbuf_return_block,
+	.close = &strbuf_close,
+};
+
+void block_source_from_strbuf(struct reftable_block_source *bs,
+			      struct strbuf *buf)
+{
+	assert(!bs->ops);
+	bs->ops = &strbuf_vtable;
+	bs->arg = buf;
+}
+
+static void malloc_return_block(void *b, struct reftable_block *dest)
+{
+	memset(dest->data, 0xff, dest->len);
+	reftable_free(dest->data);
+}
+
+static struct reftable_block_source_vtable malloc_vtable = {
+	.return_block = &malloc_return_block,
+};
+
+static struct reftable_block_source malloc_block_source_instance = {
+	.ops = &malloc_vtable,
+};
+
+struct reftable_block_source malloc_block_source(void)
+{
+	return malloc_block_source_instance;
+}
+
+struct file_block_source {
+	int fd;
+	uint64_t size;
+};
+
+static uint64_t file_size(void *b)
+{
+	return ((struct file_block_source *)b)->size;
+}
+
+static void file_return_block(void *b, struct reftable_block *dest)
+{
+	memset(dest->data, 0xff, dest->len);
+	reftable_free(dest->data);
+}
+
+static void file_close(void *b)
+{
+	int fd = ((struct file_block_source *)b)->fd;
+	if (fd > 0) {
+		close(fd);
+		((struct file_block_source *)b)->fd = 0;
+	}
+
+	reftable_free(b);
+}
+
+static int file_read_block(void *v, struct reftable_block *dest, uint64_t off,
+			   uint32_t size)
+{
+	struct file_block_source *b = v;
+	assert(off + size <= b->size);
+	dest->data = reftable_malloc(size);
+	if (pread(b->fd, dest->data, size, off) != size)
+		return -1;
+	dest->len = size;
+	return size;
+}
+
+static struct reftable_block_source_vtable file_vtable = {
+	.size = &file_size,
+	.read_block = &file_read_block,
+	.return_block = &file_return_block,
+	.close = &file_close,
+};
+
+int reftable_block_source_from_file(struct reftable_block_source *bs,
+				    const char *name)
+{
+	struct stat st = { 0 };
+	int err = 0;
+	int fd = open(name, O_RDONLY);
+	struct file_block_source *p = NULL;
+	if (fd < 0) {
+		if (errno == ENOENT) {
+			return REFTABLE_NOT_EXIST_ERROR;
+		}
+		return -1;
+	}
+
+	err = fstat(fd, &st);
+	if (err < 0)
+		return -1;
+
+	p = reftable_calloc(sizeof(struct file_block_source));
+	p->size = st.st_size;
+	p->fd = fd;
+
+	assert(!bs->ops);
+	bs->ops = &file_vtable;
+	bs->arg = p;
+	return 0;
+}
diff --git a/reftable/blocksource.h b/reftable/blocksource.h
new file mode 100644
index 0000000..072e272
--- /dev/null
+++ b/reftable/blocksource.h
@@ -0,0 +1,22 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef BLOCKSOURCE_H
+#define BLOCKSOURCE_H
+
+#include "system.h"
+
+struct reftable_block_source;
+
+/* Create an in-memory block source for reading reftables */
+void block_source_from_strbuf(struct reftable_block_source *bs,
+			      struct strbuf *buf);
+
+struct reftable_block_source malloc_block_source(void);
+
+#endif
diff --git a/reftable/constants.h b/reftable/constants.h
new file mode 100644
index 0000000..5eee72c
--- /dev/null
+++ b/reftable/constants.h
@@ -0,0 +1,21 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef CONSTANTS_H
+#define CONSTANTS_H
+
+#define BLOCK_TYPE_LOG 'g'
+#define BLOCK_TYPE_INDEX 'i'
+#define BLOCK_TYPE_REF 'r'
+#define BLOCK_TYPE_OBJ 'o'
+#define BLOCK_TYPE_ANY 0
+
+#define MAX_RESTARTS ((1 << 16) - 1)
+#define DEFAULT_BLOCK_SIZE 4096
+
+#endif
diff --git a/reftable/dump.c b/reftable/dump.c
new file mode 100644
index 0000000..155953d
--- /dev/null
+++ b/reftable/dump.c
@@ -0,0 +1,107 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "git-compat-util.h"
+#include "hash.h"
+
+#include "reftable-blocksource.h"
+#include "reftable-error.h"
+#include "reftable-merged.h"
+#include "reftable-record.h"
+#include "reftable-tests.h"
+#include "reftable-writer.h"
+#include "reftable-iterator.h"
+#include "reftable-reader.h"
+#include "reftable-stack.h"
+#include "reftable-generic.h"
+
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+static int compact_stack(const char *stackdir)
+{
+	struct reftable_stack *stack = NULL;
+	struct reftable_write_options cfg = { 0 };
+
+	int err = reftable_new_stack(&stack, stackdir, cfg);
+	if (err < 0)
+		goto done;
+
+	err = reftable_stack_compact_all(stack, NULL);
+	if (err < 0)
+		goto done;
+done:
+	if (stack) {
+		reftable_stack_destroy(stack);
+	}
+	return err;
+}
+
+static void print_help(void)
+{
+	printf("usage: dump [-cst] arg\n\n"
+	       "options: \n"
+	       "  -c compact\n"
+	       "  -t dump table\n"
+	       "  -s dump stack\n"
+	       "  -6 sha256 hash format\n"
+	       "  -h this help\n"
+	       "\n");
+}
+
+int reftable_dump_main(int argc, char *const *argv)
+{
+	int err = 0;
+	int opt_dump_table = 0;
+	int opt_dump_stack = 0;
+	int opt_compact = 0;
+	uint32_t opt_hash_id = GIT_SHA1_FORMAT_ID;
+	const char *arg = NULL, *argv0 = argv[0];
+
+	for (; argc > 1; argv++, argc--)
+		if (*argv[1] != '-')
+			break;
+		else if (!strcmp("-t", argv[1]))
+			opt_dump_table = 1;
+		else if (!strcmp("-6", argv[1]))
+			opt_hash_id = GIT_SHA256_FORMAT_ID;
+		else if (!strcmp("-s", argv[1]))
+			opt_dump_stack = 1;
+		else if (!strcmp("-c", argv[1]))
+			opt_compact = 1;
+		else if (!strcmp("-?", argv[1]) || !strcmp("-h", argv[1])) {
+			print_help();
+			return 2;
+		}
+
+	if (argc != 2) {
+		fprintf(stderr, "need argument\n");
+		print_help();
+		return 2;
+	}
+
+	arg = argv[1];
+
+	if (opt_dump_table) {
+		err = reftable_reader_print_file(arg);
+	} else if (opt_dump_stack) {
+		err = reftable_stack_print_directory(arg, opt_hash_id);
+	} else if (opt_compact) {
+		err = compact_stack(arg);
+	}
+
+	if (err < 0) {
+		fprintf(stderr, "%s: %s: %s\n", argv0, arg,
+			reftable_error_str(err));
+		return 1;
+	}
+	return 0;
+}
diff --git a/reftable/error.c b/reftable/error.c
new file mode 100644
index 0000000..f6f16de
--- /dev/null
+++ b/reftable/error.c
@@ -0,0 +1,41 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "reftable-error.h"
+
+#include <stdio.h>
+
+const char *reftable_error_str(int err)
+{
+	static char buf[250];
+	switch (err) {
+	case REFTABLE_IO_ERROR:
+		return "I/O error";
+	case REFTABLE_FORMAT_ERROR:
+		return "corrupt reftable file";
+	case REFTABLE_NOT_EXIST_ERROR:
+		return "file does not exist";
+	case REFTABLE_LOCK_ERROR:
+		return "data is outdated";
+	case REFTABLE_API_ERROR:
+		return "misuse of the reftable API";
+	case REFTABLE_ZLIB_ERROR:
+		return "zlib failure";
+	case REFTABLE_NAME_CONFLICT:
+		return "file/directory conflict";
+	case REFTABLE_EMPTY_TABLE_ERROR:
+		return "wrote empty table";
+	case REFTABLE_REFNAME_ERROR:
+		return "invalid refname";
+	case -1:
+		return "general error";
+	default:
+		snprintf(buf, sizeof(buf), "unknown error code %d", err);
+		return buf;
+	}
+}
diff --git a/reftable/generic.c b/reftable/generic.c
new file mode 100644
index 0000000..7a8a738
--- /dev/null
+++ b/reftable/generic.c
@@ -0,0 +1,169 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "basics.h"
+#include "record.h"
+#include "generic.h"
+#include "reftable-iterator.h"
+#include "reftable-generic.h"
+
+int reftable_table_seek_ref(struct reftable_table *tab,
+			    struct reftable_iterator *it, const char *name)
+{
+	struct reftable_ref_record ref = {
+		.refname = (char *)name,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, &ref);
+	return tab->ops->seek_record(tab->table_arg, it, &rec);
+}
+
+int reftable_table_seek_log(struct reftable_table *tab,
+			    struct reftable_iterator *it, const char *name)
+{
+	struct reftable_log_record log = {
+		.refname = (char *)name,
+		.update_index = ~((uint64_t)0),
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_log(&rec, &log);
+	return tab->ops->seek_record(tab->table_arg, it, &rec);
+}
+
+int reftable_table_read_ref(struct reftable_table *tab, const char *name,
+			    struct reftable_ref_record *ref)
+{
+	struct reftable_iterator it = { NULL };
+	int err = reftable_table_seek_ref(tab, &it, name);
+	if (err)
+		goto done;
+
+	err = reftable_iterator_next_ref(&it, ref);
+	if (err)
+		goto done;
+
+	if (strcmp(ref->refname, name) ||
+	    reftable_ref_record_is_deletion(ref)) {
+		reftable_ref_record_release(ref);
+		err = 1;
+		goto done;
+	}
+
+done:
+	reftable_iterator_destroy(&it);
+	return err;
+}
+
+int reftable_table_print(struct reftable_table *tab) {
+	struct reftable_iterator it = { NULL };
+	struct reftable_ref_record ref = { NULL };
+	struct reftable_log_record log = { NULL };
+	uint32_t hash_id = reftable_table_hash_id(tab);
+	int err = reftable_table_seek_ref(tab, &it, "");
+	if (err < 0) {
+		return err;
+	}
+
+	while (1) {
+		err = reftable_iterator_next_ref(&it, &ref);
+		if (err > 0) {
+			break;
+		}
+		if (err < 0) {
+			return err;
+		}
+		reftable_ref_record_print(&ref, hash_id);
+	}
+	reftable_iterator_destroy(&it);
+	reftable_ref_record_release(&ref);
+
+	err = reftable_table_seek_log(tab, &it, "");
+	if (err < 0) {
+		return err;
+	}
+	while (1) {
+		err = reftable_iterator_next_log(&it, &log);
+		if (err > 0) {
+			break;
+		}
+		if (err < 0) {
+			return err;
+		}
+		reftable_log_record_print(&log, hash_id);
+	}
+	reftable_iterator_destroy(&it);
+	reftable_log_record_release(&log);
+	return 0;
+}
+
+uint64_t reftable_table_max_update_index(struct reftable_table *tab)
+{
+	return tab->ops->max_update_index(tab->table_arg);
+}
+
+uint64_t reftable_table_min_update_index(struct reftable_table *tab)
+{
+	return tab->ops->min_update_index(tab->table_arg);
+}
+
+uint32_t reftable_table_hash_id(struct reftable_table *tab)
+{
+	return tab->ops->hash_id(tab->table_arg);
+}
+
+void reftable_iterator_destroy(struct reftable_iterator *it)
+{
+	if (!it->ops) {
+		return;
+	}
+	it->ops->close(it->iter_arg);
+	it->ops = NULL;
+	FREE_AND_NULL(it->iter_arg);
+}
+
+int reftable_iterator_next_ref(struct reftable_iterator *it,
+			       struct reftable_ref_record *ref)
+{
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, ref);
+	return iterator_next(it, &rec);
+}
+
+int reftable_iterator_next_log(struct reftable_iterator *it,
+			       struct reftable_log_record *log)
+{
+	struct reftable_record rec = { NULL };
+	reftable_record_from_log(&rec, log);
+	return iterator_next(it, &rec);
+}
+
+int iterator_next(struct reftable_iterator *it, struct reftable_record *rec)
+{
+	return it->ops->next(it->iter_arg, rec);
+}
+
+static int empty_iterator_next(void *arg, struct reftable_record *rec)
+{
+	return 1;
+}
+
+static void empty_iterator_close(void *arg)
+{
+}
+
+static struct reftable_iterator_vtable empty_vtable = {
+	.next = &empty_iterator_next,
+	.close = &empty_iterator_close,
+};
+
+void iterator_set_empty(struct reftable_iterator *it)
+{
+	assert(!it->ops);
+	it->iter_arg = NULL;
+	it->ops = &empty_vtable;
+}
diff --git a/reftable/generic.h b/reftable/generic.h
new file mode 100644
index 0000000..98886a0
--- /dev/null
+++ b/reftable/generic.h
@@ -0,0 +1,32 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef GENERIC_H
+#define GENERIC_H
+
+#include "record.h"
+#include "reftable-generic.h"
+
+/* generic interface to reftables */
+struct reftable_table_vtable {
+	int (*seek_record)(void *tab, struct reftable_iterator *it,
+			   struct reftable_record *);
+	uint32_t (*hash_id)(void *tab);
+	uint64_t (*min_update_index)(void *tab);
+	uint64_t (*max_update_index)(void *tab);
+};
+
+struct reftable_iterator_vtable {
+	int (*next)(void *iter_arg, struct reftable_record *rec);
+	void (*close)(void *iter_arg);
+};
+
+void iterator_set_empty(struct reftable_iterator *it);
+int iterator_next(struct reftable_iterator *it, struct reftable_record *rec);
+
+#endif
diff --git a/reftable/iter.c b/reftable/iter.c
new file mode 100644
index 0000000..93d04f7
--- /dev/null
+++ b/reftable/iter.c
@@ -0,0 +1,194 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "iter.h"
+
+#include "system.h"
+
+#include "block.h"
+#include "generic.h"
+#include "constants.h"
+#include "reader.h"
+#include "reftable-error.h"
+
+int iterator_is_null(struct reftable_iterator *it)
+{
+	return !it->ops;
+}
+
+static void filtering_ref_iterator_close(void *iter_arg)
+{
+	struct filtering_ref_iterator *fri = iter_arg;
+	strbuf_release(&fri->oid);
+	reftable_iterator_destroy(&fri->it);
+}
+
+static int filtering_ref_iterator_next(void *iter_arg,
+				       struct reftable_record *rec)
+{
+	struct filtering_ref_iterator *fri = iter_arg;
+	struct reftable_ref_record *ref = rec->data;
+	int err = 0;
+	while (1) {
+		err = reftable_iterator_next_ref(&fri->it, ref);
+		if (err != 0) {
+			break;
+		}
+
+		if (fri->double_check) {
+			struct reftable_iterator it = { NULL };
+
+			err = reftable_table_seek_ref(&fri->tab, &it,
+						      ref->refname);
+			if (err == 0) {
+				err = reftable_iterator_next_ref(&it, ref);
+			}
+
+			reftable_iterator_destroy(&it);
+
+			if (err < 0) {
+				break;
+			}
+
+			if (err > 0) {
+				continue;
+			}
+		}
+
+		if (ref->value_type == REFTABLE_REF_VAL2 &&
+		    (!memcmp(fri->oid.buf, ref->value.val2.target_value,
+			     fri->oid.len) ||
+		     !memcmp(fri->oid.buf, ref->value.val2.value,
+			     fri->oid.len)))
+			return 0;
+
+		if (ref->value_type == REFTABLE_REF_VAL1 &&
+		    !memcmp(fri->oid.buf, ref->value.val1, fri->oid.len)) {
+			return 0;
+		}
+	}
+
+	reftable_ref_record_release(ref);
+	return err;
+}
+
+static struct reftable_iterator_vtable filtering_ref_iterator_vtable = {
+	.next = &filtering_ref_iterator_next,
+	.close = &filtering_ref_iterator_close,
+};
+
+void iterator_from_filtering_ref_iterator(struct reftable_iterator *it,
+					  struct filtering_ref_iterator *fri)
+{
+	assert(!it->ops);
+	it->iter_arg = fri;
+	it->ops = &filtering_ref_iterator_vtable;
+}
+
+static void indexed_table_ref_iter_close(void *p)
+{
+	struct indexed_table_ref_iter *it = p;
+	block_iter_close(&it->cur);
+	reftable_block_done(&it->block_reader.block);
+	reftable_free(it->offsets);
+	strbuf_release(&it->oid);
+}
+
+static int indexed_table_ref_iter_next_block(struct indexed_table_ref_iter *it)
+{
+	uint64_t off;
+	int err = 0;
+	if (it->offset_idx == it->offset_len) {
+		it->is_finished = 1;
+		return 1;
+	}
+
+	reftable_block_done(&it->block_reader.block);
+
+	off = it->offsets[it->offset_idx++];
+	err = reader_init_block_reader(it->r, &it->block_reader, off,
+				       BLOCK_TYPE_REF);
+	if (err < 0) {
+		return err;
+	}
+	if (err > 0) {
+		/* indexed block does not exist. */
+		return REFTABLE_FORMAT_ERROR;
+	}
+	block_reader_start(&it->block_reader, &it->cur);
+	return 0;
+}
+
+static int indexed_table_ref_iter_next(void *p, struct reftable_record *rec)
+{
+	struct indexed_table_ref_iter *it = p;
+	struct reftable_ref_record *ref = rec->data;
+
+	while (1) {
+		int err = block_iter_next(&it->cur, rec);
+		if (err < 0) {
+			return err;
+		}
+
+		if (err > 0) {
+			err = indexed_table_ref_iter_next_block(it);
+			if (err < 0) {
+				return err;
+			}
+
+			if (it->is_finished) {
+				return 1;
+			}
+			continue;
+		}
+		/* BUG */
+		if (!memcmp(it->oid.buf, ref->value.val2.target_value,
+			    it->oid.len) ||
+		    !memcmp(it->oid.buf, ref->value.val2.value, it->oid.len)) {
+			return 0;
+		}
+	}
+}
+
+int new_indexed_table_ref_iter(struct indexed_table_ref_iter **dest,
+			       struct reftable_reader *r, uint8_t *oid,
+			       int oid_len, uint64_t *offsets, int offset_len)
+{
+	struct indexed_table_ref_iter empty = INDEXED_TABLE_REF_ITER_INIT;
+	struct indexed_table_ref_iter *itr =
+		reftable_calloc(sizeof(struct indexed_table_ref_iter));
+	int err = 0;
+
+	*itr = empty;
+	itr->r = r;
+	strbuf_add(&itr->oid, oid, oid_len);
+
+	itr->offsets = offsets;
+	itr->offset_len = offset_len;
+
+	err = indexed_table_ref_iter_next_block(itr);
+	if (err < 0) {
+		reftable_free(itr);
+	} else {
+		*dest = itr;
+	}
+	return err;
+}
+
+static struct reftable_iterator_vtable indexed_table_ref_iter_vtable = {
+	.next = &indexed_table_ref_iter_next,
+	.close = &indexed_table_ref_iter_close,
+};
+
+void iterator_from_indexed_table_ref_iter(struct reftable_iterator *it,
+					  struct indexed_table_ref_iter *itr)
+{
+	assert(!it->ops);
+	it->iter_arg = itr;
+	it->ops = &indexed_table_ref_iter_vtable;
+}
diff --git a/reftable/iter.h b/reftable/iter.h
new file mode 100644
index 0000000..09eb0cb
--- /dev/null
+++ b/reftable/iter.h
@@ -0,0 +1,69 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef ITER_H
+#define ITER_H
+
+#include "system.h"
+#include "block.h"
+#include "record.h"
+
+#include "reftable-iterator.h"
+#include "reftable-generic.h"
+
+/* Returns true for a zeroed out iterator, such as the one returned from
+ * iterator_destroy. */
+int iterator_is_null(struct reftable_iterator *it);
+
+/* iterator that produces only ref records that point to `oid` */
+struct filtering_ref_iterator {
+	int double_check;
+	struct reftable_table tab;
+	struct strbuf oid;
+	struct reftable_iterator it;
+};
+#define FILTERING_REF_ITERATOR_INIT \
+	{                           \
+		.oid = STRBUF_INIT  \
+	}
+
+void iterator_from_filtering_ref_iterator(struct reftable_iterator *,
+					  struct filtering_ref_iterator *);
+
+/* iterator that produces only ref records that point to `oid`,
+ * but using the object index.
+ */
+struct indexed_table_ref_iter {
+	struct reftable_reader *r;
+	struct strbuf oid;
+
+	/* mutable */
+	uint64_t *offsets;
+
+	/* Points to the next offset to read. */
+	int offset_idx;
+	int offset_len;
+	struct block_reader block_reader;
+	struct block_iter cur;
+	int is_finished;
+};
+
+#define INDEXED_TABLE_REF_ITER_INIT                                     \
+	{                                                               \
+		.cur = { .last_key = STRBUF_INIT }, .oid = STRBUF_INIT, \
+	}
+
+void iterator_from_indexed_table_ref_iter(struct reftable_iterator *it,
+					  struct indexed_table_ref_iter *itr);
+
+/* Takes ownership of `offsets` */
+int new_indexed_table_ref_iter(struct indexed_table_ref_iter **dest,
+			       struct reftable_reader *r, uint8_t *oid,
+			       int oid_len, uint64_t *offsets, int offset_len);
+
+#endif
diff --git a/reftable/merged.c b/reftable/merged.c
new file mode 100644
index 0000000..e5b53da
--- /dev/null
+++ b/reftable/merged.c
@@ -0,0 +1,362 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "merged.h"
+
+#include "constants.h"
+#include "iter.h"
+#include "pq.h"
+#include "reader.h"
+#include "record.h"
+#include "generic.h"
+#include "reftable-merged.h"
+#include "reftable-error.h"
+#include "system.h"
+
+static int merged_iter_init(struct merged_iter *mi)
+{
+	int i = 0;
+	for (i = 0; i < mi->stack_len; i++) {
+		struct reftable_record rec = reftable_new_record(mi->typ);
+		int err = iterator_next(&mi->stack[i], &rec);
+		if (err < 0) {
+			return err;
+		}
+
+		if (err > 0) {
+			reftable_iterator_destroy(&mi->stack[i]);
+			reftable_record_destroy(&rec);
+		} else {
+			struct pq_entry e = {
+				.rec = rec,
+				.index = i,
+			};
+			merged_iter_pqueue_add(&mi->pq, e);
+		}
+	}
+
+	return 0;
+}
+
+static void merged_iter_close(void *p)
+{
+	struct merged_iter *mi = p;
+	int i = 0;
+	merged_iter_pqueue_release(&mi->pq);
+	for (i = 0; i < mi->stack_len; i++) {
+		reftable_iterator_destroy(&mi->stack[i]);
+	}
+	reftable_free(mi->stack);
+}
+
+static int merged_iter_advance_nonnull_subiter(struct merged_iter *mi,
+					       size_t idx)
+{
+	struct reftable_record rec = reftable_new_record(mi->typ);
+	struct pq_entry e = {
+		.rec = rec,
+		.index = idx,
+	};
+	int err = iterator_next(&mi->stack[idx], &rec);
+	if (err < 0)
+		return err;
+
+	if (err > 0) {
+		reftable_iterator_destroy(&mi->stack[idx]);
+		reftable_record_destroy(&rec);
+		return 0;
+	}
+
+	merged_iter_pqueue_add(&mi->pq, e);
+	return 0;
+}
+
+static int merged_iter_advance_subiter(struct merged_iter *mi, size_t idx)
+{
+	if (iterator_is_null(&mi->stack[idx]))
+		return 0;
+	return merged_iter_advance_nonnull_subiter(mi, idx);
+}
+
+static int merged_iter_next_entry(struct merged_iter *mi,
+				  struct reftable_record *rec)
+{
+	struct strbuf entry_key = STRBUF_INIT;
+	struct pq_entry entry = { 0 };
+	int err = 0;
+
+	if (merged_iter_pqueue_is_empty(mi->pq))
+		return 1;
+
+	entry = merged_iter_pqueue_remove(&mi->pq);
+	err = merged_iter_advance_subiter(mi, entry.index);
+	if (err < 0)
+		return err;
+
+	/*
+	  One can also use reftable as datacenter-local storage, where the ref
+	  database is maintained in globally consistent database (eg.
+	  CockroachDB or Spanner). In this scenario, replication delays together
+	  with compaction may cause newer tables to contain older entries. In
+	  such a deployment, the loop below must be changed to collect all
+	  entries for the same key, and return new the newest one.
+	*/
+	reftable_record_key(&entry.rec, &entry_key);
+	while (!merged_iter_pqueue_is_empty(mi->pq)) {
+		struct pq_entry top = merged_iter_pqueue_top(mi->pq);
+		struct strbuf k = STRBUF_INIT;
+		int err = 0, cmp = 0;
+
+		reftable_record_key(&top.rec, &k);
+
+		cmp = strbuf_cmp(&k, &entry_key);
+		strbuf_release(&k);
+
+		if (cmp > 0) {
+			break;
+		}
+
+		merged_iter_pqueue_remove(&mi->pq);
+		err = merged_iter_advance_subiter(mi, top.index);
+		if (err < 0) {
+			return err;
+		}
+		reftable_record_destroy(&top.rec);
+	}
+
+	reftable_record_copy_from(rec, &entry.rec, hash_size(mi->hash_id));
+	reftable_record_destroy(&entry.rec);
+	strbuf_release(&entry_key);
+	return 0;
+}
+
+static int merged_iter_next(struct merged_iter *mi, struct reftable_record *rec)
+{
+	while (1) {
+		int err = merged_iter_next_entry(mi, rec);
+		if (err == 0 && mi->suppress_deletions &&
+		    reftable_record_is_deletion(rec)) {
+			continue;
+		}
+
+		return err;
+	}
+}
+
+static int merged_iter_next_void(void *p, struct reftable_record *rec)
+{
+	struct merged_iter *mi = p;
+	if (merged_iter_pqueue_is_empty(mi->pq))
+		return 1;
+
+	return merged_iter_next(mi, rec);
+}
+
+static struct reftable_iterator_vtable merged_iter_vtable = {
+	.next = &merged_iter_next_void,
+	.close = &merged_iter_close,
+};
+
+static void iterator_from_merged_iter(struct reftable_iterator *it,
+				      struct merged_iter *mi)
+{
+	assert(!it->ops);
+	it->iter_arg = mi;
+	it->ops = &merged_iter_vtable;
+}
+
+int reftable_new_merged_table(struct reftable_merged_table **dest,
+			      struct reftable_table *stack, int n,
+			      uint32_t hash_id)
+{
+	struct reftable_merged_table *m = NULL;
+	uint64_t last_max = 0;
+	uint64_t first_min = 0;
+	int i = 0;
+	for (i = 0; i < n; i++) {
+		uint64_t min = reftable_table_min_update_index(&stack[i]);
+		uint64_t max = reftable_table_max_update_index(&stack[i]);
+
+		if (reftable_table_hash_id(&stack[i]) != hash_id) {
+			return REFTABLE_FORMAT_ERROR;
+		}
+		if (i == 0 || min < first_min) {
+			first_min = min;
+		}
+		if (i == 0 || max > last_max) {
+			last_max = max;
+		}
+	}
+
+	m = reftable_calloc(sizeof(struct reftable_merged_table));
+	m->stack = stack;
+	m->stack_len = n;
+	m->min = first_min;
+	m->max = last_max;
+	m->hash_id = hash_id;
+	*dest = m;
+	return 0;
+}
+
+/* clears the list of subtable, without affecting the readers themselves. */
+void merged_table_release(struct reftable_merged_table *mt)
+{
+	FREE_AND_NULL(mt->stack);
+	mt->stack_len = 0;
+}
+
+void reftable_merged_table_free(struct reftable_merged_table *mt)
+{
+	if (!mt) {
+		return;
+	}
+	merged_table_release(mt);
+	reftable_free(mt);
+}
+
+uint64_t
+reftable_merged_table_max_update_index(struct reftable_merged_table *mt)
+{
+	return mt->max;
+}
+
+uint64_t
+reftable_merged_table_min_update_index(struct reftable_merged_table *mt)
+{
+	return mt->min;
+}
+
+static int reftable_table_seek_record(struct reftable_table *tab,
+				      struct reftable_iterator *it,
+				      struct reftable_record *rec)
+{
+	return tab->ops->seek_record(tab->table_arg, it, rec);
+}
+
+static int merged_table_seek_record(struct reftable_merged_table *mt,
+				    struct reftable_iterator *it,
+				    struct reftable_record *rec)
+{
+	struct reftable_iterator *iters = reftable_calloc(
+		sizeof(struct reftable_iterator) * mt->stack_len);
+	struct merged_iter merged = {
+		.stack = iters,
+		.typ = reftable_record_type(rec),
+		.hash_id = mt->hash_id,
+		.suppress_deletions = mt->suppress_deletions,
+	};
+	int n = 0;
+	int err = 0;
+	int i = 0;
+	for (i = 0; i < mt->stack_len && err == 0; i++) {
+		int e = reftable_table_seek_record(&mt->stack[i], &iters[n],
+						   rec);
+		if (e < 0) {
+			err = e;
+		}
+		if (e == 0) {
+			n++;
+		}
+	}
+	if (err < 0) {
+		int i = 0;
+		for (i = 0; i < n; i++) {
+			reftable_iterator_destroy(&iters[i]);
+		}
+		reftable_free(iters);
+		return err;
+	}
+
+	merged.stack_len = n;
+	err = merged_iter_init(&merged);
+	if (err < 0) {
+		merged_iter_close(&merged);
+		return err;
+	} else {
+		struct merged_iter *p =
+			reftable_malloc(sizeof(struct merged_iter));
+		*p = merged;
+		iterator_from_merged_iter(it, p);
+	}
+	return 0;
+}
+
+int reftable_merged_table_seek_ref(struct reftable_merged_table *mt,
+				   struct reftable_iterator *it,
+				   const char *name)
+{
+	struct reftable_ref_record ref = {
+		.refname = (char *)name,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, &ref);
+	return merged_table_seek_record(mt, it, &rec);
+}
+
+int reftable_merged_table_seek_log_at(struct reftable_merged_table *mt,
+				      struct reftable_iterator *it,
+				      const char *name, uint64_t update_index)
+{
+	struct reftable_log_record log = {
+		.refname = (char *)name,
+		.update_index = update_index,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_log(&rec, &log);
+	return merged_table_seek_record(mt, it, &rec);
+}
+
+int reftable_merged_table_seek_log(struct reftable_merged_table *mt,
+				   struct reftable_iterator *it,
+				   const char *name)
+{
+	uint64_t max = ~((uint64_t)0);
+	return reftable_merged_table_seek_log_at(mt, it, name, max);
+}
+
+uint32_t reftable_merged_table_hash_id(struct reftable_merged_table *mt)
+{
+	return mt->hash_id;
+}
+
+static int reftable_merged_table_seek_void(void *tab,
+					   struct reftable_iterator *it,
+					   struct reftable_record *rec)
+{
+	return merged_table_seek_record(tab, it, rec);
+}
+
+static uint32_t reftable_merged_table_hash_id_void(void *tab)
+{
+	return reftable_merged_table_hash_id(tab);
+}
+
+static uint64_t reftable_merged_table_min_update_index_void(void *tab)
+{
+	return reftable_merged_table_min_update_index(tab);
+}
+
+static uint64_t reftable_merged_table_max_update_index_void(void *tab)
+{
+	return reftable_merged_table_max_update_index(tab);
+}
+
+static struct reftable_table_vtable merged_table_vtable = {
+	.seek_record = reftable_merged_table_seek_void,
+	.hash_id = reftable_merged_table_hash_id_void,
+	.min_update_index = reftable_merged_table_min_update_index_void,
+	.max_update_index = reftable_merged_table_max_update_index_void,
+};
+
+void reftable_table_from_merged_table(struct reftable_table *tab,
+				      struct reftable_merged_table *merged)
+{
+	assert(!tab->ops);
+	tab->ops = &merged_table_vtable;
+	tab->table_arg = merged;
+}
diff --git a/reftable/merged.h b/reftable/merged.h
new file mode 100644
index 0000000..7d9f95d
--- /dev/null
+++ b/reftable/merged.h
@@ -0,0 +1,38 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef MERGED_H
+#define MERGED_H
+
+#include "pq.h"
+
+struct reftable_merged_table {
+	struct reftable_table *stack;
+	size_t stack_len;
+	uint32_t hash_id;
+
+	/* If unset, produce deletions. This is useful for compaction. For the
+	 * full stack, deletions should be produced. */
+	int suppress_deletions;
+
+	uint64_t min;
+	uint64_t max;
+};
+
+struct merged_iter {
+	struct reftable_iterator *stack;
+	uint32_t hash_id;
+	size_t stack_len;
+	uint8_t typ;
+	int suppress_deletions;
+	struct merged_iter_pqueue pq;
+};
+
+void merged_table_release(struct reftable_merged_table *mt);
+
+#endif
diff --git a/reftable/merged_test.c b/reftable/merged_test.c
new file mode 100644
index 0000000..24461e8
--- /dev/null
+++ b/reftable/merged_test.c
@@ -0,0 +1,468 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "merged.h"
+
+#include "system.h"
+
+#include "basics.h"
+#include "blocksource.h"
+#include "constants.h"
+#include "reader.h"
+#include "record.h"
+#include "test_framework.h"
+#include "reftable-merged.h"
+#include "reftable-tests.h"
+#include "reftable-generic.h"
+#include "reftable-writer.h"
+
+static void write_test_table(struct strbuf *buf,
+			     struct reftable_ref_record refs[], int n)
+{
+	int min = 0xffffffff;
+	int max = 0;
+	int i = 0;
+	int err;
+
+	struct reftable_write_options opts = {
+		.block_size = 256,
+	};
+	struct reftable_writer *w = NULL;
+	for (i = 0; i < n; i++) {
+		uint64_t ui = refs[i].update_index;
+		if (ui > max) {
+			max = ui;
+		}
+		if (ui < min) {
+			min = ui;
+		}
+	}
+
+	w = reftable_new_writer(&strbuf_add_void, buf, &opts);
+	reftable_writer_set_limits(w, min, max);
+
+	for (i = 0; i < n; i++) {
+		uint64_t before = refs[i].update_index;
+		int n = reftable_writer_add_ref(w, &refs[i]);
+		EXPECT(n == 0);
+		EXPECT(before == refs[i].update_index);
+	}
+
+	err = reftable_writer_close(w);
+	EXPECT_ERR(err);
+
+	reftable_writer_free(w);
+}
+
+static void write_test_log_table(struct strbuf *buf,
+				 struct reftable_log_record logs[], int n,
+				 uint64_t update_index)
+{
+	int i = 0;
+	int err;
+
+	struct reftable_write_options opts = {
+		.block_size = 256,
+		.exact_log_message = 1,
+	};
+	struct reftable_writer *w = NULL;
+	w = reftable_new_writer(&strbuf_add_void, buf, &opts);
+	reftable_writer_set_limits(w, update_index, update_index);
+
+	for (i = 0; i < n; i++) {
+		int err = reftable_writer_add_log(w, &logs[i]);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_writer_close(w);
+	EXPECT_ERR(err);
+
+	reftable_writer_free(w);
+}
+
+static struct reftable_merged_table *
+merged_table_from_records(struct reftable_ref_record **refs,
+			  struct reftable_block_source **source,
+			  struct reftable_reader ***readers, int *sizes,
+			  struct strbuf *buf, int n)
+{
+	int i = 0;
+	struct reftable_merged_table *mt = NULL;
+	int err;
+	struct reftable_table *tabs =
+		reftable_calloc(n * sizeof(struct reftable_table));
+	*readers = reftable_calloc(n * sizeof(struct reftable_reader *));
+	*source = reftable_calloc(n * sizeof(**source));
+	for (i = 0; i < n; i++) {
+		write_test_table(&buf[i], refs[i], sizes[i]);
+		block_source_from_strbuf(&(*source)[i], &buf[i]);
+
+		err = reftable_new_reader(&(*readers)[i], &(*source)[i],
+					  "name");
+		EXPECT_ERR(err);
+		reftable_table_from_reader(&tabs[i], (*readers)[i]);
+	}
+
+	err = reftable_new_merged_table(&mt, tabs, n, GIT_SHA1_FORMAT_ID);
+	EXPECT_ERR(err);
+	return mt;
+}
+
+static void readers_destroy(struct reftable_reader **readers, size_t n)
+{
+	int i = 0;
+	for (; i < n; i++)
+		reftable_reader_free(readers[i]);
+	reftable_free(readers);
+}
+
+static void test_merged_between(void)
+{
+	uint8_t hash1[GIT_SHA1_RAWSZ] = { 1, 2, 3, 0 };
+
+	struct reftable_ref_record r1[] = { {
+		.refname = "b",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_VAL1,
+		.value.val1 = hash1,
+	} };
+	struct reftable_ref_record r2[] = { {
+		.refname = "a",
+		.update_index = 2,
+		.value_type = REFTABLE_REF_DELETION,
+	} };
+
+	struct reftable_ref_record *refs[] = { r1, r2 };
+	int sizes[] = { 1, 1 };
+	struct strbuf bufs[2] = { STRBUF_INIT, STRBUF_INIT };
+	struct reftable_block_source *bs = NULL;
+	struct reftable_reader **readers = NULL;
+	struct reftable_merged_table *mt =
+		merged_table_from_records(refs, &bs, &readers, sizes, bufs, 2);
+	int i;
+	struct reftable_ref_record ref = { NULL };
+	struct reftable_iterator it = { NULL };
+	int err = reftable_merged_table_seek_ref(mt, &it, "a");
+	EXPECT_ERR(err);
+
+	err = reftable_iterator_next_ref(&it, &ref);
+	EXPECT_ERR(err);
+	EXPECT(ref.update_index == 2);
+	reftable_ref_record_release(&ref);
+	reftable_iterator_destroy(&it);
+	readers_destroy(readers, 2);
+	reftable_merged_table_free(mt);
+	for (i = 0; i < ARRAY_SIZE(bufs); i++) {
+		strbuf_release(&bufs[i]);
+	}
+	reftable_free(bs);
+}
+
+static void test_merged(void)
+{
+	uint8_t hash1[GIT_SHA1_RAWSZ] = { 1 };
+	uint8_t hash2[GIT_SHA1_RAWSZ] = { 2 };
+	struct reftable_ref_record r1[] = {
+		{
+			.refname = "a",
+			.update_index = 1,
+			.value_type = REFTABLE_REF_VAL1,
+			.value.val1 = hash1,
+		},
+		{
+			.refname = "b",
+			.update_index = 1,
+			.value_type = REFTABLE_REF_VAL1,
+			.value.val1 = hash1,
+		},
+		{
+			.refname = "c",
+			.update_index = 1,
+			.value_type = REFTABLE_REF_VAL1,
+			.value.val1 = hash1,
+		}
+	};
+	struct reftable_ref_record r2[] = { {
+		.refname = "a",
+		.update_index = 2,
+		.value_type = REFTABLE_REF_DELETION,
+	} };
+	struct reftable_ref_record r3[] = {
+		{
+			.refname = "c",
+			.update_index = 3,
+			.value_type = REFTABLE_REF_VAL1,
+			.value.val1 = hash2,
+		},
+		{
+			.refname = "d",
+			.update_index = 3,
+			.value_type = REFTABLE_REF_VAL1,
+			.value.val1 = hash1,
+		},
+	};
+
+	struct reftable_ref_record want[] = {
+		r2[0],
+		r1[1],
+		r3[0],
+		r3[1],
+	};
+
+	struct reftable_ref_record *refs[] = { r1, r2, r3 };
+	int sizes[3] = { 3, 1, 2 };
+	struct strbuf bufs[3] = { STRBUF_INIT, STRBUF_INIT, STRBUF_INIT };
+	struct reftable_block_source *bs = NULL;
+	struct reftable_reader **readers = NULL;
+	struct reftable_merged_table *mt =
+		merged_table_from_records(refs, &bs, &readers, sizes, bufs, 3);
+
+	struct reftable_iterator it = { NULL };
+	int err = reftable_merged_table_seek_ref(mt, &it, "a");
+	struct reftable_ref_record *out = NULL;
+	size_t len = 0;
+	size_t cap = 0;
+	int i = 0;
+
+	EXPECT_ERR(err);
+	EXPECT(reftable_merged_table_hash_id(mt) == GIT_SHA1_FORMAT_ID);
+	EXPECT(reftable_merged_table_min_update_index(mt) == 1);
+
+	while (len < 100) { /* cap loops/recursion. */
+		struct reftable_ref_record ref = { NULL };
+		int err = reftable_iterator_next_ref(&it, &ref);
+		if (err > 0) {
+			break;
+		}
+		if (len == cap) {
+			cap = 2 * cap + 1;
+			out = reftable_realloc(
+				out, sizeof(struct reftable_ref_record) * cap);
+		}
+		out[len++] = ref;
+	}
+	reftable_iterator_destroy(&it);
+
+	EXPECT(ARRAY_SIZE(want) == len);
+	for (i = 0; i < len; i++) {
+		EXPECT(reftable_ref_record_equal(&want[i], &out[i],
+						 GIT_SHA1_RAWSZ));
+	}
+	for (i = 0; i < len; i++) {
+		reftable_ref_record_release(&out[i]);
+	}
+	reftable_free(out);
+
+	for (i = 0; i < 3; i++) {
+		strbuf_release(&bufs[i]);
+	}
+	readers_destroy(readers, 3);
+	reftable_merged_table_free(mt);
+	reftable_free(bs);
+}
+
+static struct reftable_merged_table *
+merged_table_from_log_records(struct reftable_log_record **logs,
+			      struct reftable_block_source **source,
+			      struct reftable_reader ***readers, int *sizes,
+			      struct strbuf *buf, int n)
+{
+	int i = 0;
+	struct reftable_merged_table *mt = NULL;
+	int err;
+	struct reftable_table *tabs =
+		reftable_calloc(n * sizeof(struct reftable_table));
+	*readers = reftable_calloc(n * sizeof(struct reftable_reader *));
+	*source = reftable_calloc(n * sizeof(**source));
+	for (i = 0; i < n; i++) {
+		write_test_log_table(&buf[i], logs[i], sizes[i], i + 1);
+		block_source_from_strbuf(&(*source)[i], &buf[i]);
+
+		err = reftable_new_reader(&(*readers)[i], &(*source)[i],
+					  "name");
+		EXPECT_ERR(err);
+		reftable_table_from_reader(&tabs[i], (*readers)[i]);
+	}
+
+	err = reftable_new_merged_table(&mt, tabs, n, GIT_SHA1_FORMAT_ID);
+	EXPECT_ERR(err);
+	return mt;
+}
+
+static void test_merged_logs(void)
+{
+	uint8_t hash1[GIT_SHA1_RAWSZ] = { 1 };
+	uint8_t hash2[GIT_SHA1_RAWSZ] = { 2 };
+	uint8_t hash3[GIT_SHA1_RAWSZ] = { 3 };
+	struct reftable_log_record r1[] = {
+		{
+			.refname = "a",
+			.update_index = 2,
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value.update = {
+				.old_hash = hash2,
+				/* deletion */
+				.name = "jane doe",
+				.email = "jane@invalid",
+				.message = "message2",
+			}
+		},
+		{
+			.refname = "a",
+			.update_index = 1,
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value.update = {
+				.old_hash = hash1,
+				.new_hash = hash2,
+				.name = "jane doe",
+				.email = "jane@invalid",
+				.message = "message1",
+			}
+		},
+	};
+	struct reftable_log_record r2[] = {
+		{
+			.refname = "a",
+			.update_index = 3,
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value.update = {
+				.new_hash = hash3,
+				.name = "jane doe",
+				.email = "jane@invalid",
+				.message = "message3",
+			}
+		},
+	};
+	struct reftable_log_record r3[] = {
+		{
+			.refname = "a",
+			.update_index = 2,
+			.value_type = REFTABLE_LOG_DELETION,
+		},
+	};
+	struct reftable_log_record want[] = {
+		r2[0],
+		r3[0],
+		r1[1],
+	};
+
+	struct reftable_log_record *logs[] = { r1, r2, r3 };
+	int sizes[3] = { 2, 1, 1 };
+	struct strbuf bufs[3] = { STRBUF_INIT, STRBUF_INIT, STRBUF_INIT };
+	struct reftable_block_source *bs = NULL;
+	struct reftable_reader **readers = NULL;
+	struct reftable_merged_table *mt = merged_table_from_log_records(
+		logs, &bs, &readers, sizes, bufs, 3);
+
+	struct reftable_iterator it = { NULL };
+	int err = reftable_merged_table_seek_log(mt, &it, "a");
+	struct reftable_log_record *out = NULL;
+	size_t len = 0;
+	size_t cap = 0;
+	int i = 0;
+
+	EXPECT_ERR(err);
+	EXPECT(reftable_merged_table_hash_id(mt) == GIT_SHA1_FORMAT_ID);
+	EXPECT(reftable_merged_table_min_update_index(mt) == 1);
+
+	while (len < 100) { /* cap loops/recursion. */
+		struct reftable_log_record log = { NULL };
+		int err = reftable_iterator_next_log(&it, &log);
+		if (err > 0) {
+			break;
+		}
+		if (len == cap) {
+			cap = 2 * cap + 1;
+			out = reftable_realloc(
+				out, sizeof(struct reftable_log_record) * cap);
+		}
+		out[len++] = log;
+	}
+	reftable_iterator_destroy(&it);
+
+	EXPECT(ARRAY_SIZE(want) == len);
+	for (i = 0; i < len; i++) {
+		EXPECT(reftable_log_record_equal(&want[i], &out[i],
+						 GIT_SHA1_RAWSZ));
+	}
+
+	err = reftable_merged_table_seek_log_at(mt, &it, "a", 2);
+	EXPECT_ERR(err);
+	reftable_log_record_release(&out[0]);
+	err = reftable_iterator_next_log(&it, &out[0]);
+	EXPECT_ERR(err);
+	EXPECT(reftable_log_record_equal(&out[0], &r3[0], GIT_SHA1_RAWSZ));
+	reftable_iterator_destroy(&it);
+
+	for (i = 0; i < len; i++) {
+		reftable_log_record_release(&out[i]);
+	}
+	reftable_free(out);
+
+	for (i = 0; i < 3; i++) {
+		strbuf_release(&bufs[i]);
+	}
+	readers_destroy(readers, 3);
+	reftable_merged_table_free(mt);
+	reftable_free(bs);
+}
+
+static void test_default_write_opts(void)
+{
+	struct reftable_write_options opts = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+
+	struct reftable_ref_record rec = {
+		.refname = "master",
+		.update_index = 1,
+	};
+	int err;
+	struct reftable_block_source source = { NULL };
+	struct reftable_table *tab = reftable_calloc(sizeof(*tab) * 1);
+	uint32_t hash_id;
+	struct reftable_reader *rd = NULL;
+	struct reftable_merged_table *merged = NULL;
+
+	reftable_writer_set_limits(w, 1, 1);
+
+	err = reftable_writer_add_ref(w, &rec);
+	EXPECT_ERR(err);
+
+	err = reftable_writer_close(w);
+	EXPECT_ERR(err);
+	reftable_writer_free(w);
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = reftable_new_reader(&rd, &source, "filename");
+	EXPECT_ERR(err);
+
+	hash_id = reftable_reader_hash_id(rd);
+	EXPECT(hash_id == GIT_SHA1_FORMAT_ID);
+
+	reftable_table_from_reader(&tab[0], rd);
+	err = reftable_new_merged_table(&merged, tab, 1, GIT_SHA1_FORMAT_ID);
+	EXPECT_ERR(err);
+
+	reftable_reader_free(rd);
+	reftable_merged_table_free(merged);
+	strbuf_release(&buf);
+}
+
+/* XXX test refs_for(oid) */
+
+int merged_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_merged_logs);
+	RUN_TEST(test_merged_between);
+	RUN_TEST(test_merged);
+	RUN_TEST(test_default_write_opts);
+	return 0;
+}
diff --git a/reftable/pq.c b/reftable/pq.c
new file mode 100644
index 0000000..efc4740
--- /dev/null
+++ b/reftable/pq.c
@@ -0,0 +1,105 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "pq.h"
+
+#include "reftable-record.h"
+#include "system.h"
+#include "basics.h"
+
+int pq_less(struct pq_entry *a, struct pq_entry *b)
+{
+	struct strbuf ak = STRBUF_INIT;
+	struct strbuf bk = STRBUF_INIT;
+	int cmp = 0;
+	reftable_record_key(&a->rec, &ak);
+	reftable_record_key(&b->rec, &bk);
+
+	cmp = strbuf_cmp(&ak, &bk);
+
+	strbuf_release(&ak);
+	strbuf_release(&bk);
+
+	if (cmp == 0)
+		return a->index > b->index;
+
+	return cmp < 0;
+}
+
+struct pq_entry merged_iter_pqueue_top(struct merged_iter_pqueue pq)
+{
+	return pq.heap[0];
+}
+
+int merged_iter_pqueue_is_empty(struct merged_iter_pqueue pq)
+{
+	return pq.len == 0;
+}
+
+struct pq_entry merged_iter_pqueue_remove(struct merged_iter_pqueue *pq)
+{
+	int i = 0;
+	struct pq_entry e = pq->heap[0];
+	pq->heap[0] = pq->heap[pq->len - 1];
+	pq->len--;
+
+	i = 0;
+	while (i < pq->len) {
+		int min = i;
+		int j = 2 * i + 1;
+		int k = 2 * i + 2;
+		if (j < pq->len && pq_less(&pq->heap[j], &pq->heap[i])) {
+			min = j;
+		}
+		if (k < pq->len && pq_less(&pq->heap[k], &pq->heap[min])) {
+			min = k;
+		}
+
+		if (min == i) {
+			break;
+		}
+
+		SWAP(pq->heap[i], pq->heap[min]);
+		i = min;
+	}
+
+	return e;
+}
+
+void merged_iter_pqueue_add(struct merged_iter_pqueue *pq, struct pq_entry e)
+{
+	int i = 0;
+	if (pq->len == pq->cap) {
+		pq->cap = 2 * pq->cap + 1;
+		pq->heap = reftable_realloc(pq->heap,
+					    pq->cap * sizeof(struct pq_entry));
+	}
+
+	pq->heap[pq->len++] = e;
+	i = pq->len - 1;
+	while (i > 0) {
+		int j = (i - 1) / 2;
+		if (pq_less(&pq->heap[j], &pq->heap[i])) {
+			break;
+		}
+
+		SWAP(pq->heap[j], pq->heap[i]);
+
+		i = j;
+	}
+}
+
+void merged_iter_pqueue_release(struct merged_iter_pqueue *pq)
+{
+	int i = 0;
+	for (i = 0; i < pq->len; i++) {
+		reftable_record_destroy(&pq->heap[i].rec);
+	}
+	FREE_AND_NULL(pq->heap);
+	pq->len = pq->cap = 0;
+}
diff --git a/reftable/pq.h b/reftable/pq.h
new file mode 100644
index 0000000..56fc1b6
--- /dev/null
+++ b/reftable/pq.h
@@ -0,0 +1,33 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef PQ_H
+#define PQ_H
+
+#include "record.h"
+
+struct pq_entry {
+	int index;
+	struct reftable_record rec;
+};
+
+struct merged_iter_pqueue {
+	struct pq_entry *heap;
+	size_t len;
+	size_t cap;
+};
+
+struct pq_entry merged_iter_pqueue_top(struct merged_iter_pqueue pq);
+int merged_iter_pqueue_is_empty(struct merged_iter_pqueue pq);
+void merged_iter_pqueue_check(struct merged_iter_pqueue pq);
+struct pq_entry merged_iter_pqueue_remove(struct merged_iter_pqueue *pq);
+void merged_iter_pqueue_add(struct merged_iter_pqueue *pq, struct pq_entry e);
+void merged_iter_pqueue_release(struct merged_iter_pqueue *pq);
+int pq_less(struct pq_entry *a, struct pq_entry *b);
+
+#endif
diff --git a/reftable/pq_test.c b/reftable/pq_test.c
new file mode 100644
index 0000000..c9bb05e
--- /dev/null
+++ b/reftable/pq_test.c
@@ -0,0 +1,82 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+
+#include "basics.h"
+#include "constants.h"
+#include "pq.h"
+#include "record.h"
+#include "reftable-tests.h"
+#include "test_framework.h"
+
+void merged_iter_pqueue_check(struct merged_iter_pqueue pq)
+{
+	int i;
+	for (i = 1; i < pq.len; i++) {
+		int parent = (i - 1) / 2;
+
+		EXPECT(pq_less(&pq.heap[parent], &pq.heap[i]));
+	}
+}
+
+static void test_pq(void)
+{
+	char *names[54] = { NULL };
+	int N = ARRAY_SIZE(names) - 1;
+
+	struct merged_iter_pqueue pq = { NULL };
+	const char *last = NULL;
+
+	int i = 0;
+	for (i = 0; i < N; i++) {
+		char name[100];
+		snprintf(name, sizeof(name), "%02d", i);
+		names[i] = xstrdup(name);
+	}
+
+	i = 1;
+	do {
+		struct reftable_record rec =
+			reftable_new_record(BLOCK_TYPE_REF);
+		struct pq_entry e = { 0 };
+
+		reftable_record_as_ref(&rec)->refname = names[i];
+		e.rec = rec;
+		merged_iter_pqueue_add(&pq, e);
+		merged_iter_pqueue_check(pq);
+		i = (i * 7) % N;
+	} while (i != 1);
+
+	while (!merged_iter_pqueue_is_empty(pq)) {
+		struct pq_entry e = merged_iter_pqueue_remove(&pq);
+		struct reftable_ref_record *ref =
+			reftable_record_as_ref(&e.rec);
+
+		merged_iter_pqueue_check(pq);
+
+		if (last) {
+			EXPECT(strcmp(last, ref->refname) < 0);
+		}
+		last = ref->refname;
+		ref->refname = NULL;
+		reftable_free(ref);
+	}
+
+	for (i = 0; i < N; i++) {
+		reftable_free(names[i]);
+	}
+
+	merged_iter_pqueue_release(&pq);
+}
+
+int pq_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_pq);
+	return 0;
+}
diff --git a/reftable/publicbasics.c b/reftable/publicbasics.c
new file mode 100644
index 0000000..0ad7d5c
--- /dev/null
+++ b/reftable/publicbasics.c
@@ -0,0 +1,65 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "reftable-malloc.h"
+
+#include "basics.h"
+#include "system.h"
+
+static void *(*reftable_malloc_ptr)(size_t sz);
+static void *(*reftable_realloc_ptr)(void *, size_t);
+static void (*reftable_free_ptr)(void *);
+
+void *reftable_malloc(size_t sz)
+{
+	if (reftable_malloc_ptr)
+		return (*reftable_malloc_ptr)(sz);
+	return malloc(sz);
+}
+
+void *reftable_realloc(void *p, size_t sz)
+{
+	if (reftable_realloc_ptr)
+		return (*reftable_realloc_ptr)(p, sz);
+	return realloc(p, sz);
+}
+
+void reftable_free(void *p)
+{
+	if (reftable_free_ptr)
+		reftable_free_ptr(p);
+	else
+		free(p);
+}
+
+void *reftable_calloc(size_t sz)
+{
+	void *p = reftable_malloc(sz);
+	memset(p, 0, sz);
+	return p;
+}
+
+void reftable_set_alloc(void *(*malloc)(size_t),
+			void *(*realloc)(void *, size_t), void (*free)(void *))
+{
+	reftable_malloc_ptr = malloc;
+	reftable_realloc_ptr = realloc;
+	reftable_free_ptr = free;
+}
+
+int hash_size(uint32_t id)
+{
+	switch (id) {
+	case 0:
+	case GIT_SHA1_FORMAT_ID:
+		return GIT_SHA1_RAWSZ;
+	case GIT_SHA256_FORMAT_ID:
+		return GIT_SHA256_RAWSZ;
+	}
+	abort();
+}
diff --git a/reftable/reader.c b/reftable/reader.c
new file mode 100644
index 0000000..006709a
--- /dev/null
+++ b/reftable/reader.c
@@ -0,0 +1,801 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "reader.h"
+
+#include "system.h"
+#include "block.h"
+#include "constants.h"
+#include "generic.h"
+#include "iter.h"
+#include "record.h"
+#include "reftable-error.h"
+#include "reftable-generic.h"
+#include "tree.h"
+
+uint64_t block_source_size(struct reftable_block_source *source)
+{
+	return source->ops->size(source->arg);
+}
+
+int block_source_read_block(struct reftable_block_source *source,
+			    struct reftable_block *dest, uint64_t off,
+			    uint32_t size)
+{
+	int result = source->ops->read_block(source->arg, dest, off, size);
+	dest->source = *source;
+	return result;
+}
+
+void block_source_close(struct reftable_block_source *source)
+{
+	if (!source->ops) {
+		return;
+	}
+
+	source->ops->close(source->arg);
+	source->ops = NULL;
+}
+
+static struct reftable_reader_offsets *
+reader_offsets_for(struct reftable_reader *r, uint8_t typ)
+{
+	switch (typ) {
+	case BLOCK_TYPE_REF:
+		return &r->ref_offsets;
+	case BLOCK_TYPE_LOG:
+		return &r->log_offsets;
+	case BLOCK_TYPE_OBJ:
+		return &r->obj_offsets;
+	}
+	abort();
+}
+
+static int reader_get_block(struct reftable_reader *r,
+			    struct reftable_block *dest, uint64_t off,
+			    uint32_t sz)
+{
+	if (off >= r->size)
+		return 0;
+
+	if (off + sz > r->size) {
+		sz = r->size - off;
+	}
+
+	return block_source_read_block(&r->source, dest, off, sz);
+}
+
+uint32_t reftable_reader_hash_id(struct reftable_reader *r)
+{
+	return r->hash_id;
+}
+
+const char *reader_name(struct reftable_reader *r)
+{
+	return r->name;
+}
+
+static int parse_footer(struct reftable_reader *r, uint8_t *footer,
+			uint8_t *header)
+{
+	uint8_t *f = footer;
+	uint8_t first_block_typ;
+	int err = 0;
+	uint32_t computed_crc;
+	uint32_t file_crc;
+
+	if (memcmp(f, "REFT", 4)) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+	f += 4;
+
+	if (memcmp(footer, header, header_size(r->version))) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+
+	f++;
+	r->block_size = get_be24(f);
+
+	f += 3;
+	r->min_update_index = get_be64(f);
+	f += 8;
+	r->max_update_index = get_be64(f);
+	f += 8;
+
+	if (r->version == 1) {
+		r->hash_id = GIT_SHA1_FORMAT_ID;
+	} else {
+		r->hash_id = get_be32(f);
+		switch (r->hash_id) {
+		case GIT_SHA1_FORMAT_ID:
+			break;
+		case GIT_SHA256_FORMAT_ID:
+			break;
+		default:
+			err = REFTABLE_FORMAT_ERROR;
+			goto done;
+		}
+		f += 4;
+	}
+
+	r->ref_offsets.index_offset = get_be64(f);
+	f += 8;
+
+	r->obj_offsets.offset = get_be64(f);
+	f += 8;
+
+	r->object_id_len = r->obj_offsets.offset & ((1 << 5) - 1);
+	r->obj_offsets.offset >>= 5;
+
+	r->obj_offsets.index_offset = get_be64(f);
+	f += 8;
+	r->log_offsets.offset = get_be64(f);
+	f += 8;
+	r->log_offsets.index_offset = get_be64(f);
+	f += 8;
+
+	computed_crc = crc32(0, footer, f - footer);
+	file_crc = get_be32(f);
+	f += 4;
+	if (computed_crc != file_crc) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+
+	first_block_typ = header[header_size(r->version)];
+	r->ref_offsets.is_present = (first_block_typ == BLOCK_TYPE_REF);
+	r->ref_offsets.offset = 0;
+	r->log_offsets.is_present = (first_block_typ == BLOCK_TYPE_LOG ||
+				     r->log_offsets.offset > 0);
+	r->obj_offsets.is_present = r->obj_offsets.offset > 0;
+	err = 0;
+done:
+	return err;
+}
+
+int init_reader(struct reftable_reader *r, struct reftable_block_source *source,
+		const char *name)
+{
+	struct reftable_block footer = { NULL };
+	struct reftable_block header = { NULL };
+	int err = 0;
+	uint64_t file_size = block_source_size(source);
+
+	/* Need +1 to read type of first block. */
+	uint32_t read_size = header_size(2) + 1; /* read v2 because it's larger.  */
+	memset(r, 0, sizeof(struct reftable_reader));
+
+	if (read_size > file_size) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+
+	err = block_source_read_block(source, &header, 0, read_size);
+	if (err != read_size) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	if (memcmp(header.data, "REFT", 4)) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+	r->version = header.data[4];
+	if (r->version != 1 && r->version != 2) {
+		err = REFTABLE_FORMAT_ERROR;
+		goto done;
+	}
+
+	r->size = file_size - footer_size(r->version);
+	r->source = *source;
+	r->name = xstrdup(name);
+	r->hash_id = 0;
+
+	err = block_source_read_block(source, &footer, r->size,
+				      footer_size(r->version));
+	if (err != footer_size(r->version)) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	err = parse_footer(r, footer.data, header.data);
+done:
+	reftable_block_done(&footer);
+	reftable_block_done(&header);
+	return err;
+}
+
+struct table_iter {
+	struct reftable_reader *r;
+	uint8_t typ;
+	uint64_t block_off;
+	struct block_iter bi;
+	int is_finished;
+};
+#define TABLE_ITER_INIT                          \
+	{                                        \
+		.bi = {.last_key = STRBUF_INIT } \
+	}
+
+static void table_iter_copy_from(struct table_iter *dest,
+				 struct table_iter *src)
+{
+	dest->r = src->r;
+	dest->typ = src->typ;
+	dest->block_off = src->block_off;
+	dest->is_finished = src->is_finished;
+	block_iter_copy_from(&dest->bi, &src->bi);
+}
+
+static int table_iter_next_in_block(struct table_iter *ti,
+				    struct reftable_record *rec)
+{
+	int res = block_iter_next(&ti->bi, rec);
+	if (res == 0 && reftable_record_type(rec) == BLOCK_TYPE_REF) {
+		((struct reftable_ref_record *)rec->data)->update_index +=
+			ti->r->min_update_index;
+	}
+
+	return res;
+}
+
+static void table_iter_block_done(struct table_iter *ti)
+{
+	if (!ti->bi.br) {
+		return;
+	}
+	reftable_block_done(&ti->bi.br->block);
+	FREE_AND_NULL(ti->bi.br);
+
+	ti->bi.last_key.len = 0;
+	ti->bi.next_off = 0;
+}
+
+static int32_t extract_block_size(uint8_t *data, uint8_t *typ, uint64_t off,
+				  int version)
+{
+	int32_t result = 0;
+
+	if (off == 0) {
+		data += header_size(version);
+	}
+
+	*typ = data[0];
+	if (reftable_is_block_type(*typ)) {
+		result = get_be24(data + 1);
+	}
+	return result;
+}
+
+int reader_init_block_reader(struct reftable_reader *r, struct block_reader *br,
+			     uint64_t next_off, uint8_t want_typ)
+{
+	int32_t guess_block_size = r->block_size ? r->block_size :
+							 DEFAULT_BLOCK_SIZE;
+	struct reftable_block block = { NULL };
+	uint8_t block_typ = 0;
+	int err = 0;
+	uint32_t header_off = next_off ? 0 : header_size(r->version);
+	int32_t block_size = 0;
+
+	if (next_off >= r->size)
+		return 1;
+
+	err = reader_get_block(r, &block, next_off, guess_block_size);
+	if (err < 0)
+		return err;
+
+	block_size = extract_block_size(block.data, &block_typ, next_off,
+					r->version);
+	if (block_size < 0)
+		return block_size;
+
+	if (want_typ != BLOCK_TYPE_ANY && block_typ != want_typ) {
+		reftable_block_done(&block);
+		return 1;
+	}
+
+	if (block_size > guess_block_size) {
+		reftable_block_done(&block);
+		err = reader_get_block(r, &block, next_off, block_size);
+		if (err < 0) {
+			return err;
+		}
+	}
+
+	return block_reader_init(br, &block, header_off, r->block_size,
+				 hash_size(r->hash_id));
+}
+
+static int table_iter_next_block(struct table_iter *dest,
+				 struct table_iter *src)
+{
+	uint64_t next_block_off = src->block_off + src->bi.br->full_block_size;
+	struct block_reader br = { 0 };
+	int err = 0;
+
+	dest->r = src->r;
+	dest->typ = src->typ;
+	dest->block_off = next_block_off;
+
+	err = reader_init_block_reader(src->r, &br, next_block_off, src->typ);
+	if (err > 0) {
+		dest->is_finished = 1;
+		return 1;
+	}
+	if (err != 0)
+		return err;
+	else {
+		struct block_reader *brp =
+			reftable_malloc(sizeof(struct block_reader));
+		*brp = br;
+
+		dest->is_finished = 0;
+		block_reader_start(brp, &dest->bi);
+	}
+	return 0;
+}
+
+static int table_iter_next(struct table_iter *ti, struct reftable_record *rec)
+{
+	if (reftable_record_type(rec) != ti->typ)
+		return REFTABLE_API_ERROR;
+
+	while (1) {
+		struct table_iter next = TABLE_ITER_INIT;
+		int err = 0;
+		if (ti->is_finished) {
+			return 1;
+		}
+
+		err = table_iter_next_in_block(ti, rec);
+		if (err <= 0) {
+			return err;
+		}
+
+		err = table_iter_next_block(&next, ti);
+		if (err != 0) {
+			ti->is_finished = 1;
+		}
+		table_iter_block_done(ti);
+		if (err != 0) {
+			return err;
+		}
+		table_iter_copy_from(ti, &next);
+		block_iter_close(&next.bi);
+	}
+}
+
+static int table_iter_next_void(void *ti, struct reftable_record *rec)
+{
+	return table_iter_next(ti, rec);
+}
+
+static void table_iter_close(void *p)
+{
+	struct table_iter *ti = p;
+	table_iter_block_done(ti);
+	block_iter_close(&ti->bi);
+}
+
+static struct reftable_iterator_vtable table_iter_vtable = {
+	.next = &table_iter_next_void,
+	.close = &table_iter_close,
+};
+
+static void iterator_from_table_iter(struct reftable_iterator *it,
+				     struct table_iter *ti)
+{
+	assert(!it->ops);
+	it->iter_arg = ti;
+	it->ops = &table_iter_vtable;
+}
+
+static int reader_table_iter_at(struct reftable_reader *r,
+				struct table_iter *ti, uint64_t off,
+				uint8_t typ)
+{
+	struct block_reader br = { 0 };
+	struct block_reader *brp = NULL;
+
+	int err = reader_init_block_reader(r, &br, off, typ);
+	if (err != 0)
+		return err;
+
+	brp = reftable_malloc(sizeof(struct block_reader));
+	*brp = br;
+	ti->r = r;
+	ti->typ = block_reader_type(brp);
+	ti->block_off = off;
+	block_reader_start(brp, &ti->bi);
+	return 0;
+}
+
+static int reader_start(struct reftable_reader *r, struct table_iter *ti,
+			uint8_t typ, int index)
+{
+	struct reftable_reader_offsets *offs = reader_offsets_for(r, typ);
+	uint64_t off = offs->offset;
+	if (index) {
+		off = offs->index_offset;
+		if (off == 0) {
+			return 1;
+		}
+		typ = BLOCK_TYPE_INDEX;
+	}
+
+	return reader_table_iter_at(r, ti, off, typ);
+}
+
+static int reader_seek_linear(struct reftable_reader *r, struct table_iter *ti,
+			      struct reftable_record *want)
+{
+	struct reftable_record rec =
+		reftable_new_record(reftable_record_type(want));
+	struct strbuf want_key = STRBUF_INIT;
+	struct strbuf got_key = STRBUF_INIT;
+	struct table_iter next = TABLE_ITER_INIT;
+	int err = -1;
+
+	reftable_record_key(want, &want_key);
+
+	while (1) {
+		err = table_iter_next_block(&next, ti);
+		if (err < 0)
+			goto done;
+
+		if (err > 0) {
+			break;
+		}
+
+		err = block_reader_first_key(next.bi.br, &got_key);
+		if (err < 0)
+			goto done;
+
+		if (strbuf_cmp(&got_key, &want_key) > 0) {
+			table_iter_block_done(&next);
+			break;
+		}
+
+		table_iter_block_done(ti);
+		table_iter_copy_from(ti, &next);
+	}
+
+	err = block_iter_seek(&ti->bi, &want_key);
+	if (err < 0)
+		goto done;
+	err = 0;
+
+done:
+	block_iter_close(&next.bi);
+	reftable_record_destroy(&rec);
+	strbuf_release(&want_key);
+	strbuf_release(&got_key);
+	return err;
+}
+
+static int reader_seek_indexed(struct reftable_reader *r,
+			       struct reftable_iterator *it,
+			       struct reftable_record *rec)
+{
+	struct reftable_index_record want_index = { .last_key = STRBUF_INIT };
+	struct reftable_record want_index_rec = { NULL };
+	struct reftable_index_record index_result = { .last_key = STRBUF_INIT };
+	struct reftable_record index_result_rec = { NULL };
+	struct table_iter index_iter = TABLE_ITER_INIT;
+	struct table_iter next = TABLE_ITER_INIT;
+	int err = 0;
+
+	reftable_record_key(rec, &want_index.last_key);
+	reftable_record_from_index(&want_index_rec, &want_index);
+	reftable_record_from_index(&index_result_rec, &index_result);
+
+	err = reader_start(r, &index_iter, reftable_record_type(rec), 1);
+	if (err < 0)
+		goto done;
+
+	err = reader_seek_linear(r, &index_iter, &want_index_rec);
+	while (1) {
+		err = table_iter_next(&index_iter, &index_result_rec);
+		table_iter_block_done(&index_iter);
+		if (err != 0)
+			goto done;
+
+		err = reader_table_iter_at(r, &next, index_result.offset, 0);
+		if (err != 0)
+			goto done;
+
+		err = block_iter_seek(&next.bi, &want_index.last_key);
+		if (err < 0)
+			goto done;
+
+		if (next.typ == reftable_record_type(rec)) {
+			err = 0;
+			break;
+		}
+
+		if (next.typ != BLOCK_TYPE_INDEX) {
+			err = REFTABLE_FORMAT_ERROR;
+			break;
+		}
+
+		table_iter_copy_from(&index_iter, &next);
+	}
+
+	if (err == 0) {
+		struct table_iter empty = TABLE_ITER_INIT;
+		struct table_iter *malloced =
+			reftable_calloc(sizeof(struct table_iter));
+		*malloced = empty;
+		table_iter_copy_from(malloced, &next);
+		iterator_from_table_iter(it, malloced);
+	}
+done:
+	block_iter_close(&next.bi);
+	table_iter_close(&index_iter);
+	reftable_record_release(&want_index_rec);
+	reftable_record_release(&index_result_rec);
+	return err;
+}
+
+static int reader_seek_internal(struct reftable_reader *r,
+				struct reftable_iterator *it,
+				struct reftable_record *rec)
+{
+	struct reftable_reader_offsets *offs =
+		reader_offsets_for(r, reftable_record_type(rec));
+	uint64_t idx = offs->index_offset;
+	struct table_iter ti = TABLE_ITER_INIT;
+	int err = 0;
+	if (idx > 0)
+		return reader_seek_indexed(r, it, rec);
+
+	err = reader_start(r, &ti, reftable_record_type(rec), 0);
+	if (err < 0)
+		return err;
+	err = reader_seek_linear(r, &ti, rec);
+	if (err < 0)
+		return err;
+	else {
+		struct table_iter *p =
+			reftable_malloc(sizeof(struct table_iter));
+		*p = ti;
+		iterator_from_table_iter(it, p);
+	}
+
+	return 0;
+}
+
+static int reader_seek(struct reftable_reader *r, struct reftable_iterator *it,
+		       struct reftable_record *rec)
+{
+	uint8_t typ = reftable_record_type(rec);
+
+	struct reftable_reader_offsets *offs = reader_offsets_for(r, typ);
+	if (!offs->is_present) {
+		iterator_set_empty(it);
+		return 0;
+	}
+
+	return reader_seek_internal(r, it, rec);
+}
+
+int reftable_reader_seek_ref(struct reftable_reader *r,
+			     struct reftable_iterator *it, const char *name)
+{
+	struct reftable_ref_record ref = {
+		.refname = (char *)name,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, &ref);
+	return reader_seek(r, it, &rec);
+}
+
+int reftable_reader_seek_log_at(struct reftable_reader *r,
+				struct reftable_iterator *it, const char *name,
+				uint64_t update_index)
+{
+	struct reftable_log_record log = {
+		.refname = (char *)name,
+		.update_index = update_index,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_log(&rec, &log);
+	return reader_seek(r, it, &rec);
+}
+
+int reftable_reader_seek_log(struct reftable_reader *r,
+			     struct reftable_iterator *it, const char *name)
+{
+	uint64_t max = ~((uint64_t)0);
+	return reftable_reader_seek_log_at(r, it, name, max);
+}
+
+void reader_close(struct reftable_reader *r)
+{
+	block_source_close(&r->source);
+	FREE_AND_NULL(r->name);
+}
+
+int reftable_new_reader(struct reftable_reader **p,
+			struct reftable_block_source *src, char const *name)
+{
+	struct reftable_reader *rd =
+		reftable_calloc(sizeof(struct reftable_reader));
+	int err = init_reader(rd, src, name);
+	if (err == 0) {
+		*p = rd;
+	} else {
+		block_source_close(src);
+		reftable_free(rd);
+	}
+	return err;
+}
+
+void reftable_reader_free(struct reftable_reader *r)
+{
+	reader_close(r);
+	reftable_free(r);
+}
+
+static int reftable_reader_refs_for_indexed(struct reftable_reader *r,
+					    struct reftable_iterator *it,
+					    uint8_t *oid)
+{
+	struct reftable_obj_record want = {
+		.hash_prefix = oid,
+		.hash_prefix_len = r->object_id_len,
+	};
+	struct reftable_record want_rec = { NULL };
+	struct reftable_iterator oit = { NULL };
+	struct reftable_obj_record got = { NULL };
+	struct reftable_record got_rec = { NULL };
+	int err = 0;
+	struct indexed_table_ref_iter *itr = NULL;
+
+	/* Look through the reverse index. */
+	reftable_record_from_obj(&want_rec, &want);
+	err = reader_seek(r, &oit, &want_rec);
+	if (err != 0)
+		goto done;
+
+	/* read out the reftable_obj_record */
+	reftable_record_from_obj(&got_rec, &got);
+	err = iterator_next(&oit, &got_rec);
+	if (err < 0)
+		goto done;
+
+	if (err > 0 ||
+	    memcmp(want.hash_prefix, got.hash_prefix, r->object_id_len)) {
+		/* didn't find it; return empty iterator */
+		iterator_set_empty(it);
+		err = 0;
+		goto done;
+	}
+
+	err = new_indexed_table_ref_iter(&itr, r, oid, hash_size(r->hash_id),
+					 got.offsets, got.offset_len);
+	if (err < 0)
+		goto done;
+	got.offsets = NULL;
+	iterator_from_indexed_table_ref_iter(it, itr);
+
+done:
+	reftable_iterator_destroy(&oit);
+	reftable_record_release(&got_rec);
+	return err;
+}
+
+static int reftable_reader_refs_for_unindexed(struct reftable_reader *r,
+					      struct reftable_iterator *it,
+					      uint8_t *oid)
+{
+	struct table_iter ti_empty = TABLE_ITER_INIT;
+	struct table_iter *ti = reftable_calloc(sizeof(struct table_iter));
+	struct filtering_ref_iterator *filter = NULL;
+	struct filtering_ref_iterator empty = FILTERING_REF_ITERATOR_INIT;
+	int oid_len = hash_size(r->hash_id);
+	int err;
+
+	*ti = ti_empty;
+	err = reader_start(r, ti, BLOCK_TYPE_REF, 0);
+	if (err < 0) {
+		reftable_free(ti);
+		return err;
+	}
+
+	filter = reftable_malloc(sizeof(struct filtering_ref_iterator));
+	*filter = empty;
+
+	strbuf_add(&filter->oid, oid, oid_len);
+	reftable_table_from_reader(&filter->tab, r);
+	filter->double_check = 0;
+	iterator_from_table_iter(&filter->it, ti);
+
+	iterator_from_filtering_ref_iterator(it, filter);
+	return 0;
+}
+
+int reftable_reader_refs_for(struct reftable_reader *r,
+			     struct reftable_iterator *it, uint8_t *oid)
+{
+	if (r->obj_offsets.is_present)
+		return reftable_reader_refs_for_indexed(r, it, oid);
+	return reftable_reader_refs_for_unindexed(r, it, oid);
+}
+
+uint64_t reftable_reader_max_update_index(struct reftable_reader *r)
+{
+	return r->max_update_index;
+}
+
+uint64_t reftable_reader_min_update_index(struct reftable_reader *r)
+{
+	return r->min_update_index;
+}
+
+/* generic table interface. */
+
+static int reftable_reader_seek_void(void *tab, struct reftable_iterator *it,
+				     struct reftable_record *rec)
+{
+	return reader_seek(tab, it, rec);
+}
+
+static uint32_t reftable_reader_hash_id_void(void *tab)
+{
+	return reftable_reader_hash_id(tab);
+}
+
+static uint64_t reftable_reader_min_update_index_void(void *tab)
+{
+	return reftable_reader_min_update_index(tab);
+}
+
+static uint64_t reftable_reader_max_update_index_void(void *tab)
+{
+	return reftable_reader_max_update_index(tab);
+}
+
+static struct reftable_table_vtable reader_vtable = {
+	.seek_record = reftable_reader_seek_void,
+	.hash_id = reftable_reader_hash_id_void,
+	.min_update_index = reftable_reader_min_update_index_void,
+	.max_update_index = reftable_reader_max_update_index_void,
+};
+
+void reftable_table_from_reader(struct reftable_table *tab,
+				struct reftable_reader *reader)
+{
+	assert(!tab->ops);
+	tab->ops = &reader_vtable;
+	tab->table_arg = reader;
+}
+
+
+int reftable_reader_print_file(const char *tablename)
+{
+	struct reftable_block_source src = { NULL };
+	int err = reftable_block_source_from_file(&src, tablename);
+	struct reftable_reader *r = NULL;
+	struct reftable_table tab = { NULL };
+	if (err < 0)
+		goto done;
+
+	err = reftable_new_reader(&r, &src, tablename);
+	if (err < 0)
+		goto done;
+
+	reftable_table_from_reader(&tab, r);
+	err = reftable_table_print(&tab);
+done:
+	reftable_reader_free(r);
+	return err;
+}
diff --git a/reftable/reader.h b/reftable/reader.h
new file mode 100644
index 0000000..e869165
--- /dev/null
+++ b/reftable/reader.h
@@ -0,0 +1,64 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef READER_H
+#define READER_H
+
+#include "block.h"
+#include "record.h"
+#include "reftable-iterator.h"
+#include "reftable-reader.h"
+
+uint64_t block_source_size(struct reftable_block_source *source);
+
+int block_source_read_block(struct reftable_block_source *source,
+			    struct reftable_block *dest, uint64_t off,
+			    uint32_t size);
+void block_source_close(struct reftable_block_source *source);
+
+/* metadata for a block type */
+struct reftable_reader_offsets {
+	int is_present;
+	uint64_t offset;
+	uint64_t index_offset;
+};
+
+/* The state for reading a reftable file. */
+struct reftable_reader {
+	/* for convience, associate a name with the instance. */
+	char *name;
+	struct reftable_block_source source;
+
+	/* Size of the file, excluding the footer. */
+	uint64_t size;
+
+	/* 'sha1' for SHA1, 's256' for SHA-256 */
+	uint32_t hash_id;
+
+	uint32_t block_size;
+	uint64_t min_update_index;
+	uint64_t max_update_index;
+	/* Length of the OID keys in the 'o' section */
+	int object_id_len;
+	int version;
+
+	struct reftable_reader_offsets ref_offsets;
+	struct reftable_reader_offsets obj_offsets;
+	struct reftable_reader_offsets log_offsets;
+};
+
+int init_reader(struct reftable_reader *r, struct reftable_block_source *source,
+		const char *name);
+void reader_close(struct reftable_reader *r);
+const char *reader_name(struct reftable_reader *r);
+
+/* initialize a block reader to read from `r` */
+int reader_init_block_reader(struct reftable_reader *r, struct block_reader *br,
+			     uint64_t next_off, uint8_t want_typ);
+
+#endif
diff --git a/reftable/readwrite_test.c b/reftable/readwrite_test.c
new file mode 100644
index 0000000..5f6bcc2
--- /dev/null
+++ b/reftable/readwrite_test.c
@@ -0,0 +1,652 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+
+#include "basics.h"
+#include "block.h"
+#include "blocksource.h"
+#include "constants.h"
+#include "reader.h"
+#include "record.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+#include "reftable-writer.h"
+
+static const int update_index = 5;
+
+static void test_buffer(void)
+{
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_block_source source = { NULL };
+	struct reftable_block out = { NULL };
+	int n;
+	uint8_t in[] = "hello";
+	strbuf_add(&buf, in, sizeof(in));
+	block_source_from_strbuf(&source, &buf);
+	EXPECT(block_source_size(&source) == 6);
+	n = block_source_read_block(&source, &out, 0, sizeof(in));
+	EXPECT(n == sizeof(in));
+	EXPECT(!memcmp(in, out.data, n));
+	reftable_block_done(&out);
+
+	n = block_source_read_block(&source, &out, 1, 2);
+	EXPECT(n == 2);
+	EXPECT(!memcmp(out.data, "el", 2));
+
+	reftable_block_done(&out);
+	block_source_close(&source);
+	strbuf_release(&buf);
+}
+
+static void write_table(char ***names, struct strbuf *buf, int N,
+			int block_size, uint32_t hash_id)
+{
+	struct reftable_write_options opts = {
+		.block_size = block_size,
+		.hash_id = hash_id,
+	};
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, buf, &opts);
+	struct reftable_ref_record ref = { NULL };
+	int i = 0, n;
+	struct reftable_log_record log = { NULL };
+	const struct reftable_stats *stats = NULL;
+	*names = reftable_calloc(sizeof(char *) * (N + 1));
+	reftable_writer_set_limits(w, update_index, update_index);
+	for (i = 0; i < N; i++) {
+		uint8_t hash[GIT_SHA256_RAWSZ] = { 0 };
+		char name[100];
+		int n;
+
+		set_test_hash(hash, i);
+
+		snprintf(name, sizeof(name), "refs/heads/branch%02d", i);
+
+		ref.refname = name;
+		ref.update_index = update_index;
+		ref.value_type = REFTABLE_REF_VAL1;
+		ref.value.val1 = hash;
+		(*names)[i] = xstrdup(name);
+
+		n = reftable_writer_add_ref(w, &ref);
+		EXPECT(n == 0);
+	}
+
+	for (i = 0; i < N; i++) {
+		uint8_t hash[GIT_SHA256_RAWSZ] = { 0 };
+		char name[100];
+		int n;
+
+		set_test_hash(hash, i);
+
+		snprintf(name, sizeof(name), "refs/heads/branch%02d", i);
+
+		log.refname = name;
+		log.update_index = update_index;
+		log.value_type = REFTABLE_LOG_UPDATE;
+		log.value.update.new_hash = hash;
+		log.value.update.message = "message";
+
+		n = reftable_writer_add_log(w, &log);
+		EXPECT(n == 0);
+	}
+
+	n = reftable_writer_close(w);
+	EXPECT(n == 0);
+
+	stats = writer_stats(w);
+	for (i = 0; i < stats->ref_stats.blocks; i++) {
+		int off = i * opts.block_size;
+		if (off == 0) {
+			off = header_size(
+				(hash_id == GIT_SHA256_FORMAT_ID) ? 2 : 1);
+		}
+		EXPECT(buf->buf[off] == 'r');
+	}
+
+	EXPECT(stats->log_stats.blocks > 0);
+	reftable_writer_free(w);
+}
+
+static void test_log_buffer_size(void)
+{
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_write_options opts = {
+		.block_size = 4096,
+	};
+	int err;
+	int i;
+	struct reftable_log_record
+		log = { .refname = "refs/heads/master",
+			.update_index = 0xa,
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value = { .update = {
+					   .name = "Han-Wen Nienhuys",
+					   .email = "hanwen@google.com",
+					   .tz_offset = 100,
+					   .time = 0x5e430672,
+					   .message = "commit: 9\n",
+				   } } };
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+
+	/* This tests buffer extension for log compression. Must use a random
+	   hash, to ensure that the compressed part is larger than the original.
+	*/
+	uint8_t hash1[GIT_SHA1_RAWSZ], hash2[GIT_SHA1_RAWSZ];
+	for (i = 0; i < GIT_SHA1_RAWSZ; i++) {
+		hash1[i] = (uint8_t)(rand() % 256);
+		hash2[i] = (uint8_t)(rand() % 256);
+	}
+	log.value.update.old_hash = hash1;
+	log.value.update.new_hash = hash2;
+	reftable_writer_set_limits(w, update_index, update_index);
+	err = reftable_writer_add_log(w, &log);
+	EXPECT_ERR(err);
+	err = reftable_writer_close(w);
+	EXPECT_ERR(err);
+	reftable_writer_free(w);
+	strbuf_release(&buf);
+}
+
+static void test_log_write_read(void)
+{
+	int N = 2;
+	char **names = reftable_calloc(sizeof(char *) * (N + 1));
+	int err;
+	struct reftable_write_options opts = {
+		.block_size = 256,
+	};
+	struct reftable_ref_record ref = { NULL };
+	int i = 0;
+	struct reftable_log_record log = { NULL };
+	int n;
+	struct reftable_iterator it = { NULL };
+	struct reftable_reader rd = { NULL };
+	struct reftable_block_source source = { NULL };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+	const struct reftable_stats *stats = NULL;
+	reftable_writer_set_limits(w, 0, N);
+	for (i = 0; i < N; i++) {
+		char name[256];
+		struct reftable_ref_record ref = { NULL };
+		snprintf(name, sizeof(name), "b%02d%0*d", i, 130, 7);
+		names[i] = xstrdup(name);
+		ref.refname = name;
+		ref.update_index = i;
+
+		err = reftable_writer_add_ref(w, &ref);
+		EXPECT_ERR(err);
+	}
+	for (i = 0; i < N; i++) {
+		uint8_t hash1[GIT_SHA1_RAWSZ], hash2[GIT_SHA1_RAWSZ];
+		struct reftable_log_record log = { NULL };
+		set_test_hash(hash1, i);
+		set_test_hash(hash2, i + 1);
+
+		log.refname = names[i];
+		log.update_index = i;
+		log.value_type = REFTABLE_LOG_UPDATE;
+		log.value.update.old_hash = hash1;
+		log.value.update.new_hash = hash2;
+
+		err = reftable_writer_add_log(w, &log);
+		EXPECT_ERR(err);
+	}
+
+	n = reftable_writer_close(w);
+	EXPECT(n == 0);
+
+	stats = writer_stats(w);
+	EXPECT(stats->log_stats.blocks > 0);
+	reftable_writer_free(w);
+	w = NULL;
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = init_reader(&rd, &source, "file.log");
+	EXPECT_ERR(err);
+
+	err = reftable_reader_seek_ref(&rd, &it, names[N - 1]);
+	EXPECT_ERR(err);
+
+	err = reftable_iterator_next_ref(&it, &ref);
+	EXPECT_ERR(err);
+
+	/* end of iteration. */
+	err = reftable_iterator_next_ref(&it, &ref);
+	EXPECT(0 < err);
+
+	reftable_iterator_destroy(&it);
+	reftable_ref_record_release(&ref);
+
+	err = reftable_reader_seek_log(&rd, &it, "");
+	EXPECT_ERR(err);
+
+	i = 0;
+	while (1) {
+		int err = reftable_iterator_next_log(&it, &log);
+		if (err > 0) {
+			break;
+		}
+
+		EXPECT_ERR(err);
+		EXPECT_STREQ(names[i], log.refname);
+		EXPECT(i == log.update_index);
+		i++;
+		reftable_log_record_release(&log);
+	}
+
+	EXPECT(i == N);
+	reftable_iterator_destroy(&it);
+
+	/* cleanup. */
+	strbuf_release(&buf);
+	free_names(names);
+	reader_close(&rd);
+}
+
+static void test_table_read_write_sequential(void)
+{
+	char **names;
+	struct strbuf buf = STRBUF_INIT;
+	int N = 50;
+	struct reftable_iterator it = { NULL };
+	struct reftable_block_source source = { NULL };
+	struct reftable_reader rd = { NULL };
+	int err = 0;
+	int j = 0;
+
+	write_table(&names, &buf, N, 256, GIT_SHA1_FORMAT_ID);
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = init_reader(&rd, &source, "file.ref");
+	EXPECT_ERR(err);
+
+	err = reftable_reader_seek_ref(&rd, &it, "");
+	EXPECT_ERR(err);
+
+	while (1) {
+		struct reftable_ref_record ref = { NULL };
+		int r = reftable_iterator_next_ref(&it, &ref);
+		EXPECT(r >= 0);
+		if (r > 0) {
+			break;
+		}
+		EXPECT(0 == strcmp(names[j], ref.refname));
+		EXPECT(update_index == ref.update_index);
+
+		j++;
+		reftable_ref_record_release(&ref);
+	}
+	EXPECT(j == N);
+	reftable_iterator_destroy(&it);
+	strbuf_release(&buf);
+	free_names(names);
+
+	reader_close(&rd);
+}
+
+static void test_table_write_small_table(void)
+{
+	char **names;
+	struct strbuf buf = STRBUF_INIT;
+	int N = 1;
+	write_table(&names, &buf, N, 4096, GIT_SHA1_FORMAT_ID);
+	EXPECT(buf.len < 200);
+	strbuf_release(&buf);
+	free_names(names);
+}
+
+static void test_table_read_api(void)
+{
+	char **names;
+	struct strbuf buf = STRBUF_INIT;
+	int N = 50;
+	struct reftable_reader rd = { NULL };
+	struct reftable_block_source source = { NULL };
+	int err;
+	int i;
+	struct reftable_log_record log = { NULL };
+	struct reftable_iterator it = { NULL };
+
+	write_table(&names, &buf, N, 256, GIT_SHA1_FORMAT_ID);
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = init_reader(&rd, &source, "file.ref");
+	EXPECT_ERR(err);
+
+	err = reftable_reader_seek_ref(&rd, &it, names[0]);
+	EXPECT_ERR(err);
+
+	err = reftable_iterator_next_log(&it, &log);
+	EXPECT(err == REFTABLE_API_ERROR);
+
+	strbuf_release(&buf);
+	for (i = 0; i < N; i++) {
+		reftable_free(names[i]);
+	}
+	reftable_iterator_destroy(&it);
+	reftable_free(names);
+	reader_close(&rd);
+	strbuf_release(&buf);
+}
+
+static void test_table_read_write_seek(int index, int hash_id)
+{
+	char **names;
+	struct strbuf buf = STRBUF_INIT;
+	int N = 50;
+	struct reftable_reader rd = { NULL };
+	struct reftable_block_source source = { NULL };
+	int err;
+	int i = 0;
+
+	struct reftable_iterator it = { NULL };
+	struct strbuf pastLast = STRBUF_INIT;
+	struct reftable_ref_record ref = { NULL };
+
+	write_table(&names, &buf, N, 256, hash_id);
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = init_reader(&rd, &source, "file.ref");
+	EXPECT_ERR(err);
+	EXPECT(hash_id == reftable_reader_hash_id(&rd));
+
+	if (!index) {
+		rd.ref_offsets.index_offset = 0;
+	} else {
+		EXPECT(rd.ref_offsets.index_offset > 0);
+	}
+
+	for (i = 1; i < N; i++) {
+		int err = reftable_reader_seek_ref(&rd, &it, names[i]);
+		EXPECT_ERR(err);
+		err = reftable_iterator_next_ref(&it, &ref);
+		EXPECT_ERR(err);
+		EXPECT(0 == strcmp(names[i], ref.refname));
+		EXPECT(REFTABLE_REF_VAL1 == ref.value_type);
+		EXPECT(i == ref.value.val1[0]);
+
+		reftable_ref_record_release(&ref);
+		reftable_iterator_destroy(&it);
+	}
+
+	strbuf_addstr(&pastLast, names[N - 1]);
+	strbuf_addstr(&pastLast, "/");
+
+	err = reftable_reader_seek_ref(&rd, &it, pastLast.buf);
+	if (err == 0) {
+		struct reftable_ref_record ref = { NULL };
+		int err = reftable_iterator_next_ref(&it, &ref);
+		EXPECT(err > 0);
+	} else {
+		EXPECT(err > 0);
+	}
+
+	strbuf_release(&pastLast);
+	reftable_iterator_destroy(&it);
+
+	strbuf_release(&buf);
+	for (i = 0; i < N; i++) {
+		reftable_free(names[i]);
+	}
+	reftable_free(names);
+	reader_close(&rd);
+}
+
+static void test_table_read_write_seek_linear(void)
+{
+	test_table_read_write_seek(0, GIT_SHA1_FORMAT_ID);
+}
+
+static void test_table_read_write_seek_linear_sha256(void)
+{
+	test_table_read_write_seek(0, GIT_SHA256_FORMAT_ID);
+}
+
+static void test_table_read_write_seek_index(void)
+{
+	test_table_read_write_seek(1, GIT_SHA1_FORMAT_ID);
+}
+
+static void test_table_refs_for(int indexed)
+{
+	int N = 50;
+	char **want_names = reftable_calloc(sizeof(char *) * (N + 1));
+	int want_names_len = 0;
+	uint8_t want_hash[GIT_SHA1_RAWSZ];
+
+	struct reftable_write_options opts = {
+		.block_size = 256,
+	};
+	struct reftable_ref_record ref = { NULL };
+	int i = 0;
+	int n;
+	int err;
+	struct reftable_reader rd;
+	struct reftable_block_source source = { NULL };
+
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+
+	struct reftable_iterator it = { NULL };
+	int j;
+
+	set_test_hash(want_hash, 4);
+
+	for (i = 0; i < N; i++) {
+		uint8_t hash[GIT_SHA1_RAWSZ];
+		char fill[51] = { 0 };
+		char name[100];
+		uint8_t hash1[GIT_SHA1_RAWSZ];
+		uint8_t hash2[GIT_SHA1_RAWSZ];
+		struct reftable_ref_record ref = { NULL };
+
+		memset(hash, i, sizeof(hash));
+		memset(fill, 'x', 50);
+		/* Put the variable part in the start */
+		snprintf(name, sizeof(name), "br%02d%s", i, fill);
+		name[40] = 0;
+		ref.refname = name;
+
+		set_test_hash(hash1, i / 4);
+		set_test_hash(hash2, 3 + i / 4);
+		ref.value_type = REFTABLE_REF_VAL2;
+		ref.value.val2.value = hash1;
+		ref.value.val2.target_value = hash2;
+
+		/* 80 bytes / entry, so 3 entries per block. Yields 17
+		 */
+		/* blocks. */
+		n = reftable_writer_add_ref(w, &ref);
+		EXPECT(n == 0);
+
+		if (!memcmp(hash1, want_hash, GIT_SHA1_RAWSZ) ||
+		    !memcmp(hash2, want_hash, GIT_SHA1_RAWSZ)) {
+			want_names[want_names_len++] = xstrdup(name);
+		}
+	}
+
+	n = reftable_writer_close(w);
+	EXPECT(n == 0);
+
+	reftable_writer_free(w);
+	w = NULL;
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = init_reader(&rd, &source, "file.ref");
+	EXPECT_ERR(err);
+	if (!indexed) {
+		rd.obj_offsets.is_present = 0;
+	}
+
+	err = reftable_reader_seek_ref(&rd, &it, "");
+	EXPECT_ERR(err);
+	reftable_iterator_destroy(&it);
+
+	err = reftable_reader_refs_for(&rd, &it, want_hash);
+	EXPECT_ERR(err);
+
+	j = 0;
+	while (1) {
+		int err = reftable_iterator_next_ref(&it, &ref);
+		EXPECT(err >= 0);
+		if (err > 0) {
+			break;
+		}
+
+		EXPECT(j < want_names_len);
+		EXPECT(0 == strcmp(ref.refname, want_names[j]));
+		j++;
+		reftable_ref_record_release(&ref);
+	}
+	EXPECT(j == want_names_len);
+
+	strbuf_release(&buf);
+	free_names(want_names);
+	reftable_iterator_destroy(&it);
+	reader_close(&rd);
+}
+
+static void test_table_refs_for_no_index(void)
+{
+	test_table_refs_for(0);
+}
+
+static void test_table_refs_for_obj_index(void)
+{
+	test_table_refs_for(1);
+}
+
+static void test_write_empty_table(void)
+{
+	struct reftable_write_options opts = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+	struct reftable_block_source source = { NULL };
+	struct reftable_reader *rd = NULL;
+	struct reftable_ref_record rec = { NULL };
+	struct reftable_iterator it = { NULL };
+	int err;
+
+	reftable_writer_set_limits(w, 1, 1);
+
+	err = reftable_writer_close(w);
+	EXPECT(err == REFTABLE_EMPTY_TABLE_ERROR);
+	reftable_writer_free(w);
+
+	EXPECT(buf.len == header_size(1) + footer_size(1));
+
+	block_source_from_strbuf(&source, &buf);
+
+	err = reftable_new_reader(&rd, &source, "filename");
+	EXPECT_ERR(err);
+
+	err = reftable_reader_seek_ref(rd, &it, "");
+	EXPECT_ERR(err);
+
+	err = reftable_iterator_next_ref(&it, &rec);
+	EXPECT(err > 0);
+
+	reftable_iterator_destroy(&it);
+	reftable_reader_free(rd);
+	strbuf_release(&buf);
+}
+
+static void test_write_key_order(void)
+{
+	struct reftable_write_options opts = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+	struct reftable_ref_record refs[2] = {
+		{
+			.refname = "b",
+			.update_index = 1,
+			.value_type = REFTABLE_REF_SYMREF,
+			.value = {
+				.symref = "target",
+			},
+		}, {
+			.refname = "a",
+			.update_index = 1,
+			.value_type = REFTABLE_REF_SYMREF,
+			.value = {
+				.symref = "target",
+			},
+		}
+	};
+	int err;
+
+	reftable_writer_set_limits(w, 1, 1);
+	err = reftable_writer_add_ref(w, &refs[0]);
+	EXPECT_ERR(err);
+	err = reftable_writer_add_ref(w, &refs[1]);
+	printf("%d\n", err);
+	EXPECT(err == REFTABLE_API_ERROR);
+	reftable_writer_close(w);
+	reftable_writer_free(w);
+	strbuf_release(&buf);
+}
+
+static void test_corrupt_table_empty(void)
+{
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_block_source source = { NULL };
+	struct reftable_reader rd = { NULL };
+	int err;
+
+	block_source_from_strbuf(&source, &buf);
+	err = init_reader(&rd, &source, "file.log");
+	EXPECT(err == REFTABLE_FORMAT_ERROR);
+}
+
+static void test_corrupt_table(void)
+{
+	uint8_t zeros[1024] = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_block_source source = { NULL };
+	struct reftable_reader rd = { NULL };
+	int err;
+	strbuf_add(&buf, zeros, sizeof(zeros));
+
+	block_source_from_strbuf(&source, &buf);
+	err = init_reader(&rd, &source, "file.log");
+	EXPECT(err == REFTABLE_FORMAT_ERROR);
+	strbuf_release(&buf);
+}
+
+int readwrite_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_corrupt_table);
+	RUN_TEST(test_corrupt_table_empty);
+	RUN_TEST(test_log_write_read);
+	RUN_TEST(test_write_key_order);
+	RUN_TEST(test_table_read_write_seek_linear_sha256);
+	RUN_TEST(test_log_buffer_size);
+	RUN_TEST(test_table_write_small_table);
+	RUN_TEST(test_buffer);
+	RUN_TEST(test_table_read_api);
+	RUN_TEST(test_table_read_write_sequential);
+	RUN_TEST(test_table_read_write_seek_linear);
+	RUN_TEST(test_table_read_write_seek_index);
+	RUN_TEST(test_table_refs_for_no_index);
+	RUN_TEST(test_table_refs_for_obj_index);
+	RUN_TEST(test_write_empty_table);
+	return 0;
+}
diff --git a/reftable/record.c b/reftable/record.c
new file mode 100644
index 0000000..6a5dac3
--- /dev/null
+++ b/reftable/record.c
@@ -0,0 +1,1212 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+/* record.c - methods for different types of records. */
+
+#include "record.h"
+
+#include "system.h"
+#include "constants.h"
+#include "reftable-error.h"
+#include "basics.h"
+
+int get_var_int(uint64_t *dest, struct string_view *in)
+{
+	int ptr = 0;
+	uint64_t val;
+
+	if (in->len == 0)
+		return -1;
+	val = in->buf[ptr] & 0x7f;
+
+	while (in->buf[ptr] & 0x80) {
+		ptr++;
+		if (ptr > in->len) {
+			return -1;
+		}
+		val = (val + 1) << 7 | (uint64_t)(in->buf[ptr] & 0x7f);
+	}
+
+	*dest = val;
+	return ptr + 1;
+}
+
+int put_var_int(struct string_view *dest, uint64_t val)
+{
+	uint8_t buf[10] = { 0 };
+	int i = 9;
+	int n = 0;
+	buf[i] = (uint8_t)(val & 0x7f);
+	i--;
+	while (1) {
+		val >>= 7;
+		if (!val) {
+			break;
+		}
+		val--;
+		buf[i] = 0x80 | (uint8_t)(val & 0x7f);
+		i--;
+	}
+
+	n = sizeof(buf) - i - 1;
+	if (dest->len < n)
+		return -1;
+	memcpy(dest->buf, &buf[i + 1], n);
+	return n;
+}
+
+int reftable_is_block_type(uint8_t typ)
+{
+	switch (typ) {
+	case BLOCK_TYPE_REF:
+	case BLOCK_TYPE_LOG:
+	case BLOCK_TYPE_OBJ:
+	case BLOCK_TYPE_INDEX:
+		return 1;
+	}
+	return 0;
+}
+
+uint8_t *reftable_ref_record_val1(struct reftable_ref_record *rec)
+{
+	switch (rec->value_type) {
+	case REFTABLE_REF_VAL1:
+		return rec->value.val1;
+	case REFTABLE_REF_VAL2:
+		return rec->value.val2.value;
+	default:
+		return NULL;
+	}
+}
+
+uint8_t *reftable_ref_record_val2(struct reftable_ref_record *rec)
+{
+	switch (rec->value_type) {
+	case REFTABLE_REF_VAL2:
+		return rec->value.val2.target_value;
+	default:
+		return NULL;
+	}
+}
+
+static int decode_string(struct strbuf *dest, struct string_view in)
+{
+	int start_len = in.len;
+	uint64_t tsize = 0;
+	int n = get_var_int(&tsize, &in);
+	if (n <= 0)
+		return -1;
+	string_view_consume(&in, n);
+	if (in.len < tsize)
+		return -1;
+
+	strbuf_reset(dest);
+	strbuf_add(dest, in.buf, tsize);
+	string_view_consume(&in, tsize);
+
+	return start_len - in.len;
+}
+
+static int encode_string(char *str, struct string_view s)
+{
+	struct string_view start = s;
+	int l = strlen(str);
+	int n = put_var_int(&s, l);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+	if (s.len < l)
+		return -1;
+	memcpy(s.buf, str, l);
+	string_view_consume(&s, l);
+
+	return start.len - s.len;
+}
+
+int reftable_encode_key(int *restart, struct string_view dest,
+			struct strbuf prev_key, struct strbuf key,
+			uint8_t extra)
+{
+	struct string_view start = dest;
+	int prefix_len = common_prefix_size(&prev_key, &key);
+	uint64_t suffix_len = key.len - prefix_len;
+	int n = put_var_int(&dest, (uint64_t)prefix_len);
+	if (n < 0)
+		return -1;
+	string_view_consume(&dest, n);
+
+	*restart = (prefix_len == 0);
+
+	n = put_var_int(&dest, suffix_len << 3 | (uint64_t)extra);
+	if (n < 0)
+		return -1;
+	string_view_consume(&dest, n);
+
+	if (dest.len < suffix_len)
+		return -1;
+	memcpy(dest.buf, key.buf + prefix_len, suffix_len);
+	string_view_consume(&dest, suffix_len);
+
+	return start.len - dest.len;
+}
+
+int reftable_decode_key(struct strbuf *key, uint8_t *extra,
+			struct strbuf last_key, struct string_view in)
+{
+	int start_len = in.len;
+	uint64_t prefix_len = 0;
+	uint64_t suffix_len = 0;
+	int n = get_var_int(&prefix_len, &in);
+	if (n < 0)
+		return -1;
+	string_view_consume(&in, n);
+
+	if (prefix_len > last_key.len)
+		return -1;
+
+	n = get_var_int(&suffix_len, &in);
+	if (n <= 0)
+		return -1;
+	string_view_consume(&in, n);
+
+	*extra = (uint8_t)(suffix_len & 0x7);
+	suffix_len >>= 3;
+
+	if (in.len < suffix_len)
+		return -1;
+
+	strbuf_reset(key);
+	strbuf_add(key, last_key.buf, prefix_len);
+	strbuf_add(key, in.buf, suffix_len);
+	string_view_consume(&in, suffix_len);
+
+	return start_len - in.len;
+}
+
+static void reftable_ref_record_key(const void *r, struct strbuf *dest)
+{
+	const struct reftable_ref_record *rec =
+		(const struct reftable_ref_record *)r;
+	strbuf_reset(dest);
+	strbuf_addstr(dest, rec->refname);
+}
+
+static void reftable_ref_record_copy_from(void *rec, const void *src_rec,
+					  int hash_size)
+{
+	struct reftable_ref_record *ref = rec;
+	const struct reftable_ref_record *src = src_rec;
+	assert(hash_size > 0);
+
+	/* This is simple and correct, but we could probably reuse the hash
+	 * fields. */
+	reftable_ref_record_release(ref);
+	if (src->refname) {
+		ref->refname = xstrdup(src->refname);
+	}
+	ref->update_index = src->update_index;
+	ref->value_type = src->value_type;
+	switch (src->value_type) {
+	case REFTABLE_REF_DELETION:
+		break;
+	case REFTABLE_REF_VAL1:
+		ref->value.val1 = reftable_malloc(hash_size);
+		memcpy(ref->value.val1, src->value.val1, hash_size);
+		break;
+	case REFTABLE_REF_VAL2:
+		ref->value.val2.value = reftable_malloc(hash_size);
+		memcpy(ref->value.val2.value, src->value.val2.value, hash_size);
+		ref->value.val2.target_value = reftable_malloc(hash_size);
+		memcpy(ref->value.val2.target_value,
+		       src->value.val2.target_value, hash_size);
+		break;
+	case REFTABLE_REF_SYMREF:
+		ref->value.symref = xstrdup(src->value.symref);
+		break;
+	}
+}
+
+static char hexdigit(int c)
+{
+	if (c <= 9)
+		return '0' + c;
+	return 'a' + (c - 10);
+}
+
+static void hex_format(char *dest, uint8_t *src, int hash_size)
+{
+	assert(hash_size > 0);
+	if (src) {
+		int i = 0;
+		for (i = 0; i < hash_size; i++) {
+			dest[2 * i] = hexdigit(src[i] >> 4);
+			dest[2 * i + 1] = hexdigit(src[i] & 0xf);
+		}
+		dest[2 * hash_size] = 0;
+	}
+}
+
+void reftable_ref_record_print(struct reftable_ref_record *ref,
+			       uint32_t hash_id)
+{
+	char hex[2 * GIT_SHA256_RAWSZ + 1] = { 0 }; /* BUG */
+	printf("ref{%s(%" PRIu64 ") ", ref->refname, ref->update_index);
+	switch (ref->value_type) {
+	case REFTABLE_REF_SYMREF:
+		printf("=> %s", ref->value.symref);
+		break;
+	case REFTABLE_REF_VAL2:
+		hex_format(hex, ref->value.val2.value, hash_size(hash_id));
+		printf("val 2 %s", hex);
+		hex_format(hex, ref->value.val2.target_value,
+			   hash_size(hash_id));
+		printf("(T %s)", hex);
+		break;
+	case REFTABLE_REF_VAL1:
+		hex_format(hex, ref->value.val1, hash_size(hash_id));
+		printf("val 1 %s", hex);
+		break;
+	case REFTABLE_REF_DELETION:
+		printf("delete");
+		break;
+	}
+	printf("}\n");
+}
+
+static void reftable_ref_record_release_void(void *rec)
+{
+	reftable_ref_record_release(rec);
+}
+
+void reftable_ref_record_release(struct reftable_ref_record *ref)
+{
+	switch (ref->value_type) {
+	case REFTABLE_REF_SYMREF:
+		reftable_free(ref->value.symref);
+		break;
+	case REFTABLE_REF_VAL2:
+		reftable_free(ref->value.val2.target_value);
+		reftable_free(ref->value.val2.value);
+		break;
+	case REFTABLE_REF_VAL1:
+		reftable_free(ref->value.val1);
+		break;
+	case REFTABLE_REF_DELETION:
+		break;
+	default:
+		abort();
+	}
+
+	reftable_free(ref->refname);
+	memset(ref, 0, sizeof(struct reftable_ref_record));
+}
+
+static uint8_t reftable_ref_record_val_type(const void *rec)
+{
+	const struct reftable_ref_record *r =
+		(const struct reftable_ref_record *)rec;
+	return r->value_type;
+}
+
+static int reftable_ref_record_encode(const void *rec, struct string_view s,
+				      int hash_size)
+{
+	const struct reftable_ref_record *r =
+		(const struct reftable_ref_record *)rec;
+	struct string_view start = s;
+	int n = put_var_int(&s, r->update_index);
+	assert(hash_size > 0);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	switch (r->value_type) {
+	case REFTABLE_REF_SYMREF:
+		n = encode_string(r->value.symref, s);
+		if (n < 0) {
+			return -1;
+		}
+		string_view_consume(&s, n);
+		break;
+	case REFTABLE_REF_VAL2:
+		if (s.len < 2 * hash_size) {
+			return -1;
+		}
+		memcpy(s.buf, r->value.val2.value, hash_size);
+		string_view_consume(&s, hash_size);
+		memcpy(s.buf, r->value.val2.target_value, hash_size);
+		string_view_consume(&s, hash_size);
+		break;
+	case REFTABLE_REF_VAL1:
+		if (s.len < hash_size) {
+			return -1;
+		}
+		memcpy(s.buf, r->value.val1, hash_size);
+		string_view_consume(&s, hash_size);
+		break;
+	case REFTABLE_REF_DELETION:
+		break;
+	default:
+		abort();
+	}
+
+	return start.len - s.len;
+}
+
+static int reftable_ref_record_decode(void *rec, struct strbuf key,
+				      uint8_t val_type, struct string_view in,
+				      int hash_size)
+{
+	struct reftable_ref_record *r = rec;
+	struct string_view start = in;
+	uint64_t update_index = 0;
+	int n = get_var_int(&update_index, &in);
+	if (n < 0)
+		return n;
+	string_view_consume(&in, n);
+
+	reftable_ref_record_release(r);
+
+	assert(hash_size > 0);
+
+	r->refname = reftable_realloc(r->refname, key.len + 1);
+	memcpy(r->refname, key.buf, key.len);
+	r->update_index = update_index;
+	r->refname[key.len] = 0;
+	r->value_type = val_type;
+	switch (val_type) {
+	case REFTABLE_REF_VAL1:
+		if (in.len < hash_size) {
+			return -1;
+		}
+
+		r->value.val1 = reftable_malloc(hash_size);
+		memcpy(r->value.val1, in.buf, hash_size);
+		string_view_consume(&in, hash_size);
+		break;
+
+	case REFTABLE_REF_VAL2:
+		if (in.len < 2 * hash_size) {
+			return -1;
+		}
+
+		r->value.val2.value = reftable_malloc(hash_size);
+		memcpy(r->value.val2.value, in.buf, hash_size);
+		string_view_consume(&in, hash_size);
+
+		r->value.val2.target_value = reftable_malloc(hash_size);
+		memcpy(r->value.val2.target_value, in.buf, hash_size);
+		string_view_consume(&in, hash_size);
+		break;
+
+	case REFTABLE_REF_SYMREF: {
+		struct strbuf dest = STRBUF_INIT;
+		int n = decode_string(&dest, in);
+		if (n < 0) {
+			return -1;
+		}
+		string_view_consume(&in, n);
+		r->value.symref = dest.buf;
+	} break;
+
+	case REFTABLE_REF_DELETION:
+		break;
+	default:
+		abort();
+		break;
+	}
+
+	return start.len - in.len;
+}
+
+static int reftable_ref_record_is_deletion_void(const void *p)
+{
+	return reftable_ref_record_is_deletion(
+		(const struct reftable_ref_record *)p);
+}
+
+static struct reftable_record_vtable reftable_ref_record_vtable = {
+	.key = &reftable_ref_record_key,
+	.type = BLOCK_TYPE_REF,
+	.copy_from = &reftable_ref_record_copy_from,
+	.val_type = &reftable_ref_record_val_type,
+	.encode = &reftable_ref_record_encode,
+	.decode = &reftable_ref_record_decode,
+	.release = &reftable_ref_record_release_void,
+	.is_deletion = &reftable_ref_record_is_deletion_void,
+};
+
+static void reftable_obj_record_key(const void *r, struct strbuf *dest)
+{
+	const struct reftable_obj_record *rec =
+		(const struct reftable_obj_record *)r;
+	strbuf_reset(dest);
+	strbuf_add(dest, rec->hash_prefix, rec->hash_prefix_len);
+}
+
+static void reftable_obj_record_release(void *rec)
+{
+	struct reftable_obj_record *obj = rec;
+	FREE_AND_NULL(obj->hash_prefix);
+	FREE_AND_NULL(obj->offsets);
+	memset(obj, 0, sizeof(struct reftable_obj_record));
+}
+
+static void reftable_obj_record_copy_from(void *rec, const void *src_rec,
+					  int hash_size)
+{
+	struct reftable_obj_record *obj = rec;
+	const struct reftable_obj_record *src =
+		(const struct reftable_obj_record *)src_rec;
+
+	reftable_obj_record_release(obj);
+	*obj = *src;
+	obj->hash_prefix = reftable_malloc(obj->hash_prefix_len);
+	memcpy(obj->hash_prefix, src->hash_prefix, obj->hash_prefix_len);
+
+	obj->offsets = reftable_malloc(obj->offset_len * sizeof(uint64_t));
+	COPY_ARRAY(obj->offsets, src->offsets, obj->offset_len);
+}
+
+static uint8_t reftable_obj_record_val_type(const void *rec)
+{
+	const struct reftable_obj_record *r = rec;
+	if (r->offset_len > 0 && r->offset_len < 8)
+		return r->offset_len;
+	return 0;
+}
+
+static int reftable_obj_record_encode(const void *rec, struct string_view s,
+				      int hash_size)
+{
+	const struct reftable_obj_record *r = rec;
+	struct string_view start = s;
+	int i = 0;
+	int n = 0;
+	uint64_t last = 0;
+	if (r->offset_len == 0 || r->offset_len >= 8) {
+		n = put_var_int(&s, r->offset_len);
+		if (n < 0) {
+			return -1;
+		}
+		string_view_consume(&s, n);
+	}
+	if (r->offset_len == 0)
+		return start.len - s.len;
+	n = put_var_int(&s, r->offsets[0]);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	last = r->offsets[0];
+	for (i = 1; i < r->offset_len; i++) {
+		int n = put_var_int(&s, r->offsets[i] - last);
+		if (n < 0) {
+			return -1;
+		}
+		string_view_consume(&s, n);
+		last = r->offsets[i];
+	}
+	return start.len - s.len;
+}
+
+static int reftable_obj_record_decode(void *rec, struct strbuf key,
+				      uint8_t val_type, struct string_view in,
+				      int hash_size)
+{
+	struct string_view start = in;
+	struct reftable_obj_record *r = rec;
+	uint64_t count = val_type;
+	int n = 0;
+	uint64_t last;
+	int j;
+	r->hash_prefix = reftable_malloc(key.len);
+	memcpy(r->hash_prefix, key.buf, key.len);
+	r->hash_prefix_len = key.len;
+
+	if (val_type == 0) {
+		n = get_var_int(&count, &in);
+		if (n < 0) {
+			return n;
+		}
+
+		string_view_consume(&in, n);
+	}
+
+	r->offsets = NULL;
+	r->offset_len = 0;
+	if (count == 0)
+		return start.len - in.len;
+
+	r->offsets = reftable_malloc(count * sizeof(uint64_t));
+	r->offset_len = count;
+
+	n = get_var_int(&r->offsets[0], &in);
+	if (n < 0)
+		return n;
+	string_view_consume(&in, n);
+
+	last = r->offsets[0];
+	j = 1;
+	while (j < count) {
+		uint64_t delta = 0;
+		int n = get_var_int(&delta, &in);
+		if (n < 0) {
+			return n;
+		}
+		string_view_consume(&in, n);
+
+		last = r->offsets[j] = (delta + last);
+		j++;
+	}
+	return start.len - in.len;
+}
+
+static int not_a_deletion(const void *p)
+{
+	return 0;
+}
+
+static struct reftable_record_vtable reftable_obj_record_vtable = {
+	.key = &reftable_obj_record_key,
+	.type = BLOCK_TYPE_OBJ,
+	.copy_from = &reftable_obj_record_copy_from,
+	.val_type = &reftable_obj_record_val_type,
+	.encode = &reftable_obj_record_encode,
+	.decode = &reftable_obj_record_decode,
+	.release = &reftable_obj_record_release,
+	.is_deletion = not_a_deletion,
+};
+
+void reftable_log_record_print(struct reftable_log_record *log,
+			       uint32_t hash_id)
+{
+	char hex[GIT_SHA256_RAWSZ + 1] = { 0 };
+
+	switch (log->value_type) {
+	case REFTABLE_LOG_DELETION:
+		printf("log{%s(%" PRIu64 ") delete", log->refname,
+		       log->update_index);
+		break;
+	case REFTABLE_LOG_UPDATE:
+		printf("log{%s(%" PRIu64 ") %s <%s> %" PRIu64 " %04d\n",
+		       log->refname, log->update_index, log->value.update.name,
+		       log->value.update.email, log->value.update.time,
+		       log->value.update.tz_offset);
+		hex_format(hex, log->value.update.old_hash, hash_size(hash_id));
+		printf("%s => ", hex);
+		hex_format(hex, log->value.update.new_hash, hash_size(hash_id));
+		printf("%s\n\n%s\n}\n", hex, log->value.update.message);
+		break;
+	}
+}
+
+static void reftable_log_record_key(const void *r, struct strbuf *dest)
+{
+	const struct reftable_log_record *rec =
+		(const struct reftable_log_record *)r;
+	int len = strlen(rec->refname);
+	uint8_t i64[8];
+	uint64_t ts = 0;
+	strbuf_reset(dest);
+	strbuf_add(dest, (uint8_t *)rec->refname, len + 1);
+
+	ts = (~ts) - rec->update_index;
+	put_be64(&i64[0], ts);
+	strbuf_add(dest, i64, sizeof(i64));
+}
+
+static void reftable_log_record_copy_from(void *rec, const void *src_rec,
+					  int hash_size)
+{
+	struct reftable_log_record *dst = rec;
+	const struct reftable_log_record *src =
+		(const struct reftable_log_record *)src_rec;
+
+	reftable_log_record_release(dst);
+	*dst = *src;
+	if (dst->refname) {
+		dst->refname = xstrdup(dst->refname);
+	}
+	switch (dst->value_type) {
+	case REFTABLE_LOG_DELETION:
+		break;
+	case REFTABLE_LOG_UPDATE:
+		if (dst->value.update.email) {
+			dst->value.update.email =
+				xstrdup(dst->value.update.email);
+		}
+		if (dst->value.update.name) {
+			dst->value.update.name =
+				xstrdup(dst->value.update.name);
+		}
+		if (dst->value.update.message) {
+			dst->value.update.message =
+				xstrdup(dst->value.update.message);
+		}
+
+		if (dst->value.update.new_hash) {
+			dst->value.update.new_hash = reftable_malloc(hash_size);
+			memcpy(dst->value.update.new_hash,
+			       src->value.update.new_hash, hash_size);
+		}
+		if (dst->value.update.old_hash) {
+			dst->value.update.old_hash = reftable_malloc(hash_size);
+			memcpy(dst->value.update.old_hash,
+			       src->value.update.old_hash, hash_size);
+		}
+		break;
+	}
+}
+
+static void reftable_log_record_release_void(void *rec)
+{
+	struct reftable_log_record *r = rec;
+	reftable_log_record_release(r);
+}
+
+void reftable_log_record_release(struct reftable_log_record *r)
+{
+	reftable_free(r->refname);
+	switch (r->value_type) {
+	case REFTABLE_LOG_DELETION:
+		break;
+	case REFTABLE_LOG_UPDATE:
+		reftable_free(r->value.update.new_hash);
+		reftable_free(r->value.update.old_hash);
+		reftable_free(r->value.update.name);
+		reftable_free(r->value.update.email);
+		reftable_free(r->value.update.message);
+		break;
+	}
+	memset(r, 0, sizeof(struct reftable_log_record));
+}
+
+static uint8_t reftable_log_record_val_type(const void *rec)
+{
+	const struct reftable_log_record *log =
+		(const struct reftable_log_record *)rec;
+
+	return reftable_log_record_is_deletion(log) ? 0 : 1;
+}
+
+static uint8_t zero[GIT_SHA256_RAWSZ] = { 0 };
+
+static int reftable_log_record_encode(const void *rec, struct string_view s,
+				      int hash_size)
+{
+	const struct reftable_log_record *r = rec;
+	struct string_view start = s;
+	int n = 0;
+	uint8_t *oldh = NULL;
+	uint8_t *newh = NULL;
+	if (reftable_log_record_is_deletion(r))
+		return 0;
+
+	oldh = r->value.update.old_hash;
+	newh = r->value.update.new_hash;
+	if (!oldh) {
+		oldh = zero;
+	}
+	if (!newh) {
+		newh = zero;
+	}
+
+	if (s.len < 2 * hash_size)
+		return -1;
+
+	memcpy(s.buf, oldh, hash_size);
+	memcpy(s.buf + hash_size, newh, hash_size);
+	string_view_consume(&s, 2 * hash_size);
+
+	n = encode_string(r->value.update.name ? r->value.update.name : "", s);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	n = encode_string(r->value.update.email ? r->value.update.email : "",
+			  s);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	n = put_var_int(&s, r->value.update.time);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	if (s.len < 2)
+		return -1;
+
+	put_be16(s.buf, r->value.update.tz_offset);
+	string_view_consume(&s, 2);
+
+	n = encode_string(
+		r->value.update.message ? r->value.update.message : "", s);
+	if (n < 0)
+		return -1;
+	string_view_consume(&s, n);
+
+	return start.len - s.len;
+}
+
+static int reftable_log_record_decode(void *rec, struct strbuf key,
+				      uint8_t val_type, struct string_view in,
+				      int hash_size)
+{
+	struct string_view start = in;
+	struct reftable_log_record *r = rec;
+	uint64_t max = 0;
+	uint64_t ts = 0;
+	struct strbuf dest = STRBUF_INIT;
+	int n;
+
+	if (key.len <= 9 || key.buf[key.len - 9] != 0)
+		return REFTABLE_FORMAT_ERROR;
+
+	r->refname = reftable_realloc(r->refname, key.len - 8);
+	memcpy(r->refname, key.buf, key.len - 8);
+	ts = get_be64(key.buf + key.len - 8);
+
+	r->update_index = (~max) - ts;
+
+	if (val_type != r->value_type) {
+		switch (r->value_type) {
+		case REFTABLE_LOG_UPDATE:
+			FREE_AND_NULL(r->value.update.old_hash);
+			FREE_AND_NULL(r->value.update.new_hash);
+			FREE_AND_NULL(r->value.update.message);
+			FREE_AND_NULL(r->value.update.email);
+			FREE_AND_NULL(r->value.update.name);
+			break;
+		case REFTABLE_LOG_DELETION:
+			break;
+		}
+	}
+
+	r->value_type = val_type;
+	if (val_type == REFTABLE_LOG_DELETION)
+		return 0;
+
+	if (in.len < 2 * hash_size)
+		return REFTABLE_FORMAT_ERROR;
+
+	r->value.update.old_hash =
+		reftable_realloc(r->value.update.old_hash, hash_size);
+	r->value.update.new_hash =
+		reftable_realloc(r->value.update.new_hash, hash_size);
+
+	memcpy(r->value.update.old_hash, in.buf, hash_size);
+	memcpy(r->value.update.new_hash, in.buf + hash_size, hash_size);
+
+	string_view_consume(&in, 2 * hash_size);
+
+	n = decode_string(&dest, in);
+	if (n < 0)
+		goto done;
+	string_view_consume(&in, n);
+
+	r->value.update.name =
+		reftable_realloc(r->value.update.name, dest.len + 1);
+	memcpy(r->value.update.name, dest.buf, dest.len);
+	r->value.update.name[dest.len] = 0;
+
+	strbuf_reset(&dest);
+	n = decode_string(&dest, in);
+	if (n < 0)
+		goto done;
+	string_view_consume(&in, n);
+
+	r->value.update.email =
+		reftable_realloc(r->value.update.email, dest.len + 1);
+	memcpy(r->value.update.email, dest.buf, dest.len);
+	r->value.update.email[dest.len] = 0;
+
+	ts = 0;
+	n = get_var_int(&ts, &in);
+	if (n < 0)
+		goto done;
+	string_view_consume(&in, n);
+	r->value.update.time = ts;
+	if (in.len < 2)
+		goto done;
+
+	r->value.update.tz_offset = get_be16(in.buf);
+	string_view_consume(&in, 2);
+
+	strbuf_reset(&dest);
+	n = decode_string(&dest, in);
+	if (n < 0)
+		goto done;
+	string_view_consume(&in, n);
+
+	r->value.update.message =
+		reftable_realloc(r->value.update.message, dest.len + 1);
+	memcpy(r->value.update.message, dest.buf, dest.len);
+	r->value.update.message[dest.len] = 0;
+
+	strbuf_release(&dest);
+	return start.len - in.len;
+
+done:
+	strbuf_release(&dest);
+	return REFTABLE_FORMAT_ERROR;
+}
+
+static int null_streq(char *a, char *b)
+{
+	char *empty = "";
+	if (!a)
+		a = empty;
+
+	if (!b)
+		b = empty;
+
+	return 0 == strcmp(a, b);
+}
+
+static int zero_hash_eq(uint8_t *a, uint8_t *b, int sz)
+{
+	if (!a)
+		a = zero;
+
+	if (!b)
+		b = zero;
+
+	return !memcmp(a, b, sz);
+}
+
+int reftable_log_record_equal(struct reftable_log_record *a,
+			      struct reftable_log_record *b, int hash_size)
+{
+	if (!(null_streq(a->refname, b->refname) &&
+	      a->update_index == b->update_index &&
+	      a->value_type == b->value_type))
+		return 0;
+
+	switch (a->value_type) {
+	case REFTABLE_LOG_DELETION:
+		return 1;
+	case REFTABLE_LOG_UPDATE:
+		return null_streq(a->value.update.name, b->value.update.name) &&
+		       a->value.update.time == b->value.update.time &&
+		       a->value.update.tz_offset == b->value.update.tz_offset &&
+		       null_streq(a->value.update.email,
+				  b->value.update.email) &&
+		       null_streq(a->value.update.message,
+				  b->value.update.message) &&
+		       zero_hash_eq(a->value.update.old_hash,
+				    b->value.update.old_hash, hash_size) &&
+		       zero_hash_eq(a->value.update.new_hash,
+				    b->value.update.new_hash, hash_size);
+	}
+
+	abort();
+}
+
+static int reftable_log_record_is_deletion_void(const void *p)
+{
+	return reftable_log_record_is_deletion(
+		(const struct reftable_log_record *)p);
+}
+
+static struct reftable_record_vtable reftable_log_record_vtable = {
+	.key = &reftable_log_record_key,
+	.type = BLOCK_TYPE_LOG,
+	.copy_from = &reftable_log_record_copy_from,
+	.val_type = &reftable_log_record_val_type,
+	.encode = &reftable_log_record_encode,
+	.decode = &reftable_log_record_decode,
+	.release = &reftable_log_record_release_void,
+	.is_deletion = &reftable_log_record_is_deletion_void,
+};
+
+struct reftable_record reftable_new_record(uint8_t typ)
+{
+	struct reftable_record rec = { NULL };
+	switch (typ) {
+	case BLOCK_TYPE_REF: {
+		struct reftable_ref_record *r =
+			reftable_calloc(sizeof(struct reftable_ref_record));
+		reftable_record_from_ref(&rec, r);
+		return rec;
+	}
+
+	case BLOCK_TYPE_OBJ: {
+		struct reftable_obj_record *r =
+			reftable_calloc(sizeof(struct reftable_obj_record));
+		reftable_record_from_obj(&rec, r);
+		return rec;
+	}
+	case BLOCK_TYPE_LOG: {
+		struct reftable_log_record *r =
+			reftable_calloc(sizeof(struct reftable_log_record));
+		reftable_record_from_log(&rec, r);
+		return rec;
+	}
+	case BLOCK_TYPE_INDEX: {
+		struct reftable_index_record empty = { .last_key =
+							       STRBUF_INIT };
+		struct reftable_index_record *r =
+			reftable_calloc(sizeof(struct reftable_index_record));
+		*r = empty;
+		reftable_record_from_index(&rec, r);
+		return rec;
+	}
+	}
+	abort();
+	return rec;
+}
+
+/* clear out the record, yielding the reftable_record data that was
+ * encapsulated. */
+static void *reftable_record_yield(struct reftable_record *rec)
+{
+	void *p = rec->data;
+	rec->data = NULL;
+	return p;
+}
+
+void reftable_record_destroy(struct reftable_record *rec)
+{
+	reftable_record_release(rec);
+	reftable_free(reftable_record_yield(rec));
+}
+
+static void reftable_index_record_key(const void *r, struct strbuf *dest)
+{
+	const struct reftable_index_record *rec = r;
+	strbuf_reset(dest);
+	strbuf_addbuf(dest, &rec->last_key);
+}
+
+static void reftable_index_record_copy_from(void *rec, const void *src_rec,
+					    int hash_size)
+{
+	struct reftable_index_record *dst = rec;
+	const struct reftable_index_record *src = src_rec;
+
+	strbuf_reset(&dst->last_key);
+	strbuf_addbuf(&dst->last_key, &src->last_key);
+	dst->offset = src->offset;
+}
+
+static void reftable_index_record_release(void *rec)
+{
+	struct reftable_index_record *idx = rec;
+	strbuf_release(&idx->last_key);
+}
+
+static uint8_t reftable_index_record_val_type(const void *rec)
+{
+	return 0;
+}
+
+static int reftable_index_record_encode(const void *rec, struct string_view out,
+					int hash_size)
+{
+	const struct reftable_index_record *r =
+		(const struct reftable_index_record *)rec;
+	struct string_view start = out;
+
+	int n = put_var_int(&out, r->offset);
+	if (n < 0)
+		return n;
+
+	string_view_consume(&out, n);
+
+	return start.len - out.len;
+}
+
+static int reftable_index_record_decode(void *rec, struct strbuf key,
+					uint8_t val_type, struct string_view in,
+					int hash_size)
+{
+	struct string_view start = in;
+	struct reftable_index_record *r = rec;
+	int n = 0;
+
+	strbuf_reset(&r->last_key);
+	strbuf_addbuf(&r->last_key, &key);
+
+	n = get_var_int(&r->offset, &in);
+	if (n < 0)
+		return n;
+
+	string_view_consume(&in, n);
+	return start.len - in.len;
+}
+
+static struct reftable_record_vtable reftable_index_record_vtable = {
+	.key = &reftable_index_record_key,
+	.type = BLOCK_TYPE_INDEX,
+	.copy_from = &reftable_index_record_copy_from,
+	.val_type = &reftable_index_record_val_type,
+	.encode = &reftable_index_record_encode,
+	.decode = &reftable_index_record_decode,
+	.release = &reftable_index_record_release,
+	.is_deletion = &not_a_deletion,
+};
+
+void reftable_record_key(struct reftable_record *rec, struct strbuf *dest)
+{
+	rec->ops->key(rec->data, dest);
+}
+
+uint8_t reftable_record_type(struct reftable_record *rec)
+{
+	return rec->ops->type;
+}
+
+int reftable_record_encode(struct reftable_record *rec, struct string_view dest,
+			   int hash_size)
+{
+	return rec->ops->encode(rec->data, dest, hash_size);
+}
+
+void reftable_record_copy_from(struct reftable_record *rec,
+			       struct reftable_record *src, int hash_size)
+{
+	assert(src->ops->type == rec->ops->type);
+
+	rec->ops->copy_from(rec->data, src->data, hash_size);
+}
+
+uint8_t reftable_record_val_type(struct reftable_record *rec)
+{
+	return rec->ops->val_type(rec->data);
+}
+
+int reftable_record_decode(struct reftable_record *rec, struct strbuf key,
+			   uint8_t extra, struct string_view src, int hash_size)
+{
+	return rec->ops->decode(rec->data, key, extra, src, hash_size);
+}
+
+void reftable_record_release(struct reftable_record *rec)
+{
+	rec->ops->release(rec->data);
+}
+
+int reftable_record_is_deletion(struct reftable_record *rec)
+{
+	return rec->ops->is_deletion(rec->data);
+}
+
+void reftable_record_from_ref(struct reftable_record *rec,
+			      struct reftable_ref_record *ref_rec)
+{
+	assert(!rec->ops);
+	rec->data = ref_rec;
+	rec->ops = &reftable_ref_record_vtable;
+}
+
+void reftable_record_from_obj(struct reftable_record *rec,
+			      struct reftable_obj_record *obj_rec)
+{
+	assert(!rec->ops);
+	rec->data = obj_rec;
+	rec->ops = &reftable_obj_record_vtable;
+}
+
+void reftable_record_from_index(struct reftable_record *rec,
+				struct reftable_index_record *index_rec)
+{
+	assert(!rec->ops);
+	rec->data = index_rec;
+	rec->ops = &reftable_index_record_vtable;
+}
+
+void reftable_record_from_log(struct reftable_record *rec,
+			      struct reftable_log_record *log_rec)
+{
+	assert(!rec->ops);
+	rec->data = log_rec;
+	rec->ops = &reftable_log_record_vtable;
+}
+
+struct reftable_ref_record *reftable_record_as_ref(struct reftable_record *rec)
+{
+	assert(reftable_record_type(rec) == BLOCK_TYPE_REF);
+	return rec->data;
+}
+
+struct reftable_log_record *reftable_record_as_log(struct reftable_record *rec)
+{
+	assert(reftable_record_type(rec) == BLOCK_TYPE_LOG);
+	return rec->data;
+}
+
+static int hash_equal(uint8_t *a, uint8_t *b, int hash_size)
+{
+	if (a && b)
+		return !memcmp(a, b, hash_size);
+
+	return a == b;
+}
+
+int reftable_ref_record_equal(struct reftable_ref_record *a,
+			      struct reftable_ref_record *b, int hash_size)
+{
+	assert(hash_size > 0);
+	if (!(0 == strcmp(a->refname, b->refname) &&
+	      a->update_index == b->update_index &&
+	      a->value_type == b->value_type))
+		return 0;
+
+	switch (a->value_type) {
+	case REFTABLE_REF_SYMREF:
+		return !strcmp(a->value.symref, b->value.symref);
+	case REFTABLE_REF_VAL2:
+		return hash_equal(a->value.val2.value, b->value.val2.value,
+				  hash_size) &&
+		       hash_equal(a->value.val2.target_value,
+				  b->value.val2.target_value, hash_size);
+	case REFTABLE_REF_VAL1:
+		return hash_equal(a->value.val1, b->value.val1, hash_size);
+	case REFTABLE_REF_DELETION:
+		return 1;
+	default:
+		abort();
+	}
+}
+
+int reftable_ref_record_compare_name(const void *a, const void *b)
+{
+	return strcmp(((struct reftable_ref_record *)a)->refname,
+		      ((struct reftable_ref_record *)b)->refname);
+}
+
+int reftable_ref_record_is_deletion(const struct reftable_ref_record *ref)
+{
+	return ref->value_type == REFTABLE_REF_DELETION;
+}
+
+int reftable_log_record_compare_key(const void *a, const void *b)
+{
+	const struct reftable_log_record *la = a;
+	const struct reftable_log_record *lb = b;
+
+	int cmp = strcmp(la->refname, lb->refname);
+	if (cmp)
+		return cmp;
+	if (la->update_index > lb->update_index)
+		return -1;
+	return (la->update_index < lb->update_index) ? 1 : 0;
+}
+
+int reftable_log_record_is_deletion(const struct reftable_log_record *log)
+{
+	return (log->value_type == REFTABLE_LOG_DELETION);
+}
+
+void string_view_consume(struct string_view *s, int n)
+{
+	s->buf += n;
+	s->len -= n;
+}
diff --git a/reftable/record.h b/reftable/record.h
new file mode 100644
index 0000000..498e8c5
--- /dev/null
+++ b/reftable/record.h
@@ -0,0 +1,139 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef RECORD_H
+#define RECORD_H
+
+#include "system.h"
+
+#include <stdint.h>
+
+#include "reftable-record.h"
+
+/*
+ * A substring of existing string data. This structure takes no responsibility
+ * for the lifetime of the data it points to.
+ */
+struct string_view {
+	uint8_t *buf;
+	size_t len;
+};
+
+/* Advance `s.buf` by `n`, and decrease length. */
+void string_view_consume(struct string_view *s, int n);
+
+/* utilities for de/encoding varints */
+
+int get_var_int(uint64_t *dest, struct string_view *in);
+int put_var_int(struct string_view *dest, uint64_t val);
+
+/* Methods for records. */
+struct reftable_record_vtable {
+	/* encode the key of to a uint8_t strbuf. */
+	void (*key)(const void *rec, struct strbuf *dest);
+
+	/* The record type of ('r' for ref). */
+	uint8_t type;
+
+	void (*copy_from)(void *dest, const void *src, int hash_size);
+
+	/* a value of [0..7], indicating record subvariants (eg. ref vs. symref
+	 * vs ref deletion) */
+	uint8_t (*val_type)(const void *rec);
+
+	/* encodes rec into dest, returning how much space was used. */
+	int (*encode)(const void *rec, struct string_view dest, int hash_size);
+
+	/* decode data from `src` into the record. */
+	int (*decode)(void *rec, struct strbuf key, uint8_t extra,
+		      struct string_view src, int hash_size);
+
+	/* deallocate and null the record. */
+	void (*release)(void *rec);
+
+	/* is this a tombstone? */
+	int (*is_deletion)(const void *rec);
+};
+
+/* record is a generic wrapper for different types of records. */
+struct reftable_record {
+	void *data;
+	struct reftable_record_vtable *ops;
+};
+
+/* returns true for recognized block types. Block start with the block type. */
+int reftable_is_block_type(uint8_t typ);
+
+/* creates a malloced record of the given type. Dispose with record_destroy */
+struct reftable_record reftable_new_record(uint8_t typ);
+
+/* Encode `key` into `dest`. Sets `is_restart` to indicate a restart. Returns
+ * number of bytes written. */
+int reftable_encode_key(int *is_restart, struct string_view dest,
+			struct strbuf prev_key, struct strbuf key,
+			uint8_t extra);
+
+/* Decode into `key` and `extra` from `in` */
+int reftable_decode_key(struct strbuf *key, uint8_t *extra,
+			struct strbuf last_key, struct string_view in);
+
+/* reftable_index_record are used internally to speed up lookups. */
+struct reftable_index_record {
+	uint64_t offset; /* Offset of block */
+	struct strbuf last_key; /* Last key of the block. */
+};
+
+/* reftable_obj_record stores an object ID => ref mapping. */
+struct reftable_obj_record {
+	uint8_t *hash_prefix; /* leading bytes of the object ID */
+	int hash_prefix_len; /* number of leading bytes. Constant
+			      * across a single table. */
+	uint64_t *offsets; /* a vector of file offsets. */
+	int offset_len;
+};
+
+/* see struct record_vtable */
+
+void reftable_record_key(struct reftable_record *rec, struct strbuf *dest);
+uint8_t reftable_record_type(struct reftable_record *rec);
+void reftable_record_copy_from(struct reftable_record *rec,
+			       struct reftable_record *src, int hash_size);
+uint8_t reftable_record_val_type(struct reftable_record *rec);
+int reftable_record_encode(struct reftable_record *rec, struct string_view dest,
+			   int hash_size);
+int reftable_record_decode(struct reftable_record *rec, struct strbuf key,
+			   uint8_t extra, struct string_view src,
+			   int hash_size);
+int reftable_record_is_deletion(struct reftable_record *rec);
+
+/* zeroes out the embedded record */
+void reftable_record_release(struct reftable_record *rec);
+
+/* clear and deallocate embedded record, and zero `rec`. */
+void reftable_record_destroy(struct reftable_record *rec);
+
+/* initialize generic records from concrete records. The generic record should
+ * be zeroed out. */
+void reftable_record_from_obj(struct reftable_record *rec,
+			      struct reftable_obj_record *objrec);
+void reftable_record_from_index(struct reftable_record *rec,
+				struct reftable_index_record *idxrec);
+void reftable_record_from_ref(struct reftable_record *rec,
+			      struct reftable_ref_record *refrec);
+void reftable_record_from_log(struct reftable_record *rec,
+			      struct reftable_log_record *logrec);
+struct reftable_ref_record *reftable_record_as_ref(struct reftable_record *ref);
+struct reftable_log_record *reftable_record_as_log(struct reftable_record *ref);
+
+/* for qsort. */
+int reftable_ref_record_compare_name(const void *a, const void *b);
+
+/* for qsort. */
+int reftable_log_record_compare_key(const void *a, const void *b);
+
+#endif
diff --git a/reftable/record_test.c b/reftable/record_test.c
new file mode 100644
index 0000000..f4ad7ca
--- /dev/null
+++ b/reftable/record_test.c
@@ -0,0 +1,412 @@
+/*
+  Copyright 2020 Google LLC
+
+  Use of this source code is governed by a BSD-style
+  license that can be found in the LICENSE file or at
+  https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "record.h"
+
+#include "system.h"
+#include "basics.h"
+#include "constants.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+static void test_copy(struct reftable_record *rec)
+{
+	struct reftable_record copy =
+		reftable_new_record(reftable_record_type(rec));
+	reftable_record_copy_from(&copy, rec, GIT_SHA1_RAWSZ);
+	/* do it twice to catch memory leaks */
+	reftable_record_copy_from(&copy, rec, GIT_SHA1_RAWSZ);
+	switch (reftable_record_type(&copy)) {
+	case BLOCK_TYPE_REF:
+		EXPECT(reftable_ref_record_equal(reftable_record_as_ref(&copy),
+						 reftable_record_as_ref(rec),
+						 GIT_SHA1_RAWSZ));
+		break;
+	case BLOCK_TYPE_LOG:
+		EXPECT(reftable_log_record_equal(reftable_record_as_log(&copy),
+						 reftable_record_as_log(rec),
+						 GIT_SHA1_RAWSZ));
+		break;
+	}
+	reftable_record_destroy(&copy);
+}
+
+static void test_varint_roundtrip(void)
+{
+	uint64_t inputs[] = { 0,
+			      1,
+			      27,
+			      127,
+			      128,
+			      257,
+			      4096,
+			      ((uint64_t)1 << 63),
+			      ((uint64_t)1 << 63) + ((uint64_t)1 << 63) - 1 };
+	int i = 0;
+	for (i = 0; i < ARRAY_SIZE(inputs); i++) {
+		uint8_t dest[10];
+
+		struct string_view out = {
+			.buf = dest,
+			.len = sizeof(dest),
+		};
+		uint64_t in = inputs[i];
+		int n = put_var_int(&out, in);
+		uint64_t got = 0;
+
+		EXPECT(n > 0);
+		out.len = n;
+		n = get_var_int(&got, &out);
+		EXPECT(n > 0);
+
+		EXPECT(got == in);
+	}
+}
+
+static void test_common_prefix(void)
+{
+	struct {
+		const char *a, *b;
+		int want;
+	} cases[] = {
+		{ "abc", "ab", 2 },
+		{ "", "abc", 0 },
+		{ "abc", "abd", 2 },
+		{ "abc", "pqr", 0 },
+	};
+
+	int i = 0;
+	for (i = 0; i < ARRAY_SIZE(cases); i++) {
+		struct strbuf a = STRBUF_INIT;
+		struct strbuf b = STRBUF_INIT;
+		strbuf_addstr(&a, cases[i].a);
+		strbuf_addstr(&b, cases[i].b);
+		EXPECT(common_prefix_size(&a, &b) == cases[i].want);
+
+		strbuf_release(&a);
+		strbuf_release(&b);
+	}
+}
+
+static void set_hash(uint8_t *h, int j)
+{
+	int i = 0;
+	for (i = 0; i < hash_size(GIT_SHA1_FORMAT_ID); i++) {
+		h[i] = (j >> i) & 0xff;
+	}
+}
+
+static void test_reftable_ref_record_roundtrip(void)
+{
+	int i = 0;
+
+	for (i = REFTABLE_REF_DELETION; i < REFTABLE_NR_REF_VALUETYPES; i++) {
+		struct reftable_ref_record in = { NULL };
+		struct reftable_ref_record out = { NULL };
+		struct reftable_record rec_out = { NULL };
+		struct strbuf key = STRBUF_INIT;
+		struct reftable_record rec = { NULL };
+		uint8_t buffer[1024] = { 0 };
+		struct string_view dest = {
+			.buf = buffer,
+			.len = sizeof(buffer),
+		};
+
+		int n, m;
+
+		in.value_type = i;
+		switch (i) {
+		case REFTABLE_REF_DELETION:
+			break;
+		case REFTABLE_REF_VAL1:
+			in.value.val1 = reftable_malloc(GIT_SHA1_RAWSZ);
+			set_hash(in.value.val1, 1);
+			break;
+		case REFTABLE_REF_VAL2:
+			in.value.val2.value = reftable_malloc(GIT_SHA1_RAWSZ);
+			set_hash(in.value.val2.value, 1);
+			in.value.val2.target_value =
+				reftable_malloc(GIT_SHA1_RAWSZ);
+			set_hash(in.value.val2.target_value, 2);
+			break;
+		case REFTABLE_REF_SYMREF:
+			in.value.symref = xstrdup("target");
+			break;
+		}
+		in.refname = xstrdup("refs/heads/master");
+
+		reftable_record_from_ref(&rec, &in);
+		test_copy(&rec);
+
+		EXPECT(reftable_record_val_type(&rec) == i);
+
+		reftable_record_key(&rec, &key);
+		n = reftable_record_encode(&rec, dest, GIT_SHA1_RAWSZ);
+		EXPECT(n > 0);
+
+		/* decode into a non-zero reftable_record to test for leaks. */
+
+		reftable_record_from_ref(&rec_out, &out);
+		m = reftable_record_decode(&rec_out, key, i, dest,
+					   GIT_SHA1_RAWSZ);
+		EXPECT(n == m);
+
+		EXPECT(reftable_ref_record_equal(&in, &out, GIT_SHA1_RAWSZ));
+		reftable_record_release(&rec_out);
+
+		strbuf_release(&key);
+		reftable_ref_record_release(&in);
+	}
+}
+
+static void test_reftable_log_record_equal(void)
+{
+	struct reftable_log_record in[2] = {
+		{
+			.refname = xstrdup("refs/heads/master"),
+			.update_index = 42,
+		},
+		{
+			.refname = xstrdup("refs/heads/master"),
+			.update_index = 22,
+		}
+	};
+
+	EXPECT(!reftable_log_record_equal(&in[0], &in[1], GIT_SHA1_RAWSZ));
+	in[1].update_index = in[0].update_index;
+	EXPECT(reftable_log_record_equal(&in[0], &in[1], GIT_SHA1_RAWSZ));
+	reftable_log_record_release(&in[0]);
+	reftable_log_record_release(&in[1]);
+}
+
+static void test_reftable_log_record_roundtrip(void)
+{
+	int i;
+	struct reftable_log_record in[2] = {
+		{
+			.refname = xstrdup("refs/heads/master"),
+			.update_index = 42,
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value = {
+				.update = {
+					.old_hash = reftable_malloc(GIT_SHA1_RAWSZ),
+					.new_hash = reftable_malloc(GIT_SHA1_RAWSZ),
+					.name = xstrdup("han-wen"),
+					.email = xstrdup("hanwen@google.com"),
+					.message = xstrdup("test"),
+					.time = 1577123507,
+					.tz_offset = 100,
+				},
+			}
+		},
+		{
+			.refname = xstrdup("refs/heads/master"),
+			.update_index = 22,
+			.value_type = REFTABLE_LOG_DELETION,
+		}
+	};
+	set_test_hash(in[0].value.update.new_hash, 1);
+	set_test_hash(in[0].value.update.old_hash, 2);
+	for (i = 0; i < ARRAY_SIZE(in); i++) {
+		struct reftable_record rec = { NULL };
+		struct strbuf key = STRBUF_INIT;
+		uint8_t buffer[1024] = { 0 };
+		struct string_view dest = {
+			.buf = buffer,
+			.len = sizeof(buffer),
+		};
+		/* populate out, to check for leaks. */
+		struct reftable_log_record out = {
+			.refname = xstrdup("old name"),
+			.value_type = REFTABLE_LOG_UPDATE,
+			.value = {
+				.update = {
+					.new_hash = reftable_calloc(GIT_SHA1_RAWSZ),
+					.old_hash = reftable_calloc(GIT_SHA1_RAWSZ),
+					.name = xstrdup("old name"),
+					.email = xstrdup("old@email"),
+					.message = xstrdup("old message"),
+				},
+			},
+		};
+		struct reftable_record rec_out = { NULL };
+		int n, m, valtype;
+
+		reftable_record_from_log(&rec, &in[i]);
+
+		test_copy(&rec);
+
+		reftable_record_key(&rec, &key);
+
+		n = reftable_record_encode(&rec, dest, GIT_SHA1_RAWSZ);
+		EXPECT(n >= 0);
+		reftable_record_from_log(&rec_out, &out);
+		valtype = reftable_record_val_type(&rec);
+		m = reftable_record_decode(&rec_out, key, valtype, dest,
+					   GIT_SHA1_RAWSZ);
+		EXPECT(n == m);
+
+		EXPECT(reftable_log_record_equal(&in[i], &out, GIT_SHA1_RAWSZ));
+		reftable_log_record_release(&in[i]);
+		strbuf_release(&key);
+		reftable_record_release(&rec_out);
+	}
+}
+
+static void test_u24_roundtrip(void)
+{
+	uint32_t in = 0x112233;
+	uint8_t dest[3];
+	uint32_t out;
+	put_be24(dest, in);
+	out = get_be24(dest);
+	EXPECT(in == out);
+}
+
+static void test_key_roundtrip(void)
+{
+	uint8_t buffer[1024] = { 0 };
+	struct string_view dest = {
+		.buf = buffer,
+		.len = sizeof(buffer),
+	};
+	struct strbuf last_key = STRBUF_INIT;
+	struct strbuf key = STRBUF_INIT;
+	struct strbuf roundtrip = STRBUF_INIT;
+	int restart;
+	uint8_t extra;
+	int n, m;
+	uint8_t rt_extra;
+
+	strbuf_addstr(&last_key, "refs/heads/master");
+	strbuf_addstr(&key, "refs/tags/bla");
+	extra = 6;
+	n = reftable_encode_key(&restart, dest, last_key, key, extra);
+	EXPECT(!restart);
+	EXPECT(n > 0);
+
+	m = reftable_decode_key(&roundtrip, &rt_extra, last_key, dest);
+	EXPECT(n == m);
+	EXPECT(0 == strbuf_cmp(&key, &roundtrip));
+	EXPECT(rt_extra == extra);
+
+	strbuf_release(&last_key);
+	strbuf_release(&key);
+	strbuf_release(&roundtrip);
+}
+
+static void test_reftable_obj_record_roundtrip(void)
+{
+	uint8_t testHash1[GIT_SHA1_RAWSZ] = { 1, 2, 3, 4, 0 };
+	uint64_t till9[] = { 1, 2, 3, 4, 500, 600, 700, 800, 9000 };
+	struct reftable_obj_record recs[3] = { {
+						       .hash_prefix = testHash1,
+						       .hash_prefix_len = 5,
+						       .offsets = till9,
+						       .offset_len = 3,
+					       },
+					       {
+						       .hash_prefix = testHash1,
+						       .hash_prefix_len = 5,
+						       .offsets = till9,
+						       .offset_len = 9,
+					       },
+					       {
+						       .hash_prefix = testHash1,
+						       .hash_prefix_len = 5,
+					       } };
+	int i = 0;
+	for (i = 0; i < ARRAY_SIZE(recs); i++) {
+		struct reftable_obj_record in = recs[i];
+		uint8_t buffer[1024] = { 0 };
+		struct string_view dest = {
+			.buf = buffer,
+			.len = sizeof(buffer),
+		};
+		struct reftable_record rec = { NULL };
+		struct strbuf key = STRBUF_INIT;
+		struct reftable_obj_record out = { NULL };
+		struct reftable_record rec_out = { NULL };
+		int n, m;
+		uint8_t extra;
+
+		reftable_record_from_obj(&rec, &in);
+		test_copy(&rec);
+		reftable_record_key(&rec, &key);
+		n = reftable_record_encode(&rec, dest, GIT_SHA1_RAWSZ);
+		EXPECT(n > 0);
+		extra = reftable_record_val_type(&rec);
+		reftable_record_from_obj(&rec_out, &out);
+		m = reftable_record_decode(&rec_out, key, extra, dest,
+					   GIT_SHA1_RAWSZ);
+		EXPECT(n == m);
+
+		EXPECT(in.hash_prefix_len == out.hash_prefix_len);
+		EXPECT(in.offset_len == out.offset_len);
+
+		EXPECT(!memcmp(in.hash_prefix, out.hash_prefix,
+			       in.hash_prefix_len));
+		EXPECT(0 == memcmp(in.offsets, out.offsets,
+				   sizeof(uint64_t) * in.offset_len));
+		strbuf_release(&key);
+		reftable_record_release(&rec_out);
+	}
+}
+
+static void test_reftable_index_record_roundtrip(void)
+{
+	struct reftable_index_record in = {
+		.offset = 42,
+		.last_key = STRBUF_INIT,
+	};
+	uint8_t buffer[1024] = { 0 };
+	struct string_view dest = {
+		.buf = buffer,
+		.len = sizeof(buffer),
+	};
+	struct strbuf key = STRBUF_INIT;
+	struct reftable_record rec = { NULL };
+	struct reftable_index_record out = { .last_key = STRBUF_INIT };
+	struct reftable_record out_rec = { NULL };
+	int n, m;
+	uint8_t extra;
+
+	strbuf_addstr(&in.last_key, "refs/heads/master");
+	reftable_record_from_index(&rec, &in);
+	reftable_record_key(&rec, &key);
+	test_copy(&rec);
+
+	EXPECT(0 == strbuf_cmp(&key, &in.last_key));
+	n = reftable_record_encode(&rec, dest, GIT_SHA1_RAWSZ);
+	EXPECT(n > 0);
+
+	extra = reftable_record_val_type(&rec);
+	reftable_record_from_index(&out_rec, &out);
+	m = reftable_record_decode(&out_rec, key, extra, dest, GIT_SHA1_RAWSZ);
+	EXPECT(m == n);
+
+	EXPECT(in.offset == out.offset);
+
+	reftable_record_release(&out_rec);
+	strbuf_release(&key);
+	strbuf_release(&in.last_key);
+}
+
+int record_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_reftable_log_record_equal);
+	RUN_TEST(test_reftable_log_record_roundtrip);
+	RUN_TEST(test_reftable_ref_record_roundtrip);
+	RUN_TEST(test_varint_roundtrip);
+	RUN_TEST(test_key_roundtrip);
+	RUN_TEST(test_common_prefix);
+	RUN_TEST(test_reftable_obj_record_roundtrip);
+	RUN_TEST(test_reftable_index_record_roundtrip);
+	RUN_TEST(test_u24_roundtrip);
+	return 0;
+}
diff --git a/reftable/refname.c b/reftable/refname.c
new file mode 100644
index 0000000..9573496
--- /dev/null
+++ b/reftable/refname.c
@@ -0,0 +1,209 @@
+/*
+  Copyright 2020 Google LLC
+
+  Use of this source code is governed by a BSD-style
+  license that can be found in the LICENSE file or at
+  https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+#include "reftable-error.h"
+#include "basics.h"
+#include "refname.h"
+#include "reftable-iterator.h"
+
+struct find_arg {
+	char **names;
+	const char *want;
+};
+
+static int find_name(size_t k, void *arg)
+{
+	struct find_arg *f_arg = arg;
+	return strcmp(f_arg->names[k], f_arg->want) >= 0;
+}
+
+static int modification_has_ref(struct modification *mod, const char *name)
+{
+	struct reftable_ref_record ref = { NULL };
+	int err = 0;
+
+	if (mod->add_len > 0) {
+		struct find_arg arg = {
+			.names = mod->add,
+			.want = name,
+		};
+		int idx = binsearch(mod->add_len, find_name, &arg);
+		if (idx < mod->add_len && !strcmp(mod->add[idx], name)) {
+			return 0;
+		}
+	}
+
+	if (mod->del_len > 0) {
+		struct find_arg arg = {
+			.names = mod->del,
+			.want = name,
+		};
+		int idx = binsearch(mod->del_len, find_name, &arg);
+		if (idx < mod->del_len && !strcmp(mod->del[idx], name)) {
+			return 1;
+		}
+	}
+
+	err = reftable_table_read_ref(&mod->tab, name, &ref);
+	reftable_ref_record_release(&ref);
+	return err;
+}
+
+static void modification_release(struct modification *mod)
+{
+	/* don't delete the strings themselves; they're owned by ref records.
+	 */
+	FREE_AND_NULL(mod->add);
+	FREE_AND_NULL(mod->del);
+	mod->add_len = 0;
+	mod->del_len = 0;
+}
+
+static int modification_has_ref_with_prefix(struct modification *mod,
+					    const char *prefix)
+{
+	struct reftable_iterator it = { NULL };
+	struct reftable_ref_record ref = { NULL };
+	int err = 0;
+
+	if (mod->add_len > 0) {
+		struct find_arg arg = {
+			.names = mod->add,
+			.want = prefix,
+		};
+		int idx = binsearch(mod->add_len, find_name, &arg);
+		if (idx < mod->add_len &&
+		    !strncmp(prefix, mod->add[idx], strlen(prefix)))
+			goto done;
+	}
+	err = reftable_table_seek_ref(&mod->tab, &it, prefix);
+	if (err)
+		goto done;
+
+	while (1) {
+		err = reftable_iterator_next_ref(&it, &ref);
+		if (err)
+			goto done;
+
+		if (mod->del_len > 0) {
+			struct find_arg arg = {
+				.names = mod->del,
+				.want = ref.refname,
+			};
+			int idx = binsearch(mod->del_len, find_name, &arg);
+			if (idx < mod->del_len &&
+			    !strcmp(ref.refname, mod->del[idx])) {
+				continue;
+			}
+		}
+
+		if (strncmp(ref.refname, prefix, strlen(prefix))) {
+			err = 1;
+			goto done;
+		}
+		err = 0;
+		goto done;
+	}
+
+done:
+	reftable_ref_record_release(&ref);
+	reftable_iterator_destroy(&it);
+	return err;
+}
+
+static int validate_refname(const char *name)
+{
+	while (1) {
+		char *next = strchr(name, '/');
+		if (!*name) {
+			return REFTABLE_REFNAME_ERROR;
+		}
+		if (!next) {
+			return 0;
+		}
+		if (next - name == 0 || (next - name == 1 && *name == '.') ||
+		    (next - name == 2 && name[0] == '.' && name[1] == '.'))
+			return REFTABLE_REFNAME_ERROR;
+		name = next + 1;
+	}
+	return 0;
+}
+
+int validate_ref_record_addition(struct reftable_table tab,
+				 struct reftable_ref_record *recs, size_t sz)
+{
+	struct modification mod = {
+		.tab = tab,
+		.add = reftable_calloc(sizeof(char *) * sz),
+		.del = reftable_calloc(sizeof(char *) * sz),
+	};
+	int i = 0;
+	int err = 0;
+	for (; i < sz; i++) {
+		if (reftable_ref_record_is_deletion(&recs[i])) {
+			mod.del[mod.del_len++] = recs[i].refname;
+		} else {
+			mod.add[mod.add_len++] = recs[i].refname;
+		}
+	}
+
+	err = modification_validate(&mod);
+	modification_release(&mod);
+	return err;
+}
+
+static void strbuf_trim_component(struct strbuf *sl)
+{
+	while (sl->len > 0) {
+		int is_slash = (sl->buf[sl->len - 1] == '/');
+		strbuf_setlen(sl, sl->len - 1);
+		if (is_slash)
+			break;
+	}
+}
+
+int modification_validate(struct modification *mod)
+{
+	struct strbuf slashed = STRBUF_INIT;
+	int err = 0;
+	int i = 0;
+	for (; i < mod->add_len; i++) {
+		err = validate_refname(mod->add[i]);
+		if (err)
+			goto done;
+		strbuf_reset(&slashed);
+		strbuf_addstr(&slashed, mod->add[i]);
+		strbuf_addstr(&slashed, "/");
+
+		err = modification_has_ref_with_prefix(mod, slashed.buf);
+		if (err == 0) {
+			err = REFTABLE_NAME_CONFLICT;
+			goto done;
+		}
+		if (err < 0)
+			goto done;
+
+		strbuf_reset(&slashed);
+		strbuf_addstr(&slashed, mod->add[i]);
+		while (slashed.len) {
+			strbuf_trim_component(&slashed);
+			err = modification_has_ref(mod, slashed.buf);
+			if (err == 0) {
+				err = REFTABLE_NAME_CONFLICT;
+				goto done;
+			}
+			if (err < 0)
+				goto done;
+		}
+	}
+	err = 0;
+done:
+	strbuf_release(&slashed);
+	return err;
+}
diff --git a/reftable/refname.h b/reftable/refname.h
new file mode 100644
index 0000000..a24b40f
--- /dev/null
+++ b/reftable/refname.h
@@ -0,0 +1,29 @@
+/*
+  Copyright 2020 Google LLC
+
+  Use of this source code is governed by a BSD-style
+  license that can be found in the LICENSE file or at
+  https://developers.google.com/open-source/licenses/bsd
+*/
+#ifndef REFNAME_H
+#define REFNAME_H
+
+#include "reftable-record.h"
+#include "reftable-generic.h"
+
+struct modification {
+	struct reftable_table tab;
+
+	char **add;
+	size_t add_len;
+
+	char **del;
+	size_t del_len;
+};
+
+int validate_ref_record_addition(struct reftable_table tab,
+				 struct reftable_ref_record *recs, size_t sz);
+
+int modification_validate(struct modification *mod);
+
+#endif
diff --git a/reftable/refname_test.c b/reftable/refname_test.c
new file mode 100644
index 0000000..8645cd9
--- /dev/null
+++ b/reftable/refname_test.c
@@ -0,0 +1,102 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "basics.h"
+#include "block.h"
+#include "blocksource.h"
+#include "constants.h"
+#include "reader.h"
+#include "record.h"
+#include "refname.h"
+#include "reftable-error.h"
+#include "reftable-writer.h"
+#include "system.h"
+
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+struct testcase {
+	char *add;
+	char *del;
+	int error_code;
+};
+
+static void test_conflict(void)
+{
+	struct reftable_write_options opts = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	struct reftable_writer *w =
+		reftable_new_writer(&strbuf_add_void, &buf, &opts);
+	struct reftable_ref_record rec = {
+		.refname = "a/b",
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "destination", /* make sure it's not a symref.
+						*/
+		.update_index = 1,
+	};
+	int err;
+	int i;
+	struct reftable_block_source source = { NULL };
+	struct reftable_reader *rd = NULL;
+	struct reftable_table tab = { NULL };
+	struct testcase cases[] = {
+		{ "a/b/c", NULL, REFTABLE_NAME_CONFLICT },
+		{ "b", NULL, 0 },
+		{ "a", NULL, REFTABLE_NAME_CONFLICT },
+		{ "a", "a/b", 0 },
+
+		{ "p/", NULL, REFTABLE_REFNAME_ERROR },
+		{ "p//q", NULL, REFTABLE_REFNAME_ERROR },
+		{ "p/./q", NULL, REFTABLE_REFNAME_ERROR },
+		{ "p/../q", NULL, REFTABLE_REFNAME_ERROR },
+
+		{ "a/b/c", "a/b", 0 },
+		{ NULL, "a//b", 0 },
+	};
+	reftable_writer_set_limits(w, 1, 1);
+
+	err = reftable_writer_add_ref(w, &rec);
+	EXPECT_ERR(err);
+
+	err = reftable_writer_close(w);
+	EXPECT_ERR(err);
+	reftable_writer_free(w);
+
+	block_source_from_strbuf(&source, &buf);
+	err = reftable_new_reader(&rd, &source, "filename");
+	EXPECT_ERR(err);
+
+	reftable_table_from_reader(&tab, rd);
+
+	for (i = 0; i < ARRAY_SIZE(cases); i++) {
+		struct modification mod = {
+			.tab = tab,
+		};
+
+		if (cases[i].add) {
+			mod.add = &cases[i].add;
+			mod.add_len = 1;
+		}
+		if (cases[i].del) {
+			mod.del = &cases[i].del;
+			mod.del_len = 1;
+		}
+
+		err = modification_validate(&mod);
+		EXPECT(err == cases[i].error_code);
+	}
+
+	reftable_reader_free(rd);
+	strbuf_release(&buf);
+}
+
+int refname_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_conflict);
+	return 0;
+}
diff --git a/reftable/reftable-blocksource.h b/reftable/reftable-blocksource.h
new file mode 100644
index 0000000..5aa3990
--- /dev/null
+++ b/reftable/reftable-blocksource.h
@@ -0,0 +1,49 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_BLOCKSOURCE_H
+#define REFTABLE_BLOCKSOURCE_H
+
+#include <stdint.h>
+
+/* block_source is a generic wrapper for a seekable readable file.
+ */
+struct reftable_block_source {
+	struct reftable_block_source_vtable *ops;
+	void *arg;
+};
+
+/* a contiguous segment of bytes. It keeps track of its generating block_source
+ * so it can return itself into the pool. */
+struct reftable_block {
+	uint8_t *data;
+	int len;
+	struct reftable_block_source source;
+};
+
+/* block_source_vtable are the operations that make up block_source */
+struct reftable_block_source_vtable {
+	/* returns the size of a block source */
+	uint64_t (*size)(void *source);
+
+	/* reads a segment from the block source. It is an error to read
+	   beyond the end of the block */
+	int (*read_block)(void *source, struct reftable_block *dest,
+			  uint64_t off, uint32_t size);
+	/* mark the block as read; may return the data back to malloc */
+	void (*return_block)(void *source, struct reftable_block *blockp);
+
+	/* release all resources associated with the block source */
+	void (*close)(void *source);
+};
+
+/* opens a file on the file system as a block_source */
+int reftable_block_source_from_file(struct reftable_block_source *block_src,
+				    const char *name);
+
+#endif
diff --git a/reftable/reftable-error.h b/reftable/reftable-error.h
new file mode 100644
index 0000000..6f89bed
--- /dev/null
+++ b/reftable/reftable-error.h
@@ -0,0 +1,62 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_ERROR_H
+#define REFTABLE_ERROR_H
+
+/*
+ * Errors in reftable calls are signaled with negative integer return values. 0
+ * means success.
+ */
+enum reftable_error {
+	/* Unexpected file system behavior */
+	REFTABLE_IO_ERROR = -2,
+
+	/* Format inconsistency on reading data */
+	REFTABLE_FORMAT_ERROR = -3,
+
+	/* File does not exist. Returned from block_source_from_file(), because
+	 * it needs special handling in stack.
+	 */
+	REFTABLE_NOT_EXIST_ERROR = -4,
+
+	/* Trying to write out-of-date data. */
+	REFTABLE_LOCK_ERROR = -5,
+
+	/* Misuse of the API:
+	 *  - on writing a record with NULL refname.
+	 *  - on writing a reftable_ref_record outside the table limits
+	 *  - on writing a ref or log record before the stack's
+	 * next_update_inde*x
+	 *  - on writing a log record with multiline message with
+	 *  exact_log_message unset
+	 *  - on reading a reftable_ref_record from log iterator, or vice versa.
+	 *
+	 * When a call misuses the API, the internal state of the library is
+	 * kept unchanged.
+	 */
+	REFTABLE_API_ERROR = -6,
+
+	/* Decompression error */
+	REFTABLE_ZLIB_ERROR = -7,
+
+	/* Wrote a table without blocks. */
+	REFTABLE_EMPTY_TABLE_ERROR = -8,
+
+	/* Dir/file conflict. */
+	REFTABLE_NAME_CONFLICT = -9,
+
+	/* Invalid ref name. */
+	REFTABLE_REFNAME_ERROR = -10,
+};
+
+/* convert the numeric error code to a string. The string should not be
+ * deallocated. */
+const char *reftable_error_str(int err);
+
+#endif
diff --git a/reftable/reftable-generic.h b/reftable/reftable-generic.h
new file mode 100644
index 0000000..d239751
--- /dev/null
+++ b/reftable/reftable-generic.h
@@ -0,0 +1,47 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_GENERIC_H
+#define REFTABLE_GENERIC_H
+
+#include "reftable-iterator.h"
+
+struct reftable_table_vtable;
+
+/*
+ * Provides a unified API for reading tables, either merged tables, or single
+ * readers. */
+struct reftable_table {
+	struct reftable_table_vtable *ops;
+	void *table_arg;
+};
+
+int reftable_table_seek_log(struct reftable_table *tab,
+			    struct reftable_iterator *it, const char *name);
+
+int reftable_table_seek_ref(struct reftable_table *tab,
+			    struct reftable_iterator *it, const char *name);
+
+/* returns the hash ID from a generic reftable_table */
+uint32_t reftable_table_hash_id(struct reftable_table *tab);
+
+/* returns the max update_index covered by this table. */
+uint64_t reftable_table_max_update_index(struct reftable_table *tab);
+
+/* returns the min update_index covered by this table. */
+uint64_t reftable_table_min_update_index(struct reftable_table *tab);
+
+/* convenience function to read a single ref. Returns < 0 for error, 0
+   for success, and 1 if ref not found. */
+int reftable_table_read_ref(struct reftable_table *tab, const char *name,
+			    struct reftable_ref_record *ref);
+
+/* dump table contents onto stdout for debugging */
+int reftable_table_print(struct reftable_table *tab);
+
+#endif
diff --git a/reftable/reftable-iterator.h b/reftable/reftable-iterator.h
new file mode 100644
index 0000000..d3eee7a
--- /dev/null
+++ b/reftable/reftable-iterator.h
@@ -0,0 +1,39 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_ITERATOR_H
+#define REFTABLE_ITERATOR_H
+
+#include "reftable-record.h"
+
+struct reftable_iterator_vtable;
+
+/* iterator is the generic interface for walking over data stored in a
+ * reftable.
+ */
+struct reftable_iterator {
+	struct reftable_iterator_vtable *ops;
+	void *iter_arg;
+};
+
+/* reads the next reftable_ref_record. Returns < 0 for error, 0 for OK and > 0:
+ * end of iteration.
+ */
+int reftable_iterator_next_ref(struct reftable_iterator *it,
+			       struct reftable_ref_record *ref);
+
+/* reads the next reftable_log_record. Returns < 0 for error, 0 for OK and > 0:
+ * end of iteration.
+ */
+int reftable_iterator_next_log(struct reftable_iterator *it,
+			       struct reftable_log_record *log);
+
+/* releases resources associated with an iterator. */
+void reftable_iterator_destroy(struct reftable_iterator *it);
+
+#endif
diff --git a/reftable/reftable-malloc.h b/reftable/reftable-malloc.h
new file mode 100644
index 0000000..5f2185f
--- /dev/null
+++ b/reftable/reftable-malloc.h
@@ -0,0 +1,18 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_H
+#define REFTABLE_H
+
+#include <stddef.h>
+
+/* Overrides the functions to use for memory management. */
+void reftable_set_alloc(void *(*malloc)(size_t),
+			void *(*realloc)(void *, size_t), void (*free)(void *));
+
+#endif
diff --git a/reftable/reftable-merged.h b/reftable/reftable-merged.h
new file mode 100644
index 0000000..1a6d169
--- /dev/null
+++ b/reftable/reftable-merged.h
@@ -0,0 +1,72 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_MERGED_H
+#define REFTABLE_MERGED_H
+
+#include "reftable-iterator.h"
+
+/*
+ * Merged tables
+ *
+ * A ref database kept in a sequence of table files. The merged_table presents a
+ * unified view to reading (seeking, iterating) a sequence of immutable tables.
+ *
+ * The merged tables are on purpose kept disconnected from their actual storage
+ * (eg. files on disk), because it is useful to merge tables aren't files. For
+ * example, the per-workspace and global ref namespace can be implemented as a
+ * merged table of two stacks of file-backed reftables.
+ */
+
+/* A merged table is implements seeking/iterating over a stack of tables. */
+struct reftable_merged_table;
+
+/* A generic reftable; see below. */
+struct reftable_table;
+
+/* reftable_new_merged_table creates a new merged table. It takes ownership of
+   the stack array.
+*/
+int reftable_new_merged_table(struct reftable_merged_table **dest,
+			      struct reftable_table *stack, int n,
+			      uint32_t hash_id);
+
+/* returns an iterator positioned just before 'name' */
+int reftable_merged_table_seek_ref(struct reftable_merged_table *mt,
+				   struct reftable_iterator *it,
+				   const char *name);
+
+/* returns an iterator for log entry, at given update_index */
+int reftable_merged_table_seek_log_at(struct reftable_merged_table *mt,
+				      struct reftable_iterator *it,
+				      const char *name, uint64_t update_index);
+
+/* like reftable_merged_table_seek_log_at but look for the newest entry. */
+int reftable_merged_table_seek_log(struct reftable_merged_table *mt,
+				   struct reftable_iterator *it,
+				   const char *name);
+
+/* returns the max update_index covered by this merged table. */
+uint64_t
+reftable_merged_table_max_update_index(struct reftable_merged_table *mt);
+
+/* returns the min update_index covered by this merged table. */
+uint64_t
+reftable_merged_table_min_update_index(struct reftable_merged_table *mt);
+
+/* releases memory for the merged_table */
+void reftable_merged_table_free(struct reftable_merged_table *m);
+
+/* return the hash ID of the merged table. */
+uint32_t reftable_merged_table_hash_id(struct reftable_merged_table *m);
+
+/* create a generic table from reftable_merged_table */
+void reftable_table_from_merged_table(struct reftable_table *tab,
+				      struct reftable_merged_table *table);
+
+#endif
diff --git a/reftable/reftable-reader.h b/reftable/reftable-reader.h
new file mode 100644
index 0000000..4a4bc2f
--- /dev/null
+++ b/reftable/reftable-reader.h
@@ -0,0 +1,101 @@
+/*
+  Copyright 2020 Google LLC
+
+  Use of this source code is governed by a BSD-style
+  license that can be found in the LICENSE file or at
+  https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_READER_H
+#define REFTABLE_READER_H
+
+#include "reftable-iterator.h"
+#include "reftable-blocksource.h"
+
+/*
+ * Reading single tables
+ *
+ * The follow routines are for reading single files. For an
+ * application-level interface, skip ahead to struct
+ * reftable_merged_table and struct reftable_stack.
+ */
+
+/* The reader struct is a handle to an open reftable file. */
+struct reftable_reader;
+
+/* Generic table. */
+struct reftable_table;
+
+/* reftable_new_reader opens a reftable for reading. If successful,
+ * returns 0 code and sets pp. The name is used for creating a
+ * stack. Typically, it is the basename of the file. The block source
+ * `src` is owned by the reader, and is closed on calling
+ * reftable_reader_destroy(). On error, the block source `src` is
+ * closed as well.
+ */
+int reftable_new_reader(struct reftable_reader **pp,
+			struct reftable_block_source *src, const char *name);
+
+/* reftable_reader_seek_ref returns an iterator where 'name' would be inserted
+   in the table.  To seek to the start of the table, use name = "".
+
+   example:
+
+   struct reftable_reader *r = NULL;
+   int err = reftable_new_reader(&r, &src, "filename");
+   if (err < 0) { ... }
+   struct reftable_iterator it  = {0};
+   err = reftable_reader_seek_ref(r, &it, "refs/heads/master");
+   if (err < 0) { ... }
+   struct reftable_ref_record ref  = {0};
+   while (1) {
+   err = reftable_iterator_next_ref(&it, &ref);
+   if (err > 0) {
+   break;
+   }
+   if (err < 0) {
+   ..error handling..
+   }
+   ..found..
+   }
+   reftable_iterator_destroy(&it);
+   reftable_ref_record_release(&ref);
+*/
+int reftable_reader_seek_ref(struct reftable_reader *r,
+			     struct reftable_iterator *it, const char *name);
+
+/* returns the hash ID used in this table. */
+uint32_t reftable_reader_hash_id(struct reftable_reader *r);
+
+/* seek to logs for the given name, older than update_index. To seek to the
+   start of the table, use name = "".
+*/
+int reftable_reader_seek_log_at(struct reftable_reader *r,
+				struct reftable_iterator *it, const char *name,
+				uint64_t update_index);
+
+/* seek to newest log entry for given name. */
+int reftable_reader_seek_log(struct reftable_reader *r,
+			     struct reftable_iterator *it, const char *name);
+
+/* closes and deallocates a reader. */
+void reftable_reader_free(struct reftable_reader *);
+
+/* return an iterator for the refs pointing to `oid`. */
+int reftable_reader_refs_for(struct reftable_reader *r,
+			     struct reftable_iterator *it, uint8_t *oid);
+
+/* return the max_update_index for a table */
+uint64_t reftable_reader_max_update_index(struct reftable_reader *r);
+
+/* return the min_update_index for a table */
+uint64_t reftable_reader_min_update_index(struct reftable_reader *r);
+
+/* creates a generic table from a file reader. */
+void reftable_table_from_reader(struct reftable_table *tab,
+				struct reftable_reader *reader);
+
+/* print table onto stdout for debugging. */
+int reftable_reader_print_file(const char *tablename);
+
+#endif
diff --git a/reftable/reftable-record.h b/reftable/reftable-record.h
new file mode 100644
index 0000000..5370d22
--- /dev/null
+++ b/reftable/reftable-record.h
@@ -0,0 +1,114 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_RECORD_H
+#define REFTABLE_RECORD_H
+
+#include <stdint.h>
+
+/*
+ * Basic data types
+ *
+ * Reftables store the state of each ref in struct reftable_ref_record, and they
+ * store a sequence of reflog updates in struct reftable_log_record.
+ */
+
+/* reftable_ref_record holds a ref database entry target_value */
+struct reftable_ref_record {
+	char *refname; /* Name of the ref, malloced. */
+	uint64_t update_index; /* Logical timestamp at which this value is
+				* written */
+
+	enum {
+		/* tombstone to hide deletions from earlier tables */
+		REFTABLE_REF_DELETION = 0x0,
+
+		/* a simple ref */
+		REFTABLE_REF_VAL1 = 0x1,
+		/* a tag, plus its peeled hash */
+		REFTABLE_REF_VAL2 = 0x2,
+
+		/* a symbolic reference */
+		REFTABLE_REF_SYMREF = 0x3,
+#define REFTABLE_NR_REF_VALUETYPES 4
+	} value_type;
+	union {
+		uint8_t *val1; /* malloced hash. */
+		struct {
+			uint8_t *value; /* first value, malloced hash  */
+			uint8_t *target_value; /* second value, malloced hash */
+		} val2;
+		char *symref; /* referent, malloced 0-terminated string */
+	} value;
+};
+
+/* Returns the first hash, or NULL if `rec` is not of type
+ * REFTABLE_REF_VAL1 or REFTABLE_REF_VAL2. */
+uint8_t *reftable_ref_record_val1(struct reftable_ref_record *rec);
+
+/* Returns the second hash, or NULL if `rec` is not of type
+ * REFTABLE_REF_VAL2. */
+uint8_t *reftable_ref_record_val2(struct reftable_ref_record *rec);
+
+/* returns whether 'ref' represents a deletion */
+int reftable_ref_record_is_deletion(const struct reftable_ref_record *ref);
+
+/* prints a reftable_ref_record onto stdout. Useful for debugging. */
+void reftable_ref_record_print(struct reftable_ref_record *ref,
+			       uint32_t hash_id);
+
+/* frees and nulls all pointer values inside `ref`. */
+void reftable_ref_record_release(struct reftable_ref_record *ref);
+
+/* returns whether two reftable_ref_records are the same. Useful for testing. */
+int reftable_ref_record_equal(struct reftable_ref_record *a,
+			      struct reftable_ref_record *b, int hash_size);
+
+/* reftable_log_record holds a reflog entry */
+struct reftable_log_record {
+	char *refname;
+	uint64_t update_index; /* logical timestamp of a transactional update.
+				*/
+
+	enum {
+		/* tombstone to hide deletions from earlier tables */
+		REFTABLE_LOG_DELETION = 0x0,
+
+		/* a simple update */
+		REFTABLE_LOG_UPDATE = 0x1,
+#define REFTABLE_NR_LOG_VALUETYPES 2
+	} value_type;
+
+	union {
+		struct {
+			uint8_t *new_hash;
+			uint8_t *old_hash;
+			char *name;
+			char *email;
+			uint64_t time;
+			int16_t tz_offset;
+			char *message;
+		} update;
+	} value;
+};
+
+/* returns whether 'ref' represents the deletion of a log record. */
+int reftable_log_record_is_deletion(const struct reftable_log_record *log);
+
+/* frees and nulls all pointer values. */
+void reftable_log_record_release(struct reftable_log_record *log);
+
+/* returns whether two records are equal. Useful for testing. */
+int reftable_log_record_equal(struct reftable_log_record *a,
+			      struct reftable_log_record *b, int hash_size);
+
+/* dumps a reftable_log_record on stdout, for debugging/testing. */
+void reftable_log_record_print(struct reftable_log_record *log,
+			       uint32_t hash_id);
+
+#endif
diff --git a/reftable/reftable-stack.h b/reftable/reftable-stack.h
new file mode 100644
index 0000000..1b602dd
--- /dev/null
+++ b/reftable/reftable-stack.h
@@ -0,0 +1,128 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_STACK_H
+#define REFTABLE_STACK_H
+
+#include "reftable-writer.h"
+
+/*
+ * The stack presents an interface to a mutable sequence of reftables.
+
+ * A stack can be mutated by pushing a table to the top of the stack.
+
+ * The reftable_stack automatically compacts files on disk to ensure good
+ * amortized performance.
+ *
+ * For windows and other platforms that cannot have open files as rename
+ * destinations, concurrent access from multiple processes needs the rand()
+ * random seed to be randomized.
+ */
+struct reftable_stack;
+
+/* open a new reftable stack. The tables along with the table list will be
+ *  stored in 'dir'. Typically, this should be .git/reftables.
+ */
+int reftable_new_stack(struct reftable_stack **dest, const char *dir,
+		       struct reftable_write_options config);
+
+/* returns the update_index at which a next table should be written. */
+uint64_t reftable_stack_next_update_index(struct reftable_stack *st);
+
+/* holds a transaction to add tables at the top of a stack. */
+struct reftable_addition;
+
+/*
+ * returns a new transaction to add reftables to the given stack. As a side
+ * effect, the ref database is locked.
+ */
+int reftable_stack_new_addition(struct reftable_addition **dest,
+				struct reftable_stack *st);
+
+/* Adds a reftable to transaction. */
+int reftable_addition_add(struct reftable_addition *add,
+			  int (*write_table)(struct reftable_writer *wr,
+					     void *arg),
+			  void *arg);
+
+/* Commits the transaction, releasing the lock. After calling this,
+ * reftable_addition_destroy should still be called.
+ */
+int reftable_addition_commit(struct reftable_addition *add);
+
+/* Release all non-committed data from the transaction, and deallocate the
+ * transaction. Releases the lock if held. */
+void reftable_addition_destroy(struct reftable_addition *add);
+
+/* add a new table to the stack. The write_table function must call
+ * reftable_writer_set_limits, add refs and return an error value. */
+int reftable_stack_add(struct reftable_stack *st,
+		       int (*write_table)(struct reftable_writer *wr,
+					  void *write_arg),
+		       void *write_arg);
+
+/* returns the merged_table for seeking. This table is valid until the
+ * next write or reload, and should not be closed or deleted.
+ */
+struct reftable_merged_table *
+reftable_stack_merged_table(struct reftable_stack *st);
+
+/* frees all resources associated with the stack. */
+void reftable_stack_destroy(struct reftable_stack *st);
+
+/* Reloads the stack if necessary. This is very cheap to run if the stack was up
+ * to date */
+int reftable_stack_reload(struct reftable_stack *st);
+
+/* Policy for expiring reflog entries. */
+struct reftable_log_expiry_config {
+	/* Drop entries older than this timestamp */
+	uint64_t time;
+
+	/* Drop older entries */
+	uint64_t min_update_index;
+};
+
+/* compacts all reftables into a giant table. Expire reflog entries if config is
+ * non-NULL */
+int reftable_stack_compact_all(struct reftable_stack *st,
+			       struct reftable_log_expiry_config *config);
+
+/* heuristically compact unbalanced table stack. */
+int reftable_stack_auto_compact(struct reftable_stack *st);
+
+/* delete stale .ref tables. */
+int reftable_stack_clean(struct reftable_stack *st);
+
+/* convenience function to read a single ref. Returns < 0 for error, 0 for
+ * success, and 1 if ref not found. */
+int reftable_stack_read_ref(struct reftable_stack *st, const char *refname,
+			    struct reftable_ref_record *ref);
+
+/* convenience function to read a single log. Returns < 0 for error, 0 for
+ * success, and 1 if ref not found. */
+int reftable_stack_read_log(struct reftable_stack *st, const char *refname,
+			    struct reftable_log_record *log);
+
+/* statistics on past compactions. */
+struct reftable_compaction_stats {
+	uint64_t bytes; /* total number of bytes written */
+	uint64_t entries_written; /* total number of entries written, including
+				     failures. */
+	int attempts; /* how often we tried to compact */
+	int failures; /* failures happen on concurrent updates */
+};
+
+/* return statistics for compaction up till now. */
+struct reftable_compaction_stats *
+reftable_stack_compaction_stats(struct reftable_stack *st);
+
+/* print the entire stack represented by the directory */
+int reftable_stack_print_directory(const char *stackdir, uint32_t hash_id);
+
+#endif
diff --git a/reftable/reftable-tests.h b/reftable/reftable-tests.h
new file mode 100644
index 0000000..0019cbc
--- /dev/null
+++ b/reftable/reftable-tests.h
@@ -0,0 +1,23 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_TESTS_H
+#define REFTABLE_TESTS_H
+
+int basics_test_main(int argc, const char **argv);
+int block_test_main(int argc, const char **argv);
+int merged_test_main(int argc, const char **argv);
+int pq_test_main(int argc, const char **argv);
+int record_test_main(int argc, const char **argv);
+int refname_test_main(int argc, const char **argv);
+int readwrite_test_main(int argc, const char **argv);
+int stack_test_main(int argc, const char **argv);
+int tree_test_main(int argc, const char **argv);
+int reftable_dump_main(int argc, char *const *argv);
+
+#endif
diff --git a/reftable/reftable-writer.h b/reftable/reftable-writer.h
new file mode 100644
index 0000000..af36462
--- /dev/null
+++ b/reftable/reftable-writer.h
@@ -0,0 +1,148 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef REFTABLE_WRITER_H
+#define REFTABLE_WRITER_H
+
+#include "reftable-record.h"
+
+#include <stdint.h>
+#include <unistd.h> /* ssize_t */
+
+/* Writing single reftables */
+
+/* reftable_write_options sets options for writing a single reftable. */
+struct reftable_write_options {
+	/* boolean: do not pad out blocks to block size. */
+	unsigned unpadded : 1;
+
+	/* the blocksize. Should be less than 2^24. */
+	uint32_t block_size;
+
+	/* boolean: do not generate a SHA1 => ref index. */
+	unsigned skip_index_objects : 1;
+
+	/* how often to write complete keys in each block. */
+	int restart_interval;
+
+	/* 4-byte identifier ("sha1", "s256") of the hash.
+	 * Defaults to SHA1 if unset
+	 */
+	uint32_t hash_id;
+
+	/* boolean: do not check ref names for validity or dir/file conflicts.
+	 */
+	unsigned skip_name_check : 1;
+
+	/* boolean: copy log messages exactly. If unset, check that the message
+	 *   is a single line, and add '\n' if missing.
+	 */
+	unsigned exact_log_message : 1;
+};
+
+/* reftable_block_stats holds statistics for a single block type */
+struct reftable_block_stats {
+	/* total number of entries written */
+	int entries;
+	/* total number of key restarts */
+	int restarts;
+	/* total number of blocks */
+	int blocks;
+	/* total number of index blocks */
+	int index_blocks;
+	/* depth of the index */
+	int max_index_level;
+
+	/* offset of the first block for this type */
+	uint64_t offset;
+	/* offset of the top level index block for this type, or 0 if not
+	 * present */
+	uint64_t index_offset;
+};
+
+/* stats holds overall statistics for a single reftable */
+struct reftable_stats {
+	/* total number of blocks written. */
+	int blocks;
+	/* stats for ref data */
+	struct reftable_block_stats ref_stats;
+	/* stats for the SHA1 to ref map. */
+	struct reftable_block_stats obj_stats;
+	/* stats for index blocks */
+	struct reftable_block_stats idx_stats;
+	/* stats for log blocks */
+	struct reftable_block_stats log_stats;
+
+	/* disambiguation length of shortened object IDs. */
+	int object_id_len;
+};
+
+/* reftable_new_writer creates a new writer */
+struct reftable_writer *
+reftable_new_writer(ssize_t (*writer_func)(void *, const void *, size_t),
+		    void *writer_arg, struct reftable_write_options *opts);
+
+/* Set the range of update indices for the records we will add. When writing a
+   table into a stack, the min should be at least
+   reftable_stack_next_update_index(), or REFTABLE_API_ERROR is returned.
+
+   For transactional updates to a stack, typically min==max, and the
+   update_index can be obtained by inspeciting the stack. When converting an
+   existing ref database into a single reftable, this would be a range of
+   update-index timestamps.
+ */
+void reftable_writer_set_limits(struct reftable_writer *w, uint64_t min,
+				uint64_t max);
+
+/*
+  Add a reftable_ref_record. The record should have names that come after
+  already added records.
+
+  The update_index must be within the limits set by
+  reftable_writer_set_limits(), or REFTABLE_API_ERROR is returned. It is an
+  REFTABLE_API_ERROR error to write a ref record after a log record.
+*/
+int reftable_writer_add_ref(struct reftable_writer *w,
+			    struct reftable_ref_record *ref);
+
+/*
+  Convenience function to add multiple reftable_ref_records; the function sorts
+  the records before adding them, reordering the records array passed in.
+*/
+int reftable_writer_add_refs(struct reftable_writer *w,
+			     struct reftable_ref_record *refs, int n);
+
+/*
+  adds reftable_log_records. Log records are keyed by (refname, decreasing
+  update_index). The key for the record added must come after the already added
+  log records.
+*/
+int reftable_writer_add_log(struct reftable_writer *w,
+			    struct reftable_log_record *log);
+
+/*
+  Convenience function to add multiple reftable_log_records; the function sorts
+  the records before adding them, reordering records array passed in.
+*/
+int reftable_writer_add_logs(struct reftable_writer *w,
+			     struct reftable_log_record *logs, int n);
+
+/* reftable_writer_close finalizes the reftable. The writer is retained so
+ * statistics can be inspected. */
+int reftable_writer_close(struct reftable_writer *w);
+
+/* writer_stats returns the statistics on the reftable being written.
+
+   This struct becomes invalid when the writer is freed.
+ */
+const struct reftable_stats *writer_stats(struct reftable_writer *w);
+
+/* reftable_writer_free deallocates memory for the writer */
+void reftable_writer_free(struct reftable_writer *w);
+
+#endif
diff --git a/reftable/reftable.c b/reftable/reftable.c
new file mode 100644
index 0000000..0e4607a
--- /dev/null
+++ b/reftable/reftable.c
@@ -0,0 +1,115 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "basics.h"
+#include "record.h"
+#include "generic.h"
+#include "reftable-iterator.h"
+#include "reftable-generic.h"
+
+int reftable_table_seek_ref(struct reftable_table *tab,
+			    struct reftable_iterator *it, const char *name)
+{
+	struct reftable_ref_record ref = {
+		.refname = (char *)name,
+	};
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, &ref);
+	return tab->ops->seek_record(tab->table_arg, it, &rec);
+}
+
+int reftable_table_read_ref(struct reftable_table *tab, const char *name,
+			    struct reftable_ref_record *ref)
+{
+	struct reftable_iterator it = { NULL };
+	int err = reftable_table_seek_ref(tab, &it, name);
+	if (err)
+		goto done;
+
+	err = reftable_iterator_next_ref(&it, ref);
+	if (err)
+		goto done;
+
+	if (strcmp(ref->refname, name) ||
+	    reftable_ref_record_is_deletion(ref)) {
+		reftable_ref_record_release(ref);
+		err = 1;
+		goto done;
+	}
+
+done:
+	reftable_iterator_destroy(&it);
+	return err;
+}
+
+uint64_t reftable_table_max_update_index(struct reftable_table *tab)
+{
+	return tab->ops->max_update_index(tab->table_arg);
+}
+
+uint64_t reftable_table_min_update_index(struct reftable_table *tab)
+{
+	return tab->ops->min_update_index(tab->table_arg);
+}
+
+uint32_t reftable_table_hash_id(struct reftable_table *tab)
+{
+	return tab->ops->hash_id(tab->table_arg);
+}
+
+void reftable_iterator_destroy(struct reftable_iterator *it)
+{
+	if (!it->ops) {
+		return;
+	}
+	it->ops->close(it->iter_arg);
+	it->ops = NULL;
+	FREE_AND_NULL(it->iter_arg);
+}
+
+int reftable_iterator_next_ref(struct reftable_iterator *it,
+			       struct reftable_ref_record *ref)
+{
+	struct reftable_record rec = { NULL };
+	reftable_record_from_ref(&rec, ref);
+	return iterator_next(it, &rec);
+}
+
+int reftable_iterator_next_log(struct reftable_iterator *it,
+			       struct reftable_log_record *log)
+{
+	struct reftable_record rec = { NULL };
+	reftable_record_from_log(&rec, log);
+	return iterator_next(it, &rec);
+}
+
+int iterator_next(struct reftable_iterator *it, struct reftable_record *rec)
+{
+	return it->ops->next(it->iter_arg, rec);
+}
+
+static int empty_iterator_next(void *arg, struct reftable_record *rec)
+{
+	return 1;
+}
+
+static void empty_iterator_close(void *arg)
+{
+}
+
+static struct reftable_iterator_vtable empty_vtable = {
+	.next = &empty_iterator_next,
+	.close = &empty_iterator_close,
+};
+
+void iterator_set_empty(struct reftable_iterator *it)
+{
+	assert(!it->ops);
+	it->iter_arg = NULL;
+	it->ops = &empty_vtable;
+}
diff --git a/reftable/stack.c b/reftable/stack.c
new file mode 100644
index 0000000..df5021e
--- /dev/null
+++ b/reftable/stack.c
@@ -0,0 +1,1396 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "stack.h"
+
+#include "system.h"
+#include "merged.h"
+#include "reader.h"
+#include "refname.h"
+#include "reftable-error.h"
+#include "reftable-record.h"
+#include "reftable-merged.h"
+#include "writer.h"
+
+static int stack_try_add(struct reftable_stack *st,
+			 int (*write_table)(struct reftable_writer *wr,
+					    void *arg),
+			 void *arg);
+static int stack_write_compact(struct reftable_stack *st,
+			       struct reftable_writer *wr, int first, int last,
+			       struct reftable_log_expiry_config *config);
+static int stack_check_addition(struct reftable_stack *st,
+				const char *new_tab_name);
+static void reftable_addition_close(struct reftable_addition *add);
+static int reftable_stack_reload_maybe_reuse(struct reftable_stack *st,
+					     int reuse_open);
+
+static void stack_filename(struct strbuf *dest, struct reftable_stack *st,
+			   const char *name)
+{
+	strbuf_reset(dest);
+	strbuf_addstr(dest, st->reftable_dir);
+	strbuf_addstr(dest, "/");
+	strbuf_addstr(dest, name);
+}
+
+static ssize_t reftable_fd_write(void *arg, const void *data, size_t sz)
+{
+	int *fdp = (int *)arg;
+	return write(*fdp, data, sz);
+}
+
+int reftable_new_stack(struct reftable_stack **dest, const char *dir,
+		       struct reftable_write_options config)
+{
+	struct reftable_stack *p =
+		reftable_calloc(sizeof(struct reftable_stack));
+	struct strbuf list_file_name = STRBUF_INIT;
+	int err = 0;
+
+	if (config.hash_id == 0) {
+		config.hash_id = GIT_SHA1_FORMAT_ID;
+	}
+
+	*dest = NULL;
+
+	strbuf_reset(&list_file_name);
+	strbuf_addstr(&list_file_name, dir);
+	strbuf_addstr(&list_file_name, "/tables.list");
+
+	p->list_file = strbuf_detach(&list_file_name, NULL);
+	p->reftable_dir = xstrdup(dir);
+	p->config = config;
+
+	err = reftable_stack_reload_maybe_reuse(p, 1);
+	if (err < 0) {
+		reftable_stack_destroy(p);
+	} else {
+		*dest = p;
+	}
+	return err;
+}
+
+static int fd_read_lines(int fd, char ***namesp)
+{
+	off_t size = lseek(fd, 0, SEEK_END);
+	char *buf = NULL;
+	int err = 0;
+	if (size < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+	err = lseek(fd, 0, SEEK_SET);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	buf = reftable_malloc(size + 1);
+	if (read(fd, buf, size) != size) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+	buf[size] = 0;
+
+	parse_names(buf, size, namesp);
+
+done:
+	reftable_free(buf);
+	return err;
+}
+
+int read_lines(const char *filename, char ***namesp)
+{
+	int fd = open(filename, O_RDONLY);
+	int err = 0;
+	if (fd < 0) {
+		if (errno == ENOENT) {
+			*namesp = reftable_calloc(sizeof(char *));
+			return 0;
+		}
+
+		return REFTABLE_IO_ERROR;
+	}
+	err = fd_read_lines(fd, namesp);
+	close(fd);
+	return err;
+}
+
+struct reftable_merged_table *
+reftable_stack_merged_table(struct reftable_stack *st)
+{
+	return st->merged;
+}
+
+static int has_name(char **names, const char *name)
+{
+	while (*names) {
+		if (!strcmp(*names, name))
+			return 1;
+		names++;
+	}
+	return 0;
+}
+
+/* Close and free the stack */
+void reftable_stack_destroy(struct reftable_stack *st)
+{
+	char **names = NULL;
+	int err = 0;
+	if (st->merged) {
+		reftable_merged_table_free(st->merged);
+		st->merged = NULL;
+	}
+
+	err = read_lines(st->list_file, &names);
+	if (err < 0) {
+		FREE_AND_NULL(names);
+	}
+
+	if (st->readers) {
+		int i = 0;
+		struct strbuf filename = STRBUF_INIT;
+		for (i = 0; i < st->readers_len; i++) {
+			const char *name = reader_name(st->readers[i]);
+			strbuf_reset(&filename);
+			if (names && !has_name(names, name)) {
+				stack_filename(&filename, st, name);
+			}
+			reftable_reader_free(st->readers[i]);
+
+			if (filename.len) {
+				/* On Windows, can only unlink after closing. */
+				unlink(filename.buf);
+			}
+		}
+		strbuf_release(&filename);
+		st->readers_len = 0;
+		FREE_AND_NULL(st->readers);
+	}
+	FREE_AND_NULL(st->list_file);
+	FREE_AND_NULL(st->reftable_dir);
+	reftable_free(st);
+	free_names(names);
+}
+
+static struct reftable_reader **stack_copy_readers(struct reftable_stack *st,
+						   int cur_len)
+{
+	struct reftable_reader **cur =
+		reftable_calloc(sizeof(struct reftable_reader *) * cur_len);
+	int i = 0;
+	for (i = 0; i < cur_len; i++) {
+		cur[i] = st->readers[i];
+	}
+	return cur;
+}
+
+static int reftable_stack_reload_once(struct reftable_stack *st, char **names,
+				      int reuse_open)
+{
+	int cur_len = !st->merged ? 0 : st->merged->stack_len;
+	struct reftable_reader **cur = stack_copy_readers(st, cur_len);
+	int err = 0;
+	int names_len = names_length(names);
+	struct reftable_reader **new_readers =
+		reftable_calloc(sizeof(struct reftable_reader *) * names_len);
+	struct reftable_table *new_tables =
+		reftable_calloc(sizeof(struct reftable_table) * names_len);
+	int new_readers_len = 0;
+	struct reftable_merged_table *new_merged = NULL;
+	int i;
+
+	while (*names) {
+		struct reftable_reader *rd = NULL;
+		char *name = *names++;
+
+		/* this is linear; we assume compaction keeps the number of
+		   tables under control so this is not quadratic. */
+		int j = 0;
+		for (j = 0; reuse_open && j < cur_len; j++) {
+			if (cur[j] && 0 == strcmp(cur[j]->name, name)) {
+				rd = cur[j];
+				cur[j] = NULL;
+				break;
+			}
+		}
+
+		if (!rd) {
+			struct reftable_block_source src = { NULL };
+			struct strbuf table_path = STRBUF_INIT;
+			stack_filename(&table_path, st, name);
+
+			err = reftable_block_source_from_file(&src,
+							      table_path.buf);
+			strbuf_release(&table_path);
+
+			if (err < 0)
+				goto done;
+
+			err = reftable_new_reader(&rd, &src, name);
+			if (err < 0)
+				goto done;
+		}
+
+		new_readers[new_readers_len] = rd;
+		reftable_table_from_reader(&new_tables[new_readers_len], rd);
+		new_readers_len++;
+	}
+
+	/* success! */
+	err = reftable_new_merged_table(&new_merged, new_tables,
+					new_readers_len, st->config.hash_id);
+	if (err < 0)
+		goto done;
+
+	new_tables = NULL;
+	st->readers_len = new_readers_len;
+	if (st->merged) {
+		merged_table_release(st->merged);
+		reftable_merged_table_free(st->merged);
+	}
+	if (st->readers) {
+		reftable_free(st->readers);
+	}
+	st->readers = new_readers;
+	new_readers = NULL;
+	new_readers_len = 0;
+
+	new_merged->suppress_deletions = 1;
+	st->merged = new_merged;
+	for (i = 0; i < cur_len; i++) {
+		if (cur[i]) {
+			const char *name = reader_name(cur[i]);
+			struct strbuf filename = STRBUF_INIT;
+			stack_filename(&filename, st, name);
+
+			reader_close(cur[i]);
+			reftable_reader_free(cur[i]);
+
+			/* On Windows, can only unlink after closing. */
+			unlink(filename.buf);
+
+			strbuf_release(&filename);
+		}
+	}
+
+done:
+	for (i = 0; i < new_readers_len; i++) {
+		reader_close(new_readers[i]);
+		reftable_reader_free(new_readers[i]);
+	}
+	reftable_free(new_readers);
+	reftable_free(new_tables);
+	reftable_free(cur);
+	return err;
+}
+
+/* return negative if a before b. */
+static int tv_cmp(struct timeval *a, struct timeval *b)
+{
+	time_t diff = a->tv_sec - b->tv_sec;
+	int udiff = a->tv_usec - b->tv_usec;
+
+	if (diff != 0)
+		return diff;
+
+	return udiff;
+}
+
+static int reftable_stack_reload_maybe_reuse(struct reftable_stack *st,
+					     int reuse_open)
+{
+	struct timeval deadline = { 0 };
+	int err = gettimeofday(&deadline, NULL);
+	int64_t delay = 0;
+	int tries = 0;
+	if (err < 0)
+		return err;
+
+	deadline.tv_sec += 3;
+	while (1) {
+		char **names = NULL;
+		char **names_after = NULL;
+		struct timeval now = { 0 };
+		int err = gettimeofday(&now, NULL);
+		int err2 = 0;
+		if (err < 0) {
+			return err;
+		}
+
+		/* Only look at deadlines after the first few times. This
+		   simplifies debugging in GDB */
+		tries++;
+		if (tries > 3 && tv_cmp(&now, &deadline) >= 0) {
+			break;
+		}
+
+		err = read_lines(st->list_file, &names);
+		if (err < 0) {
+			free_names(names);
+			return err;
+		}
+		err = reftable_stack_reload_once(st, names, reuse_open);
+		if (err == 0) {
+			free_names(names);
+			break;
+		}
+		if (err != REFTABLE_NOT_EXIST_ERROR) {
+			free_names(names);
+			return err;
+		}
+
+		/* err == REFTABLE_NOT_EXIST_ERROR can be caused by a concurrent
+		   writer. Check if there was one by checking if the name list
+		   changed.
+		*/
+		err2 = read_lines(st->list_file, &names_after);
+		if (err2 < 0) {
+			free_names(names);
+			return err2;
+		}
+
+		if (names_equal(names_after, names)) {
+			free_names(names);
+			free_names(names_after);
+			return err;
+		}
+		free_names(names);
+		free_names(names_after);
+
+		delay = delay + (delay * rand()) / RAND_MAX + 1;
+		sleep_millisec(delay);
+	}
+
+	return 0;
+}
+
+/* -1 = error
+ 0 = up to date
+ 1 = changed. */
+static int stack_uptodate(struct reftable_stack *st)
+{
+	char **names = NULL;
+	int err = read_lines(st->list_file, &names);
+	int i = 0;
+	if (err < 0)
+		return err;
+
+	for (i = 0; i < st->readers_len; i++) {
+		if (!names[i]) {
+			err = 1;
+			goto done;
+		}
+
+		if (strcmp(st->readers[i]->name, names[i])) {
+			err = 1;
+			goto done;
+		}
+	}
+
+	if (names[st->merged->stack_len]) {
+		err = 1;
+		goto done;
+	}
+
+done:
+	free_names(names);
+	return err;
+}
+
+int reftable_stack_reload(struct reftable_stack *st)
+{
+	int err = stack_uptodate(st);
+	if (err > 0)
+		return reftable_stack_reload_maybe_reuse(st, 1);
+	return err;
+}
+
+int reftable_stack_add(struct reftable_stack *st,
+		       int (*write)(struct reftable_writer *wr, void *arg),
+		       void *arg)
+{
+	int err = stack_try_add(st, write, arg);
+	if (err < 0) {
+		if (err == REFTABLE_LOCK_ERROR) {
+			/* Ignore error return, we want to propagate
+			   REFTABLE_LOCK_ERROR.
+			*/
+			reftable_stack_reload(st);
+		}
+		return err;
+	}
+
+	if (!st->disable_auto_compact)
+		return reftable_stack_auto_compact(st);
+
+	return 0;
+}
+
+static void format_name(struct strbuf *dest, uint64_t min, uint64_t max)
+{
+	char buf[100];
+	uint32_t rnd = (uint32_t)rand();
+	snprintf(buf, sizeof(buf), "0x%012" PRIx64 "-0x%012" PRIx64 "-%08x",
+		 min, max, rnd);
+	strbuf_reset(dest);
+	strbuf_addstr(dest, buf);
+}
+
+struct reftable_addition {
+	int lock_file_fd;
+	struct strbuf lock_file_name;
+	struct reftable_stack *stack;
+
+	char **new_tables;
+	int new_tables_len;
+	uint64_t next_update_index;
+};
+
+#define REFTABLE_ADDITION_INIT                \
+	{                                     \
+		.lock_file_name = STRBUF_INIT \
+	}
+
+static int reftable_stack_init_addition(struct reftable_addition *add,
+					struct reftable_stack *st)
+{
+	int err = 0;
+	add->stack = st;
+
+	strbuf_reset(&add->lock_file_name);
+	strbuf_addstr(&add->lock_file_name, st->list_file);
+	strbuf_addstr(&add->lock_file_name, ".lock");
+
+	add->lock_file_fd = open(add->lock_file_name.buf,
+				 O_EXCL | O_CREAT | O_WRONLY, 0644);
+	if (add->lock_file_fd < 0) {
+		if (errno == EEXIST) {
+			err = REFTABLE_LOCK_ERROR;
+		} else {
+			err = REFTABLE_IO_ERROR;
+		}
+		goto done;
+	}
+	err = stack_uptodate(st);
+	if (err < 0)
+		goto done;
+
+	if (err > 1) {
+		err = REFTABLE_LOCK_ERROR;
+		goto done;
+	}
+
+	add->next_update_index = reftable_stack_next_update_index(st);
+done:
+	if (err) {
+		reftable_addition_close(add);
+	}
+	return err;
+}
+
+static void reftable_addition_close(struct reftable_addition *add)
+{
+	int i = 0;
+	struct strbuf nm = STRBUF_INIT;
+	for (i = 0; i < add->new_tables_len; i++) {
+		stack_filename(&nm, add->stack, add->new_tables[i]);
+		unlink(nm.buf);
+		reftable_free(add->new_tables[i]);
+		add->new_tables[i] = NULL;
+	}
+	reftable_free(add->new_tables);
+	add->new_tables = NULL;
+	add->new_tables_len = 0;
+
+	if (add->lock_file_fd > 0) {
+		close(add->lock_file_fd);
+		add->lock_file_fd = 0;
+	}
+	if (add->lock_file_name.len > 0) {
+		unlink(add->lock_file_name.buf);
+		strbuf_release(&add->lock_file_name);
+	}
+
+	strbuf_release(&nm);
+}
+
+void reftable_addition_destroy(struct reftable_addition *add)
+{
+	if (!add) {
+		return;
+	}
+	reftable_addition_close(add);
+	reftable_free(add);
+}
+
+int reftable_addition_commit(struct reftable_addition *add)
+{
+	struct strbuf table_list = STRBUF_INIT;
+	int i = 0;
+	int err = 0;
+	if (add->new_tables_len == 0)
+		goto done;
+
+	for (i = 0; i < add->stack->merged->stack_len; i++) {
+		strbuf_addstr(&table_list, add->stack->readers[i]->name);
+		strbuf_addstr(&table_list, "\n");
+	}
+	for (i = 0; i < add->new_tables_len; i++) {
+		strbuf_addstr(&table_list, add->new_tables[i]);
+		strbuf_addstr(&table_list, "\n");
+	}
+
+	err = write(add->lock_file_fd, table_list.buf, table_list.len);
+	strbuf_release(&table_list);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	err = close(add->lock_file_fd);
+	add->lock_file_fd = 0;
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	err = rename(add->lock_file_name.buf, add->stack->list_file);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	/* success, no more state to clean up. */
+	strbuf_release(&add->lock_file_name);
+	for (i = 0; i < add->new_tables_len; i++) {
+		reftable_free(add->new_tables[i]);
+	}
+	reftable_free(add->new_tables);
+	add->new_tables = NULL;
+	add->new_tables_len = 0;
+
+	err = reftable_stack_reload(add->stack);
+done:
+	reftable_addition_close(add);
+	return err;
+}
+
+int reftable_stack_new_addition(struct reftable_addition **dest,
+				struct reftable_stack *st)
+{
+	int err = 0;
+	struct reftable_addition empty = REFTABLE_ADDITION_INIT;
+	*dest = reftable_calloc(sizeof(**dest));
+	**dest = empty;
+	err = reftable_stack_init_addition(*dest, st);
+	if (err) {
+		reftable_free(*dest);
+		*dest = NULL;
+	}
+	return err;
+}
+
+static int stack_try_add(struct reftable_stack *st,
+			 int (*write_table)(struct reftable_writer *wr,
+					    void *arg),
+			 void *arg)
+{
+	struct reftable_addition add = REFTABLE_ADDITION_INIT;
+	int err = reftable_stack_init_addition(&add, st);
+	if (err < 0)
+		goto done;
+	if (err > 0) {
+		err = REFTABLE_LOCK_ERROR;
+		goto done;
+	}
+
+	err = reftable_addition_add(&add, write_table, arg);
+	if (err < 0)
+		goto done;
+
+	err = reftable_addition_commit(&add);
+done:
+	reftable_addition_close(&add);
+	return err;
+}
+
+int reftable_addition_add(struct reftable_addition *add,
+			  int (*write_table)(struct reftable_writer *wr,
+					     void *arg),
+			  void *arg)
+{
+	struct strbuf temp_tab_file_name = STRBUF_INIT;
+	struct strbuf tab_file_name = STRBUF_INIT;
+	struct strbuf next_name = STRBUF_INIT;
+	struct reftable_writer *wr = NULL;
+	int err = 0;
+	int tab_fd = 0;
+
+	strbuf_reset(&next_name);
+	format_name(&next_name, add->next_update_index, add->next_update_index);
+
+	stack_filename(&temp_tab_file_name, add->stack, next_name.buf);
+	strbuf_addstr(&temp_tab_file_name, ".temp.XXXXXX");
+
+	tab_fd = mkstemp(temp_tab_file_name.buf);
+	if (tab_fd < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	wr = reftable_new_writer(reftable_fd_write, &tab_fd,
+				 &add->stack->config);
+	err = write_table(wr, arg);
+	if (err < 0)
+		goto done;
+
+	err = reftable_writer_close(wr);
+	if (err == REFTABLE_EMPTY_TABLE_ERROR) {
+		err = 0;
+		goto done;
+	}
+	if (err < 0)
+		goto done;
+
+	err = close(tab_fd);
+	tab_fd = 0;
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	err = stack_check_addition(add->stack, temp_tab_file_name.buf);
+	if (err < 0)
+		goto done;
+
+	if (wr->min_update_index < add->next_update_index) {
+		err = REFTABLE_API_ERROR;
+		goto done;
+	}
+
+	format_name(&next_name, wr->min_update_index, wr->max_update_index);
+	strbuf_addstr(&next_name, ".ref");
+
+	stack_filename(&tab_file_name, add->stack, next_name.buf);
+
+	/*
+	  On windows, this relies on rand() picking a unique destination name.
+	  Maybe we should do retry loop as well?
+	 */
+	err = rename(temp_tab_file_name.buf, tab_file_name.buf);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		goto done;
+	}
+
+	add->new_tables = reftable_realloc(add->new_tables,
+					   sizeof(*add->new_tables) *
+						   (add->new_tables_len + 1));
+	add->new_tables[add->new_tables_len] = strbuf_detach(&next_name, NULL);
+	add->new_tables_len++;
+done:
+	if (tab_fd > 0) {
+		close(tab_fd);
+		tab_fd = 0;
+	}
+	if (temp_tab_file_name.len > 0) {
+		unlink(temp_tab_file_name.buf);
+	}
+
+	strbuf_release(&temp_tab_file_name);
+	strbuf_release(&tab_file_name);
+	strbuf_release(&next_name);
+	reftable_writer_free(wr);
+	return err;
+}
+
+uint64_t reftable_stack_next_update_index(struct reftable_stack *st)
+{
+	int sz = st->merged->stack_len;
+	if (sz > 0)
+		return reftable_reader_max_update_index(st->readers[sz - 1]) +
+		       1;
+	return 1;
+}
+
+static int stack_compact_locked(struct reftable_stack *st, int first, int last,
+				struct strbuf *temp_tab,
+				struct reftable_log_expiry_config *config)
+{
+	struct strbuf next_name = STRBUF_INIT;
+	int tab_fd = -1;
+	struct reftable_writer *wr = NULL;
+	int err = 0;
+
+	format_name(&next_name,
+		    reftable_reader_min_update_index(st->readers[first]),
+		    reftable_reader_max_update_index(st->readers[last]));
+
+	stack_filename(temp_tab, st, next_name.buf);
+	strbuf_addstr(temp_tab, ".temp.XXXXXX");
+
+	tab_fd = mkstemp(temp_tab->buf);
+	wr = reftable_new_writer(reftable_fd_write, &tab_fd, &st->config);
+
+	err = stack_write_compact(st, wr, first, last, config);
+	if (err < 0)
+		goto done;
+	err = reftable_writer_close(wr);
+	if (err < 0)
+		goto done;
+
+	err = close(tab_fd);
+	tab_fd = 0;
+
+done:
+	reftable_writer_free(wr);
+	if (tab_fd > 0) {
+		close(tab_fd);
+		tab_fd = 0;
+	}
+	if (err != 0 && temp_tab->len > 0) {
+		unlink(temp_tab->buf);
+		strbuf_release(temp_tab);
+	}
+	strbuf_release(&next_name);
+	return err;
+}
+
+static int stack_write_compact(struct reftable_stack *st,
+			       struct reftable_writer *wr, int first, int last,
+			       struct reftable_log_expiry_config *config)
+{
+	int subtabs_len = last - first + 1;
+	struct reftable_table *subtabs = reftable_calloc(
+		sizeof(struct reftable_table) * (last - first + 1));
+	struct reftable_merged_table *mt = NULL;
+	int err = 0;
+	struct reftable_iterator it = { NULL };
+	struct reftable_ref_record ref = { NULL };
+	struct reftable_log_record log = { NULL };
+
+	uint64_t entries = 0;
+
+	int i = 0, j = 0;
+	for (i = first, j = 0; i <= last; i++) {
+		struct reftable_reader *t = st->readers[i];
+		reftable_table_from_reader(&subtabs[j++], t);
+		st->stats.bytes += t->size;
+	}
+	reftable_writer_set_limits(wr, st->readers[first]->min_update_index,
+				   st->readers[last]->max_update_index);
+
+	err = reftable_new_merged_table(&mt, subtabs, subtabs_len,
+					st->config.hash_id);
+	if (err < 0) {
+		reftable_free(subtabs);
+		goto done;
+	}
+
+	err = reftable_merged_table_seek_ref(mt, &it, "");
+	if (err < 0)
+		goto done;
+
+	while (1) {
+		err = reftable_iterator_next_ref(&it, &ref);
+		if (err > 0) {
+			err = 0;
+			break;
+		}
+		if (err < 0) {
+			break;
+		}
+
+		if (first == 0 && reftable_ref_record_is_deletion(&ref)) {
+			continue;
+		}
+
+		err = reftable_writer_add_ref(wr, &ref);
+		if (err < 0) {
+			break;
+		}
+		entries++;
+	}
+	reftable_iterator_destroy(&it);
+
+	err = reftable_merged_table_seek_log(mt, &it, "");
+	if (err < 0)
+		goto done;
+
+	while (1) {
+		err = reftable_iterator_next_log(&it, &log);
+		if (err > 0) {
+			err = 0;
+			break;
+		}
+		if (err < 0) {
+			break;
+		}
+		if (first == 0 && reftable_log_record_is_deletion(&log)) {
+			continue;
+		}
+
+		if (config && config->min_update_index > 0 &&
+		    log.update_index < config->min_update_index) {
+			continue;
+		}
+
+		if (config && config->time > 0 &&
+		    log.value.update.time < config->time) {
+			continue;
+		}
+
+		err = reftable_writer_add_log(wr, &log);
+		if (err < 0) {
+			break;
+		}
+		entries++;
+	}
+
+done:
+	reftable_iterator_destroy(&it);
+	if (mt) {
+		merged_table_release(mt);
+		reftable_merged_table_free(mt);
+	}
+	reftable_ref_record_release(&ref);
+	reftable_log_record_release(&log);
+	st->stats.entries_written += entries;
+	return err;
+}
+
+/* <  0: error. 0 == OK, > 0 attempt failed; could retry. */
+static int stack_compact_range(struct reftable_stack *st, int first, int last,
+			       struct reftable_log_expiry_config *expiry)
+{
+	struct strbuf temp_tab_file_name = STRBUF_INIT;
+	struct strbuf new_table_name = STRBUF_INIT;
+	struct strbuf lock_file_name = STRBUF_INIT;
+	struct strbuf ref_list_contents = STRBUF_INIT;
+	struct strbuf new_table_path = STRBUF_INIT;
+	int err = 0;
+	int have_lock = 0;
+	int lock_file_fd = 0;
+	int compact_count = last - first + 1;
+	char **listp = NULL;
+	char **delete_on_success =
+		reftable_calloc(sizeof(char *) * (compact_count + 1));
+	char **subtable_locks =
+		reftable_calloc(sizeof(char *) * (compact_count + 1));
+	int i = 0;
+	int j = 0;
+	int is_empty_table = 0;
+
+	if (first > last || (!expiry && first == last)) {
+		err = 0;
+		goto done;
+	}
+
+	st->stats.attempts++;
+
+	strbuf_reset(&lock_file_name);
+	strbuf_addstr(&lock_file_name, st->list_file);
+	strbuf_addstr(&lock_file_name, ".lock");
+
+	lock_file_fd =
+		open(lock_file_name.buf, O_EXCL | O_CREAT | O_WRONLY, 0644);
+	if (lock_file_fd < 0) {
+		if (errno == EEXIST) {
+			err = 1;
+		} else {
+			err = REFTABLE_IO_ERROR;
+		}
+		goto done;
+	}
+	/* Don't want to write to the lock for now.  */
+	close(lock_file_fd);
+	lock_file_fd = 0;
+
+	have_lock = 1;
+	err = stack_uptodate(st);
+	if (err != 0)
+		goto done;
+
+	for (i = first, j = 0; i <= last; i++) {
+		struct strbuf subtab_file_name = STRBUF_INIT;
+		struct strbuf subtab_lock = STRBUF_INIT;
+		int sublock_file_fd = -1;
+
+		stack_filename(&subtab_file_name, st,
+			       reader_name(st->readers[i]));
+
+		strbuf_reset(&subtab_lock);
+		strbuf_addbuf(&subtab_lock, &subtab_file_name);
+		strbuf_addstr(&subtab_lock, ".lock");
+
+		sublock_file_fd = open(subtab_lock.buf,
+				       O_EXCL | O_CREAT | O_WRONLY, 0644);
+		if (sublock_file_fd > 0) {
+			close(sublock_file_fd);
+		} else if (sublock_file_fd < 0) {
+			if (errno == EEXIST) {
+				err = 1;
+			} else {
+				err = REFTABLE_IO_ERROR;
+			}
+		}
+
+		subtable_locks[j] = subtab_lock.buf;
+		delete_on_success[j] = subtab_file_name.buf;
+		j++;
+
+		if (err != 0)
+			goto done;
+	}
+
+	err = unlink(lock_file_name.buf);
+	if (err < 0)
+		goto done;
+	have_lock = 0;
+
+	err = stack_compact_locked(st, first, last, &temp_tab_file_name,
+				   expiry);
+	/* Compaction + tombstones can create an empty table out of non-empty
+	 * tables. */
+	is_empty_table = (err == REFTABLE_EMPTY_TABLE_ERROR);
+	if (is_empty_table) {
+		err = 0;
+	}
+	if (err < 0)
+		goto done;
+
+	lock_file_fd =
+		open(lock_file_name.buf, O_EXCL | O_CREAT | O_WRONLY, 0644);
+	if (lock_file_fd < 0) {
+		if (errno == EEXIST) {
+			err = 1;
+		} else {
+			err = REFTABLE_IO_ERROR;
+		}
+		goto done;
+	}
+	have_lock = 1;
+
+	format_name(&new_table_name, st->readers[first]->min_update_index,
+		    st->readers[last]->max_update_index);
+	strbuf_addstr(&new_table_name, ".ref");
+
+	stack_filename(&new_table_path, st, new_table_name.buf);
+
+	if (!is_empty_table) {
+		/* retry? */
+		err = rename(temp_tab_file_name.buf, new_table_path.buf);
+		if (err < 0) {
+			err = REFTABLE_IO_ERROR;
+			goto done;
+		}
+	}
+
+	for (i = 0; i < first; i++) {
+		strbuf_addstr(&ref_list_contents, st->readers[i]->name);
+		strbuf_addstr(&ref_list_contents, "\n");
+	}
+	if (!is_empty_table) {
+		strbuf_addbuf(&ref_list_contents, &new_table_name);
+		strbuf_addstr(&ref_list_contents, "\n");
+	}
+	for (i = last + 1; i < st->merged->stack_len; i++) {
+		strbuf_addstr(&ref_list_contents, st->readers[i]->name);
+		strbuf_addstr(&ref_list_contents, "\n");
+	}
+
+	err = write(lock_file_fd, ref_list_contents.buf, ref_list_contents.len);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		unlink(new_table_path.buf);
+		goto done;
+	}
+	err = close(lock_file_fd);
+	lock_file_fd = 0;
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		unlink(new_table_path.buf);
+		goto done;
+	}
+
+	err = rename(lock_file_name.buf, st->list_file);
+	if (err < 0) {
+		err = REFTABLE_IO_ERROR;
+		unlink(new_table_path.buf);
+		goto done;
+	}
+	have_lock = 0;
+
+	/* Reload the stack before deleting. On windows, we can only delete the
+	   files after we closed them.
+	*/
+	err = reftable_stack_reload_maybe_reuse(st, first < last);
+
+	listp = delete_on_success;
+	while (*listp) {
+		if (strcmp(*listp, new_table_path.buf)) {
+			unlink(*listp);
+		}
+		listp++;
+	}
+
+done:
+	free_names(delete_on_success);
+
+	listp = subtable_locks;
+	while (*listp) {
+		unlink(*listp);
+		listp++;
+	}
+	free_names(subtable_locks);
+	if (lock_file_fd > 0) {
+		close(lock_file_fd);
+		lock_file_fd = 0;
+	}
+	if (have_lock) {
+		unlink(lock_file_name.buf);
+	}
+	strbuf_release(&new_table_name);
+	strbuf_release(&new_table_path);
+	strbuf_release(&ref_list_contents);
+	strbuf_release(&temp_tab_file_name);
+	strbuf_release(&lock_file_name);
+	return err;
+}
+
+int reftable_stack_compact_all(struct reftable_stack *st,
+			       struct reftable_log_expiry_config *config)
+{
+	return stack_compact_range(st, 0, st->merged->stack_len - 1, config);
+}
+
+static int stack_compact_range_stats(struct reftable_stack *st, int first,
+				     int last,
+				     struct reftable_log_expiry_config *config)
+{
+	int err = stack_compact_range(st, first, last, config);
+	if (err > 0) {
+		st->stats.failures++;
+	}
+	return err;
+}
+
+static int segment_size(struct segment *s)
+{
+	return s->end - s->start;
+}
+
+int fastlog2(uint64_t sz)
+{
+	int l = 0;
+	if (sz == 0)
+		return 0;
+	for (; sz; sz /= 2) {
+		l++;
+	}
+	return l - 1;
+}
+
+struct segment *sizes_to_segments(int *seglen, uint64_t *sizes, int n)
+{
+	struct segment *segs = reftable_calloc(sizeof(struct segment) * n);
+	int next = 0;
+	struct segment cur = { 0 };
+	int i = 0;
+
+	if (n == 0) {
+		*seglen = 0;
+		return segs;
+	}
+	for (i = 0; i < n; i++) {
+		int log = fastlog2(sizes[i]);
+		if (cur.log != log && cur.bytes > 0) {
+			struct segment fresh = {
+				.start = i,
+			};
+
+			segs[next++] = cur;
+			cur = fresh;
+		}
+
+		cur.log = log;
+		cur.end = i + 1;
+		cur.bytes += sizes[i];
+	}
+	segs[next++] = cur;
+	*seglen = next;
+	return segs;
+}
+
+struct segment suggest_compaction_segment(uint64_t *sizes, int n)
+{
+	int seglen = 0;
+	struct segment *segs = sizes_to_segments(&seglen, sizes, n);
+	struct segment min_seg = {
+		.log = 64,
+	};
+	int i = 0;
+	for (i = 0; i < seglen; i++) {
+		if (segment_size(&segs[i]) == 1) {
+			continue;
+		}
+
+		if (segs[i].log < min_seg.log) {
+			min_seg = segs[i];
+		}
+	}
+
+	while (min_seg.start > 0) {
+		int prev = min_seg.start - 1;
+		if (fastlog2(min_seg.bytes) < fastlog2(sizes[prev])) {
+			break;
+		}
+
+		min_seg.start = prev;
+		min_seg.bytes += sizes[prev];
+	}
+
+	reftable_free(segs);
+	return min_seg;
+}
+
+static uint64_t *stack_table_sizes_for_compaction(struct reftable_stack *st)
+{
+	uint64_t *sizes =
+		reftable_calloc(sizeof(uint64_t) * st->merged->stack_len);
+	int version = (st->config.hash_id == GIT_SHA1_FORMAT_ID) ? 1 : 2;
+	int overhead = header_size(version) - 1;
+	int i = 0;
+	for (i = 0; i < st->merged->stack_len; i++) {
+		sizes[i] = st->readers[i]->size - overhead;
+	}
+	return sizes;
+}
+
+int reftable_stack_auto_compact(struct reftable_stack *st)
+{
+	uint64_t *sizes = stack_table_sizes_for_compaction(st);
+	struct segment seg =
+		suggest_compaction_segment(sizes, st->merged->stack_len);
+	reftable_free(sizes);
+	if (segment_size(&seg) > 0)
+		return stack_compact_range_stats(st, seg.start, seg.end - 1,
+						 NULL);
+
+	return 0;
+}
+
+struct reftable_compaction_stats *
+reftable_stack_compaction_stats(struct reftable_stack *st)
+{
+	return &st->stats;
+}
+
+int reftable_stack_read_ref(struct reftable_stack *st, const char *refname,
+			    struct reftable_ref_record *ref)
+{
+	struct reftable_table tab = { NULL };
+	reftable_table_from_merged_table(&tab, reftable_stack_merged_table(st));
+	return reftable_table_read_ref(&tab, refname, ref);
+}
+
+int reftable_stack_read_log(struct reftable_stack *st, const char *refname,
+			    struct reftable_log_record *log)
+{
+	struct reftable_iterator it = { NULL };
+	struct reftable_merged_table *mt = reftable_stack_merged_table(st);
+	int err = reftable_merged_table_seek_log(mt, &it, refname);
+	if (err)
+		goto done;
+
+	err = reftable_iterator_next_log(&it, log);
+	if (err)
+		goto done;
+
+	if (strcmp(log->refname, refname) ||
+	    reftable_log_record_is_deletion(log)) {
+		err = 1;
+		goto done;
+	}
+
+done:
+	if (err) {
+		reftable_log_record_release(log);
+	}
+	reftable_iterator_destroy(&it);
+	return err;
+}
+
+static int stack_check_addition(struct reftable_stack *st,
+				const char *new_tab_name)
+{
+	int err = 0;
+	struct reftable_block_source src = { NULL };
+	struct reftable_reader *rd = NULL;
+	struct reftable_table tab = { NULL };
+	struct reftable_ref_record *refs = NULL;
+	struct reftable_iterator it = { NULL };
+	int cap = 0;
+	int len = 0;
+	int i = 0;
+
+	if (st->config.skip_name_check)
+		return 0;
+
+	err = reftable_block_source_from_file(&src, new_tab_name);
+	if (err < 0)
+		goto done;
+
+	err = reftable_new_reader(&rd, &src, new_tab_name);
+	if (err < 0)
+		goto done;
+
+	err = reftable_reader_seek_ref(rd, &it, "");
+	if (err > 0) {
+		err = 0;
+		goto done;
+	}
+	if (err < 0)
+		goto done;
+
+	while (1) {
+		struct reftable_ref_record ref = { NULL };
+		err = reftable_iterator_next_ref(&it, &ref);
+		if (err > 0) {
+			break;
+		}
+		if (err < 0)
+			goto done;
+
+		if (len >= cap) {
+			cap = 2 * cap + 1;
+			refs = reftable_realloc(refs, cap * sizeof(refs[0]));
+		}
+
+		refs[len++] = ref;
+	}
+
+	reftable_table_from_merged_table(&tab, reftable_stack_merged_table(st));
+
+	err = validate_ref_record_addition(tab, refs, len);
+
+done:
+	for (i = 0; i < len; i++) {
+		reftable_ref_record_release(&refs[i]);
+	}
+
+	free(refs);
+	reftable_iterator_destroy(&it);
+	reftable_reader_free(rd);
+	return err;
+}
+
+static int is_table_name(const char *s)
+{
+	const char *dot = strrchr(s, '.');
+	return dot && !strcmp(dot, ".ref");
+}
+
+static void remove_maybe_stale_table(struct reftable_stack *st, uint64_t max,
+				     const char *name)
+{
+	int err = 0;
+	uint64_t update_idx = 0;
+	struct reftable_block_source src = { NULL };
+	struct reftable_reader *rd = NULL;
+	struct strbuf table_path = STRBUF_INIT;
+	stack_filename(&table_path, st, name);
+
+	err = reftable_block_source_from_file(&src, table_path.buf);
+	if (err < 0)
+		goto done;
+
+	err = reftable_new_reader(&rd, &src, name);
+	if (err < 0)
+		goto done;
+
+	update_idx = reftable_reader_max_update_index(rd);
+	reftable_reader_free(rd);
+
+	if (update_idx <= max) {
+		unlink(table_path.buf);
+	}
+done:
+	strbuf_release(&table_path);
+}
+
+static int reftable_stack_clean_locked(struct reftable_stack *st)
+{
+	uint64_t max = reftable_merged_table_max_update_index(
+		reftable_stack_merged_table(st));
+	DIR *dir = opendir(st->reftable_dir);
+	struct dirent *d = NULL;
+	if (!dir) {
+		return REFTABLE_IO_ERROR;
+	}
+
+	while ((d = readdir(dir))) {
+		int i = 0;
+		int found = 0;
+		if (!is_table_name(d->d_name))
+			continue;
+
+		for (i = 0; !found && i < st->readers_len; i++) {
+			found = !strcmp(reader_name(st->readers[i]), d->d_name);
+		}
+		if (found)
+			continue;
+
+		remove_maybe_stale_table(st, max, d->d_name);
+	}
+
+	closedir(dir);
+	return 0;
+}
+
+int reftable_stack_clean(struct reftable_stack *st)
+{
+	struct reftable_addition *add = NULL;
+	int err = reftable_stack_new_addition(&add, st);
+	if (err < 0) {
+		goto done;
+	}
+
+	err = reftable_stack_reload(st);
+	if (err < 0) {
+		goto done;
+	}
+
+	err = reftable_stack_clean_locked(st);
+
+done:
+	reftable_addition_destroy(add);
+	return err;
+}
+
+int reftable_stack_print_directory(const char *stackdir, uint32_t hash_id)
+{
+	struct reftable_stack *stack = NULL;
+	struct reftable_write_options cfg = { .hash_id = hash_id };
+	struct reftable_merged_table *merged = NULL;
+	struct reftable_table table = { NULL };
+
+	int err = reftable_new_stack(&stack, stackdir, cfg);
+	if (err < 0)
+		goto done;
+
+	merged = reftable_stack_merged_table(stack);
+	reftable_table_from_merged_table(&table, merged);
+	err = reftable_table_print(&table);
+done:
+	if (stack)
+		reftable_stack_destroy(stack);
+	return err;
+}
diff --git a/reftable/stack.h b/reftable/stack.h
new file mode 100644
index 0000000..f570058
--- /dev/null
+++ b/reftable/stack.h
@@ -0,0 +1,41 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef STACK_H
+#define STACK_H
+
+#include "system.h"
+#include "reftable-writer.h"
+#include "reftable-stack.h"
+
+struct reftable_stack {
+	char *list_file;
+	char *reftable_dir;
+	int disable_auto_compact;
+
+	struct reftable_write_options config;
+
+	struct reftable_reader **readers;
+	size_t readers_len;
+	struct reftable_merged_table *merged;
+	struct reftable_compaction_stats stats;
+};
+
+int read_lines(const char *filename, char ***lines);
+
+struct segment {
+	int start, end;
+	int log;
+	uint64_t bytes;
+};
+
+int fastlog2(uint64_t sz);
+struct segment *sizes_to_segments(int *seglen, uint64_t *sizes, int n);
+struct segment suggest_compaction_segment(uint64_t *sizes, int n);
+
+#endif
diff --git a/reftable/stack_test.c b/reftable/stack_test.c
new file mode 100644
index 0000000..eb0b722
--- /dev/null
+++ b/reftable/stack_test.c
@@ -0,0 +1,953 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "stack.h"
+
+#include "system.h"
+
+#include "reftable-reader.h"
+#include "merged.h"
+#include "basics.h"
+#include "constants.h"
+#include "record.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+#include <sys/types.h>
+#include <dirent.h>
+
+static void clear_dir(const char *dirname)
+{
+	struct strbuf path = STRBUF_INIT;
+	strbuf_addstr(&path, dirname);
+	remove_dir_recursively(&path, 0);
+	strbuf_release(&path);
+}
+
+static int count_dir_entries(const char *dirname)
+{
+	DIR *dir = opendir(dirname);
+	int len = 0;
+	struct dirent *d;
+	if (dir == NULL)
+		return 0;
+
+	while ((d = readdir(dir))) {
+		if (!strcmp(d->d_name, "..") || !strcmp(d->d_name, "."))
+			continue;
+		len++;
+	}
+	closedir(dir);
+	return len;
+}
+
+/*
+ * Work linenumber into the tempdir, so we can see which tests forget to
+ * cleanup.
+ */
+static char *get_tmp_template(int linenumber)
+{
+	const char *tmp = getenv("TMPDIR");
+	static char template[1024];
+	snprintf(template, sizeof(template) - 1, "%s/stack_test-%d.XXXXXX",
+		 tmp ? tmp : "/tmp", linenumber);
+	return template;
+}
+
+static char *get_tmp_dir(int linenumber)
+{
+	char *dir = get_tmp_template(linenumber);
+	EXPECT(mkdtemp(dir));
+	return dir;
+}
+
+static void test_read_file(void)
+{
+	char *fn = get_tmp_template(__LINE__);
+	int fd = mkstemp(fn);
+	char out[1024] = "line1\n\nline2\nline3";
+	int n, err;
+	char **names = NULL;
+	char *want[] = { "line1", "line2", "line3" };
+	int i = 0;
+
+	EXPECT(fd > 0);
+	n = write(fd, out, strlen(out));
+	EXPECT(n == strlen(out));
+	err = close(fd);
+	EXPECT(err >= 0);
+
+	err = read_lines(fn, &names);
+	EXPECT_ERR(err);
+
+	for (i = 0; names[i]; i++) {
+		EXPECT(0 == strcmp(want[i], names[i]));
+	}
+	free_names(names);
+	remove(fn);
+}
+
+static void test_parse_names(void)
+{
+	char buf[] = "line\n";
+	char **names = NULL;
+	parse_names(buf, strlen(buf), &names);
+
+	EXPECT(NULL != names[0]);
+	EXPECT(0 == strcmp(names[0], "line"));
+	EXPECT(NULL == names[1]);
+	free_names(names);
+}
+
+static void test_names_equal(void)
+{
+	char *a[] = { "a", "b", "c", NULL };
+	char *b[] = { "a", "b", "d", NULL };
+	char *c[] = { "a", "b", NULL };
+
+	EXPECT(names_equal(a, a));
+	EXPECT(!names_equal(a, b));
+	EXPECT(!names_equal(a, c));
+}
+
+static int write_test_ref(struct reftable_writer *wr, void *arg)
+{
+	struct reftable_ref_record *ref = arg;
+	reftable_writer_set_limits(wr, ref->update_index, ref->update_index);
+	return reftable_writer_add_ref(wr, ref);
+}
+
+struct write_log_arg {
+	struct reftable_log_record *log;
+	uint64_t update_index;
+};
+
+static int write_test_log(struct reftable_writer *wr, void *arg)
+{
+	struct write_log_arg *wla = arg;
+
+	reftable_writer_set_limits(wr, wla->update_index, wla->update_index);
+	return reftable_writer_add_log(wr, wla->log);
+}
+
+static void test_reftable_stack_add_one(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	struct reftable_ref_record ref = {
+		.refname = "HEAD",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+	struct reftable_ref_record dest = { NULL };
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_test_ref, &ref);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_ref(st, ref.refname, &dest);
+	EXPECT_ERR(err);
+	EXPECT(0 == strcmp("master", dest.value.symref));
+
+	printf("testing print functionality:\n");
+	err = reftable_stack_print_directory(dir, GIT_SHA1_FORMAT_ID);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_print_directory(dir, GIT_SHA256_FORMAT_ID);
+	EXPECT(err == REFTABLE_FORMAT_ERROR);
+
+	reftable_ref_record_release(&dest);
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_uptodate(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st1 = NULL;
+	struct reftable_stack *st2 = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	int err;
+	struct reftable_ref_record ref1 = {
+		.refname = "HEAD",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+	struct reftable_ref_record ref2 = {
+		.refname = "branch2",
+		.update_index = 2,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+
+
+	/* simulate multi-process access to the same stack
+	   by creating two stacks for the same directory.
+	 */
+	err = reftable_new_stack(&st1, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_new_stack(&st2, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st1, &write_test_ref, &ref1);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st2, &write_test_ref, &ref2);
+	EXPECT(err == REFTABLE_LOCK_ERROR);
+
+	err = reftable_stack_reload(st2);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st2, &write_test_ref, &ref2);
+	EXPECT_ERR(err);
+	reftable_stack_destroy(st1);
+	reftable_stack_destroy(st2);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_transaction_api(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	struct reftable_addition *add = NULL;
+
+	struct reftable_ref_record ref = {
+		.refname = "HEAD",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+	struct reftable_ref_record dest = { NULL };
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	reftable_addition_destroy(add);
+
+	err = reftable_stack_new_addition(&add, st);
+	EXPECT_ERR(err);
+
+	err = reftable_addition_add(add, &write_test_ref, &ref);
+	EXPECT_ERR(err);
+
+	err = reftable_addition_commit(add);
+	EXPECT_ERR(err);
+
+	reftable_addition_destroy(add);
+
+	err = reftable_stack_read_ref(st, ref.refname, &dest);
+	EXPECT_ERR(err);
+	EXPECT(REFTABLE_REF_SYMREF == dest.value_type);
+	EXPECT(0 == strcmp("master", dest.value.symref));
+
+	reftable_ref_record_release(&dest);
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_validate_refname(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	char *dir = get_tmp_dir(__LINE__);
+
+	int i;
+	struct reftable_ref_record ref = {
+		.refname = "a/b",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+	char *additions[] = { "a", "a/b/c" };
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_test_ref, &ref);
+	EXPECT_ERR(err);
+
+	for (i = 0; i < ARRAY_SIZE(additions); i++) {
+		struct reftable_ref_record ref = {
+			.refname = additions[i],
+			.update_index = 1,
+			.value_type = REFTABLE_REF_SYMREF,
+			.value.symref = "master",
+		};
+
+		err = reftable_stack_add(st, &write_test_ref, &ref);
+		EXPECT(err == REFTABLE_NAME_CONFLICT);
+	}
+
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static int write_error(struct reftable_writer *wr, void *arg)
+{
+	return *((int *)arg);
+}
+
+static void test_reftable_stack_update_index_check(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	struct reftable_ref_record ref1 = {
+		.refname = "name1",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+	struct reftable_ref_record ref2 = {
+		.refname = "name2",
+		.update_index = 1,
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "master",
+	};
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_test_ref, &ref1);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_test_ref, &ref2);
+	EXPECT(err == REFTABLE_API_ERROR);
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_lock_failure(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err, i;
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+	for (i = -1; i != REFTABLE_EMPTY_TABLE_ERROR; i--) {
+		err = reftable_stack_add(st, &write_error, &i);
+		EXPECT(err == i);
+	}
+
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_add(void)
+{
+	int i = 0;
+	int err = 0;
+	struct reftable_write_options cfg = {
+		.exact_log_message = 1,
+	};
+	struct reftable_stack *st = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_ref_record refs[2] = { { NULL } };
+	struct reftable_log_record logs[2] = { { NULL } };
+	int N = ARRAY_SIZE(refs);
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+	st->disable_auto_compact = 1;
+
+	for (i = 0; i < N; i++) {
+		char buf[256];
+		snprintf(buf, sizeof(buf), "branch%02d", i);
+		refs[i].refname = xstrdup(buf);
+		refs[i].update_index = i + 1;
+		refs[i].value_type = REFTABLE_REF_VAL1;
+		refs[i].value.val1 = reftable_malloc(GIT_SHA1_RAWSZ);
+		set_test_hash(refs[i].value.val1, i);
+
+		logs[i].refname = xstrdup(buf);
+		logs[i].update_index = N + i + 1;
+		logs[i].value_type = REFTABLE_LOG_UPDATE;
+
+		logs[i].value.update.new_hash = reftable_malloc(GIT_SHA1_RAWSZ);
+		logs[i].value.update.email = xstrdup("identity@invalid");
+		set_test_hash(logs[i].value.update.new_hash, i);
+	}
+
+	for (i = 0; i < N; i++) {
+		int err = reftable_stack_add(st, &write_test_ref, &refs[i]);
+		EXPECT_ERR(err);
+	}
+
+	for (i = 0; i < N; i++) {
+		struct write_log_arg arg = {
+			.log = &logs[i],
+			.update_index = reftable_stack_next_update_index(st),
+		};
+		int err = reftable_stack_add(st, &write_test_log, &arg);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_stack_compact_all(st, NULL);
+	EXPECT_ERR(err);
+
+	for (i = 0; i < N; i++) {
+		struct reftable_ref_record dest = { NULL };
+
+		int err = reftable_stack_read_ref(st, refs[i].refname, &dest);
+		EXPECT_ERR(err);
+		EXPECT(reftable_ref_record_equal(&dest, refs + i,
+						 GIT_SHA1_RAWSZ));
+		reftable_ref_record_release(&dest);
+	}
+
+	for (i = 0; i < N; i++) {
+		struct reftable_log_record dest = { NULL };
+		int err = reftable_stack_read_log(st, refs[i].refname, &dest);
+		EXPECT_ERR(err);
+		EXPECT(reftable_log_record_equal(&dest, logs + i,
+						 GIT_SHA1_RAWSZ));
+		reftable_log_record_release(&dest);
+	}
+
+	/* cleanup */
+	reftable_stack_destroy(st);
+	for (i = 0; i < N; i++) {
+		reftable_ref_record_release(&refs[i]);
+		reftable_log_record_release(&logs[i]);
+	}
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_log_normalize(void)
+{
+	int err = 0;
+	struct reftable_write_options cfg = {
+		0,
+	};
+	struct reftable_stack *st = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	uint8_t h1[GIT_SHA1_RAWSZ] = { 0x01 }, h2[GIT_SHA1_RAWSZ] = { 0x02 };
+
+	struct reftable_log_record input = { .refname = "branch",
+					     .update_index = 1,
+					     .value_type = REFTABLE_LOG_UPDATE,
+					     .value = { .update = {
+								.new_hash = h1,
+								.old_hash = h2,
+							} } };
+	struct reftable_log_record dest = {
+		.update_index = 0,
+	};
+	struct write_log_arg arg = {
+		.log = &input,
+		.update_index = 1,
+	};
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	input.value.update.message = "one\ntwo";
+	err = reftable_stack_add(st, &write_test_log, &arg);
+	EXPECT(err == REFTABLE_API_ERROR);
+
+	input.value.update.message = "one";
+	err = reftable_stack_add(st, &write_test_log, &arg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_log(st, input.refname, &dest);
+	EXPECT_ERR(err);
+	EXPECT(0 == strcmp(dest.value.update.message, "one\n"));
+
+	input.value.update.message = "two\n";
+	arg.update_index = 2;
+	err = reftable_stack_add(st, &write_test_log, &arg);
+	EXPECT_ERR(err);
+	err = reftable_stack_read_log(st, input.refname, &dest);
+	EXPECT_ERR(err);
+	EXPECT(0 == strcmp(dest.value.update.message, "two\n"));
+
+	/* cleanup */
+	reftable_stack_destroy(st);
+	reftable_log_record_release(&dest);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_tombstone(void)
+{
+	int i = 0;
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	struct reftable_ref_record refs[2] = { { NULL } };
+	struct reftable_log_record logs[2] = { { NULL } };
+	int N = ARRAY_SIZE(refs);
+	struct reftable_ref_record dest = { NULL };
+	struct reftable_log_record log_dest = { NULL };
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	/* even entries add the refs, odd entries delete them. */
+	for (i = 0; i < N; i++) {
+		const char *buf = "branch";
+		refs[i].refname = xstrdup(buf);
+		refs[i].update_index = i + 1;
+		if (i % 2 == 0) {
+			refs[i].value_type = REFTABLE_REF_VAL1;
+			refs[i].value.val1 = reftable_malloc(GIT_SHA1_RAWSZ);
+			set_test_hash(refs[i].value.val1, i);
+		}
+
+		logs[i].refname = xstrdup(buf);
+		/* update_index is part of the key. */
+		logs[i].update_index = 42;
+		if (i % 2 == 0) {
+			logs[i].value_type = REFTABLE_LOG_UPDATE;
+			logs[i].value.update.new_hash =
+				reftable_malloc(GIT_SHA1_RAWSZ);
+			set_test_hash(logs[i].value.update.new_hash, i);
+			logs[i].value.update.email =
+				xstrdup("identity@invalid");
+		}
+	}
+	for (i = 0; i < N; i++) {
+		int err = reftable_stack_add(st, &write_test_ref, &refs[i]);
+		EXPECT_ERR(err);
+	}
+
+	for (i = 0; i < N; i++) {
+		struct write_log_arg arg = {
+			.log = &logs[i],
+			.update_index = reftable_stack_next_update_index(st),
+		};
+		int err = reftable_stack_add(st, &write_test_log, &arg);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_stack_read_ref(st, "branch", &dest);
+	EXPECT(err == 1);
+	reftable_ref_record_release(&dest);
+
+	err = reftable_stack_read_log(st, "branch", &log_dest);
+	EXPECT(err == 1);
+	reftable_log_record_release(&log_dest);
+
+	err = reftable_stack_compact_all(st, NULL);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_ref(st, "branch", &dest);
+	EXPECT(err == 1);
+
+	err = reftable_stack_read_log(st, "branch", &log_dest);
+	EXPECT(err == 1);
+	reftable_ref_record_release(&dest);
+	reftable_log_record_release(&log_dest);
+
+	/* cleanup */
+	reftable_stack_destroy(st);
+	for (i = 0; i < N; i++) {
+		reftable_ref_record_release(&refs[i]);
+		reftable_log_record_release(&logs[i]);
+	}
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_hash_id(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+
+	struct reftable_ref_record ref = {
+		.refname = "master",
+		.value_type = REFTABLE_REF_SYMREF,
+		.value.symref = "target",
+		.update_index = 1,
+	};
+	struct reftable_write_options cfg32 = { .hash_id = GIT_SHA256_FORMAT_ID };
+	struct reftable_stack *st32 = NULL;
+	struct reftable_write_options cfg_default = { 0 };
+	struct reftable_stack *st_default = NULL;
+	struct reftable_ref_record dest = { NULL };
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_test_ref, &ref);
+	EXPECT_ERR(err);
+
+	/* can't read it with the wrong hash ID. */
+	err = reftable_new_stack(&st32, dir, cfg32);
+	EXPECT(err == REFTABLE_FORMAT_ERROR);
+
+	/* check that we can read it back with default config too. */
+	err = reftable_new_stack(&st_default, dir, cfg_default);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_ref(st_default, "master", &dest);
+	EXPECT_ERR(err);
+
+	EXPECT(reftable_ref_record_equal(&ref, &dest, GIT_SHA1_RAWSZ));
+	reftable_ref_record_release(&dest);
+	reftable_stack_destroy(st);
+	reftable_stack_destroy(st_default);
+	clear_dir(dir);
+}
+
+static void test_log2(void)
+{
+	EXPECT(1 == fastlog2(3));
+	EXPECT(2 == fastlog2(4));
+	EXPECT(2 == fastlog2(5));
+}
+
+static void test_sizes_to_segments(void)
+{
+	uint64_t sizes[] = { 2, 3, 4, 5, 7, 9 };
+	/* .................0  1  2  3  4  5 */
+
+	int seglen = 0;
+	struct segment *segs =
+		sizes_to_segments(&seglen, sizes, ARRAY_SIZE(sizes));
+	EXPECT(segs[2].log == 3);
+	EXPECT(segs[2].start == 5);
+	EXPECT(segs[2].end == 6);
+
+	EXPECT(segs[1].log == 2);
+	EXPECT(segs[1].start == 2);
+	EXPECT(segs[1].end == 5);
+	reftable_free(segs);
+}
+
+static void test_sizes_to_segments_empty(void)
+{
+	int seglen = 0;
+	struct segment *segs = sizes_to_segments(&seglen, NULL, 0);
+	EXPECT(seglen == 0);
+	reftable_free(segs);
+}
+
+static void test_sizes_to_segments_all_equal(void)
+{
+	uint64_t sizes[] = { 5, 5 };
+
+	int seglen = 0;
+	struct segment *segs =
+		sizes_to_segments(&seglen, sizes, ARRAY_SIZE(sizes));
+	EXPECT(seglen == 1);
+	EXPECT(segs[0].start == 0);
+	EXPECT(segs[0].end == 2);
+	reftable_free(segs);
+}
+
+static void test_suggest_compaction_segment(void)
+{
+	uint64_t sizes[] = { 128, 64, 17, 16, 9, 9, 9, 16, 16 };
+	/* .................0    1    2  3   4  5  6 */
+	struct segment min =
+		suggest_compaction_segment(sizes, ARRAY_SIZE(sizes));
+	EXPECT(min.start == 2);
+	EXPECT(min.end == 7);
+}
+
+static void test_suggest_compaction_segment_nothing(void)
+{
+	uint64_t sizes[] = { 64, 32, 16, 8, 4, 2 };
+	struct segment result =
+		suggest_compaction_segment(sizes, ARRAY_SIZE(sizes));
+	EXPECT(result.start == result.end);
+}
+
+static void test_reflog_expire(void)
+{
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	struct reftable_log_record logs[20] = { { NULL } };
+	int N = ARRAY_SIZE(logs) - 1;
+	int i = 0;
+	int err;
+	struct reftable_log_expiry_config expiry = {
+		.time = 10,
+	};
+	struct reftable_log_record log = { NULL };
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	for (i = 1; i <= N; i++) {
+		char buf[256];
+		snprintf(buf, sizeof(buf), "branch%02d", i);
+
+		logs[i].refname = xstrdup(buf);
+		logs[i].update_index = i;
+		logs[i].value_type = REFTABLE_LOG_UPDATE;
+		logs[i].value.update.time = i;
+		logs[i].value.update.new_hash = reftable_malloc(GIT_SHA1_RAWSZ);
+		logs[i].value.update.email = xstrdup("identity@invalid");
+		set_test_hash(logs[i].value.update.new_hash, i);
+	}
+
+	for (i = 1; i <= N; i++) {
+		struct write_log_arg arg = {
+			.log = &logs[i],
+			.update_index = reftable_stack_next_update_index(st),
+		};
+		int err = reftable_stack_add(st, &write_test_log, &arg);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_stack_compact_all(st, NULL);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_compact_all(st, &expiry);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_log(st, logs[9].refname, &log);
+	EXPECT(err == 1);
+
+	err = reftable_stack_read_log(st, logs[11].refname, &log);
+	EXPECT_ERR(err);
+
+	expiry.min_update_index = 15;
+	err = reftable_stack_compact_all(st, &expiry);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_read_log(st, logs[14].refname, &log);
+	EXPECT(err == 1);
+
+	err = reftable_stack_read_log(st, logs[16].refname, &log);
+	EXPECT_ERR(err);
+
+	/* cleanup */
+	reftable_stack_destroy(st);
+	for (i = 0; i <= N; i++) {
+		reftable_log_record_release(&logs[i]);
+	}
+	clear_dir(dir);
+	reftable_log_record_release(&log);
+}
+
+static int write_nothing(struct reftable_writer *wr, void *arg)
+{
+	reftable_writer_set_limits(wr, 1, 1);
+	return 0;
+}
+
+static void test_empty_add(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	int err;
+	char *dir = get_tmp_dir(__LINE__);
+
+	struct reftable_stack *st2 = NULL;
+
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_add(st, &write_nothing, NULL);
+	EXPECT_ERR(err);
+
+	err = reftable_new_stack(&st2, dir, cfg);
+	EXPECT_ERR(err);
+	clear_dir(dir);
+	reftable_stack_destroy(st);
+	reftable_stack_destroy(st2);
+}
+
+static void test_reftable_stack_auto_compaction(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	int err, i;
+	int N = 100;
+
+	err = reftable_new_stack(&st, dir, cfg);
+	EXPECT_ERR(err);
+
+	st->disable_auto_compact = 1; /* call manually below for coverage. */
+	for (i = 0; i < N; i++) {
+		char name[100];
+		struct reftable_ref_record ref = {
+			.refname = name,
+			.update_index = reftable_stack_next_update_index(st),
+			.value_type = REFTABLE_REF_SYMREF,
+			.value.symref = "master",
+		};
+		snprintf(name, sizeof(name), "branch%04d", i);
+
+		err = reftable_stack_add(st, &write_test_ref, &ref);
+		EXPECT_ERR(err);
+
+		err = reftable_stack_auto_compact(st);
+		EXPECT(i < 3 || st->merged->stack_len < 2 * fastlog2(i));
+	}
+
+	EXPECT(reftable_stack_compaction_stats(st)->entries_written <
+	       (uint64_t)(N * fastlog2(N)));
+
+	reftable_stack_destroy(st);
+	clear_dir(dir);
+}
+
+static void test_reftable_stack_compaction_concurrent(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st1 = NULL, *st2 = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	int err, i;
+	int N = 3;
+
+	err = reftable_new_stack(&st1, dir, cfg);
+	EXPECT_ERR(err);
+
+	for (i = 0; i < N; i++) {
+		char name[100];
+		struct reftable_ref_record ref = {
+			.refname = name,
+			.update_index = reftable_stack_next_update_index(st1),
+			.value_type = REFTABLE_REF_SYMREF,
+			.value.symref = "master",
+		};
+		snprintf(name, sizeof(name), "branch%04d", i);
+
+		err = reftable_stack_add(st1, &write_test_ref, &ref);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_new_stack(&st2, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_compact_all(st1, NULL);
+	EXPECT_ERR(err);
+
+	reftable_stack_destroy(st1);
+	reftable_stack_destroy(st2);
+
+	EXPECT(count_dir_entries(dir) == 2);
+	clear_dir(dir);
+}
+
+static void unclean_stack_close(struct reftable_stack *st)
+{
+	/* break abstraction boundary to simulate unclean shutdown. */
+	int i = 0;
+	for (; i < st->readers_len; i++) {
+		reftable_reader_free(st->readers[i]);
+	}
+	st->readers_len = 0;
+	FREE_AND_NULL(st->readers);
+}
+
+static void test_reftable_stack_compaction_concurrent_clean(void)
+{
+	struct reftable_write_options cfg = { 0 };
+	struct reftable_stack *st1 = NULL, *st2 = NULL, *st3 = NULL;
+	char *dir = get_tmp_dir(__LINE__);
+
+	int err, i;
+	int N = 3;
+
+	err = reftable_new_stack(&st1, dir, cfg);
+	EXPECT_ERR(err);
+
+	for (i = 0; i < N; i++) {
+		char name[100];
+		struct reftable_ref_record ref = {
+			.refname = name,
+			.update_index = reftable_stack_next_update_index(st1),
+			.value_type = REFTABLE_REF_SYMREF,
+			.value.symref = "master",
+		};
+		snprintf(name, sizeof(name), "branch%04d", i);
+
+		err = reftable_stack_add(st1, &write_test_ref, &ref);
+		EXPECT_ERR(err);
+	}
+
+	err = reftable_new_stack(&st2, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_compact_all(st1, NULL);
+	EXPECT_ERR(err);
+
+	unclean_stack_close(st1);
+	unclean_stack_close(st2);
+
+	err = reftable_new_stack(&st3, dir, cfg);
+	EXPECT_ERR(err);
+
+	err = reftable_stack_clean(st3);
+	EXPECT_ERR(err);
+	EXPECT(count_dir_entries(dir) == 2);
+
+	reftable_stack_destroy(st1);
+	reftable_stack_destroy(st2);
+	reftable_stack_destroy(st3);
+
+	clear_dir(dir);
+}
+
+int stack_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_empty_add);
+	RUN_TEST(test_log2);
+	RUN_TEST(test_names_equal);
+	RUN_TEST(test_parse_names);
+	RUN_TEST(test_read_file);
+	RUN_TEST(test_reflog_expire);
+	RUN_TEST(test_reftable_stack_add);
+	RUN_TEST(test_reftable_stack_add_one);
+	RUN_TEST(test_reftable_stack_auto_compaction);
+	RUN_TEST(test_reftable_stack_compaction_concurrent);
+	RUN_TEST(test_reftable_stack_compaction_concurrent_clean);
+	RUN_TEST(test_reftable_stack_hash_id);
+	RUN_TEST(test_reftable_stack_lock_failure);
+	RUN_TEST(test_reftable_stack_log_normalize);
+	RUN_TEST(test_reftable_stack_tombstone);
+	RUN_TEST(test_reftable_stack_transaction_api);
+	RUN_TEST(test_reftable_stack_update_index_check);
+	RUN_TEST(test_reftable_stack_uptodate);
+	RUN_TEST(test_reftable_stack_validate_refname);
+	RUN_TEST(test_sizes_to_segments);
+	RUN_TEST(test_sizes_to_segments_all_equal);
+	RUN_TEST(test_sizes_to_segments_empty);
+	RUN_TEST(test_suggest_compaction_segment);
+	RUN_TEST(test_suggest_compaction_segment_nothing);
+	return 0;
+}
diff --git a/reftable/system.h b/reftable/system.h
new file mode 100644
index 0000000..4907306
--- /dev/null
+++ b/reftable/system.h
@@ -0,0 +1,32 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef SYSTEM_H
+#define SYSTEM_H
+
+/* This header glues the reftable library to the rest of Git */
+
+#include "git-compat-util.h"
+#include "strbuf.h"
+#include "hash.h" /* hash ID, sizes.*/
+#include "dir.h" /* remove_dir_recursively, for tests.*/
+
+#include <zlib.h>
+
+#ifdef NO_UNCOMPRESS2
+/*
+ * This is uncompress2, which is only available in zlib >= 1.2.9
+ * (released as of early 2017)
+ */
+int uncompress2(Bytef *dest, uLongf *destLen, const Bytef *source,
+		uLong *sourceLen);
+#endif
+
+int hash_size(uint32_t id);
+
+#endif
diff --git a/reftable/test_framework.c b/reftable/test_framework.c
new file mode 100644
index 0000000..84ac972
--- /dev/null
+++ b/reftable/test_framework.c
@@ -0,0 +1,23 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "system.h"
+#include "test_framework.h"
+
+#include "basics.h"
+
+void set_test_hash(uint8_t *p, int i)
+{
+	memset(p, (uint8_t)i, hash_size(GIT_SHA1_FORMAT_ID));
+}
+
+ssize_t strbuf_add_void(void *b, const void *data, size_t sz)
+{
+	strbuf_add(b, data, sz);
+	return sz;
+}
diff --git a/reftable/test_framework.h b/reftable/test_framework.h
new file mode 100644
index 0000000..774cb27
--- /dev/null
+++ b/reftable/test_framework.h
@@ -0,0 +1,53 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef TEST_FRAMEWORK_H
+#define TEST_FRAMEWORK_H
+
+#include "system.h"
+#include "reftable-error.h"
+
+#define EXPECT_ERR(c)                                                  \
+	if (c != 0) {                                                  \
+		fflush(stderr);                                        \
+		fflush(stdout);                                        \
+		fprintf(stderr, "%s: %d: error == %d (%s), want 0\n",  \
+			__FILE__, __LINE__, c, reftable_error_str(c)); \
+		abort();                                               \
+	}
+
+#define EXPECT_STREQ(a, b)                                               \
+	if (strcmp(a, b)) {                                              \
+		fflush(stderr);                                          \
+		fflush(stdout);                                          \
+		fprintf(stderr, "%s:%d: %s (%s) != %s (%s)\n", __FILE__, \
+			__LINE__, #a, a, #b, b);                         \
+		abort();                                                 \
+	}
+
+#define EXPECT(c)                                                          \
+	if (!(c)) {                                                        \
+		fflush(stderr);                                            \
+		fflush(stdout);                                            \
+		fprintf(stderr, "%s: %d: failed assertion %s\n", __FILE__, \
+			__LINE__, #c);                                     \
+		abort();                                                   \
+	}
+
+#define RUN_TEST(f)                          \
+	fprintf(stderr, "running %s\n", #f); \
+	fflush(stderr);                      \
+	f();
+
+void set_test_hash(uint8_t *p, int i);
+
+/* Like strbuf_add, but suitable for passing to reftable_new_writer
+ */
+ssize_t strbuf_add_void(void *b, const void *data, size_t sz);
+
+#endif
diff --git a/reftable/tree.c b/reftable/tree.c
new file mode 100644
index 0000000..82db799
--- /dev/null
+++ b/reftable/tree.c
@@ -0,0 +1,63 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "tree.h"
+
+#include "basics.h"
+#include "system.h"
+
+struct tree_node *tree_search(void *key, struct tree_node **rootp,
+			      int (*compare)(const void *, const void *),
+			      int insert)
+{
+	int res;
+	if (*rootp == NULL) {
+		if (!insert) {
+			return NULL;
+		} else {
+			struct tree_node *n =
+				reftable_calloc(sizeof(struct tree_node));
+			n->key = key;
+			*rootp = n;
+			return *rootp;
+		}
+	}
+
+	res = compare(key, (*rootp)->key);
+	if (res < 0)
+		return tree_search(key, &(*rootp)->left, compare, insert);
+	else if (res > 0)
+		return tree_search(key, &(*rootp)->right, compare, insert);
+	return *rootp;
+}
+
+void infix_walk(struct tree_node *t, void (*action)(void *arg, void *key),
+		void *arg)
+{
+	if (t->left) {
+		infix_walk(t->left, action, arg);
+	}
+	action(arg, t->key);
+	if (t->right) {
+		infix_walk(t->right, action, arg);
+	}
+}
+
+void tree_free(struct tree_node *t)
+{
+	if (t == NULL) {
+		return;
+	}
+	if (t->left) {
+		tree_free(t->left);
+	}
+	if (t->right) {
+		tree_free(t->right);
+	}
+	reftable_free(t);
+}
diff --git a/reftable/tree.h b/reftable/tree.h
new file mode 100644
index 0000000..fbdd002
--- /dev/null
+++ b/reftable/tree.h
@@ -0,0 +1,34 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef TREE_H
+#define TREE_H
+
+/* tree_node is a generic binary search tree. */
+struct tree_node {
+	void *key;
+	struct tree_node *left, *right;
+};
+
+/* looks for `key` in `rootp` using `compare` as comparison function. If insert
+ * is set, insert the key if it's not found. Else, return NULL.
+ */
+struct tree_node *tree_search(void *key, struct tree_node **rootp,
+			      int (*compare)(const void *, const void *),
+			      int insert);
+
+/* performs an infix walk of the tree. */
+void infix_walk(struct tree_node *t, void (*action)(void *arg, void *key),
+		void *arg);
+
+/*
+ * deallocates the tree nodes recursively. Keys should be deallocated separately
+ * by walking over the tree. */
+void tree_free(struct tree_node *t);
+
+#endif
diff --git a/reftable/tree_test.c b/reftable/tree_test.c
new file mode 100644
index 0000000..cbff125
--- /dev/null
+++ b/reftable/tree_test.c
@@ -0,0 +1,61 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "tree.h"
+
+#include "basics.h"
+#include "record.h"
+#include "test_framework.h"
+#include "reftable-tests.h"
+
+static int test_compare(const void *a, const void *b)
+{
+	return (char *)a - (char *)b;
+}
+
+struct curry {
+	void *last;
+};
+
+static void check_increasing(void *arg, void *key)
+{
+	struct curry *c = arg;
+	if (c->last) {
+		EXPECT(test_compare(c->last, key) < 0);
+	}
+	c->last = key;
+}
+
+static void test_tree(void)
+{
+	struct tree_node *root = NULL;
+
+	void *values[11] = { NULL };
+	struct tree_node *nodes[11] = { NULL };
+	int i = 1;
+	struct curry c = { NULL };
+	do {
+		nodes[i] = tree_search(values + i, &root, &test_compare, 1);
+		i = (i * 7) % 11;
+	} while (i != 1);
+
+	for (i = 1; i < ARRAY_SIZE(nodes); i++) {
+		EXPECT(values + i == nodes[i]->key);
+		EXPECT(nodes[i] ==
+		       tree_search(values + i, &root, &test_compare, 0));
+	}
+
+	infix_walk(root, check_increasing, &c);
+	tree_free(root);
+}
+
+int tree_test_main(int argc, const char *argv[])
+{
+	RUN_TEST(test_tree);
+	return 0;
+}
diff --git a/reftable/writer.c b/reftable/writer.c
new file mode 100644
index 0000000..3ca721e
--- /dev/null
+++ b/reftable/writer.c
@@ -0,0 +1,690 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#include "writer.h"
+
+#include "system.h"
+
+#include "block.h"
+#include "constants.h"
+#include "record.h"
+#include "tree.h"
+#include "reftable-error.h"
+
+/* finishes a block, and writes it to storage */
+static int writer_flush_block(struct reftable_writer *w);
+
+/* deallocates memory related to the index */
+static void writer_clear_index(struct reftable_writer *w);
+
+/* finishes writing a 'r' (refs) or 'g' (reflogs) section */
+static int writer_finish_public_section(struct reftable_writer *w);
+
+static struct reftable_block_stats *
+writer_reftable_block_stats(struct reftable_writer *w, uint8_t typ)
+{
+	switch (typ) {
+	case 'r':
+		return &w->stats.ref_stats;
+	case 'o':
+		return &w->stats.obj_stats;
+	case 'i':
+		return &w->stats.idx_stats;
+	case 'g':
+		return &w->stats.log_stats;
+	}
+	abort();
+	return NULL;
+}
+
+/* write data, queuing the padding for the next write. Returns negative for
+ * error. */
+static int padded_write(struct reftable_writer *w, uint8_t *data, size_t len,
+			int padding)
+{
+	int n = 0;
+	if (w->pending_padding > 0) {
+		uint8_t *zeroed = reftable_calloc(w->pending_padding);
+		int n = w->write(w->write_arg, zeroed, w->pending_padding);
+		if (n < 0)
+			return n;
+
+		w->pending_padding = 0;
+		reftable_free(zeroed);
+	}
+
+	w->pending_padding = padding;
+	n = w->write(w->write_arg, data, len);
+	if (n < 0)
+		return n;
+	n += padding;
+	return 0;
+}
+
+static void options_set_defaults(struct reftable_write_options *opts)
+{
+	if (opts->restart_interval == 0) {
+		opts->restart_interval = 16;
+	}
+
+	if (opts->hash_id == 0) {
+		opts->hash_id = GIT_SHA1_FORMAT_ID;
+	}
+	if (opts->block_size == 0) {
+		opts->block_size = DEFAULT_BLOCK_SIZE;
+	}
+}
+
+static int writer_version(struct reftable_writer *w)
+{
+	return (w->opts.hash_id == 0 || w->opts.hash_id == GIT_SHA1_FORMAT_ID) ?
+			     1 :
+			     2;
+}
+
+static int writer_write_header(struct reftable_writer *w, uint8_t *dest)
+{
+	memcpy(dest, "REFT", 4);
+
+	dest[4] = writer_version(w);
+
+	put_be24(dest + 5, w->opts.block_size);
+	put_be64(dest + 8, w->min_update_index);
+	put_be64(dest + 16, w->max_update_index);
+	if (writer_version(w) == 2) {
+		put_be32(dest + 24, w->opts.hash_id);
+	}
+	return header_size(writer_version(w));
+}
+
+static void writer_reinit_block_writer(struct reftable_writer *w, uint8_t typ)
+{
+	int block_start = 0;
+	if (w->next == 0) {
+		block_start = header_size(writer_version(w));
+	}
+
+	strbuf_release(&w->last_key);
+	block_writer_init(&w->block_writer_data, typ, w->block,
+			  w->opts.block_size, block_start,
+			  hash_size(w->opts.hash_id));
+	w->block_writer = &w->block_writer_data;
+	w->block_writer->restart_interval = w->opts.restart_interval;
+}
+
+static struct strbuf reftable_empty_strbuf = STRBUF_INIT;
+
+struct reftable_writer *
+reftable_new_writer(ssize_t (*writer_func)(void *, const void *, size_t),
+		    void *writer_arg, struct reftable_write_options *opts)
+{
+	struct reftable_writer *wp =
+		reftable_calloc(sizeof(struct reftable_writer));
+	strbuf_init(&wp->block_writer_data.last_key, 0);
+	options_set_defaults(opts);
+	if (opts->block_size >= (1 << 24)) {
+		/* TODO - error return? */
+		abort();
+	}
+	wp->last_key = reftable_empty_strbuf;
+	wp->block = reftable_calloc(opts->block_size);
+	wp->write = writer_func;
+	wp->write_arg = writer_arg;
+	wp->opts = *opts;
+	writer_reinit_block_writer(wp, BLOCK_TYPE_REF);
+
+	return wp;
+}
+
+void reftable_writer_set_limits(struct reftable_writer *w, uint64_t min,
+				uint64_t max)
+{
+	w->min_update_index = min;
+	w->max_update_index = max;
+}
+
+void reftable_writer_free(struct reftable_writer *w)
+{
+	reftable_free(w->block);
+	reftable_free(w);
+}
+
+struct obj_index_tree_node {
+	struct strbuf hash;
+	uint64_t *offsets;
+	size_t offset_len;
+	size_t offset_cap;
+};
+
+#define OBJ_INDEX_TREE_NODE_INIT    \
+	{                           \
+		.hash = STRBUF_INIT \
+	}
+
+static int obj_index_tree_node_compare(const void *a, const void *b)
+{
+	return strbuf_cmp(&((const struct obj_index_tree_node *)a)->hash,
+			  &((const struct obj_index_tree_node *)b)->hash);
+}
+
+static void writer_index_hash(struct reftable_writer *w, struct strbuf *hash)
+{
+	uint64_t off = w->next;
+
+	struct obj_index_tree_node want = { .hash = *hash };
+
+	struct tree_node *node = tree_search(&want, &w->obj_index_tree,
+					     &obj_index_tree_node_compare, 0);
+	struct obj_index_tree_node *key = NULL;
+	if (node == NULL) {
+		struct obj_index_tree_node empty = OBJ_INDEX_TREE_NODE_INIT;
+		key = reftable_malloc(sizeof(struct obj_index_tree_node));
+		*key = empty;
+
+		strbuf_reset(&key->hash);
+		strbuf_addbuf(&key->hash, hash);
+		tree_search((void *)key, &w->obj_index_tree,
+			    &obj_index_tree_node_compare, 1);
+	} else {
+		key = node->key;
+	}
+
+	if (key->offset_len > 0 && key->offsets[key->offset_len - 1] == off) {
+		return;
+	}
+
+	if (key->offset_len == key->offset_cap) {
+		key->offset_cap = 2 * key->offset_cap + 1;
+		key->offsets = reftable_realloc(
+			key->offsets, sizeof(uint64_t) * key->offset_cap);
+	}
+
+	key->offsets[key->offset_len++] = off;
+}
+
+static int writer_add_record(struct reftable_writer *w,
+			     struct reftable_record *rec)
+{
+	struct strbuf key = STRBUF_INIT;
+	int err = -1;
+	reftable_record_key(rec, &key);
+	if (strbuf_cmp(&w->last_key, &key) >= 0) {
+		err = REFTABLE_API_ERROR;
+		goto done;
+	}
+
+	strbuf_reset(&w->last_key);
+	strbuf_addbuf(&w->last_key, &key);
+	if (w->block_writer == NULL) {
+		writer_reinit_block_writer(w, reftable_record_type(rec));
+	}
+
+	assert(block_writer_type(w->block_writer) == reftable_record_type(rec));
+
+	if (block_writer_add(w->block_writer, rec) == 0) {
+		err = 0;
+		goto done;
+	}
+
+	err = writer_flush_block(w);
+	if (err < 0) {
+		goto done;
+	}
+
+	writer_reinit_block_writer(w, reftable_record_type(rec));
+	err = block_writer_add(w->block_writer, rec);
+	if (err < 0) {
+		goto done;
+	}
+
+	err = 0;
+done:
+	strbuf_release(&key);
+	return err;
+}
+
+int reftable_writer_add_ref(struct reftable_writer *w,
+			    struct reftable_ref_record *ref)
+{
+	struct reftable_record rec = { NULL };
+	struct reftable_ref_record copy = *ref;
+	int err = 0;
+
+	if (ref->refname == NULL)
+		return REFTABLE_API_ERROR;
+	if (ref->update_index < w->min_update_index ||
+	    ref->update_index > w->max_update_index)
+		return REFTABLE_API_ERROR;
+
+	reftable_record_from_ref(&rec, &copy);
+	copy.update_index -= w->min_update_index;
+
+	err = writer_add_record(w, &rec);
+	if (err < 0)
+		return err;
+
+	if (!w->opts.skip_index_objects && reftable_ref_record_val1(ref)) {
+		struct strbuf h = STRBUF_INIT;
+		strbuf_add(&h, (char *)reftable_ref_record_val1(ref),
+			   hash_size(w->opts.hash_id));
+		writer_index_hash(w, &h);
+		strbuf_release(&h);
+	}
+
+	if (!w->opts.skip_index_objects && reftable_ref_record_val2(ref)) {
+		struct strbuf h = STRBUF_INIT;
+		strbuf_add(&h, reftable_ref_record_val2(ref),
+			   hash_size(w->opts.hash_id));
+		writer_index_hash(w, &h);
+		strbuf_release(&h);
+	}
+	return 0;
+}
+
+int reftable_writer_add_refs(struct reftable_writer *w,
+			     struct reftable_ref_record *refs, int n)
+{
+	int err = 0;
+	int i = 0;
+	QSORT(refs, n, reftable_ref_record_compare_name);
+	for (i = 0; err == 0 && i < n; i++) {
+		err = reftable_writer_add_ref(w, &refs[i]);
+	}
+	return err;
+}
+
+static int reftable_writer_add_log_verbatim(struct reftable_writer *w,
+					    struct reftable_log_record *log)
+{
+	struct reftable_record rec = { NULL };
+	if (w->block_writer &&
+	    block_writer_type(w->block_writer) == BLOCK_TYPE_REF) {
+		int err = writer_finish_public_section(w);
+		if (err < 0)
+			return err;
+	}
+
+	w->next -= w->pending_padding;
+	w->pending_padding = 0;
+
+	reftable_record_from_log(&rec, log);
+	return writer_add_record(w, &rec);
+}
+
+int reftable_writer_add_log(struct reftable_writer *w,
+			    struct reftable_log_record *log)
+{
+	char *input_log_message = NULL;
+	struct strbuf cleaned_message = STRBUF_INIT;
+	int err = 0;
+
+	if (log->value_type == REFTABLE_LOG_DELETION)
+		return reftable_writer_add_log_verbatim(w, log);
+
+	if (log->refname == NULL)
+		return REFTABLE_API_ERROR;
+
+	input_log_message = log->value.update.message;
+	if (!w->opts.exact_log_message && log->value.update.message) {
+		strbuf_addstr(&cleaned_message, log->value.update.message);
+		while (cleaned_message.len &&
+		       cleaned_message.buf[cleaned_message.len - 1] == '\n')
+			strbuf_setlen(&cleaned_message,
+				      cleaned_message.len - 1);
+		if (strchr(cleaned_message.buf, '\n')) {
+			/* multiple lines not allowed. */
+			err = REFTABLE_API_ERROR;
+			goto done;
+		}
+		strbuf_addstr(&cleaned_message, "\n");
+		log->value.update.message = cleaned_message.buf;
+	}
+
+	err = reftable_writer_add_log_verbatim(w, log);
+	log->value.update.message = input_log_message;
+done:
+	strbuf_release(&cleaned_message);
+	return err;
+}
+
+int reftable_writer_add_logs(struct reftable_writer *w,
+			     struct reftable_log_record *logs, int n)
+{
+	int err = 0;
+	int i = 0;
+	QSORT(logs, n, reftable_log_record_compare_key);
+
+	for (i = 0; err == 0 && i < n; i++) {
+		err = reftable_writer_add_log(w, &logs[i]);
+	}
+	return err;
+}
+
+static int writer_finish_section(struct reftable_writer *w)
+{
+	uint8_t typ = block_writer_type(w->block_writer);
+	uint64_t index_start = 0;
+	int max_level = 0;
+	int threshold = w->opts.unpadded ? 1 : 3;
+	int before_blocks = w->stats.idx_stats.blocks;
+	int err = writer_flush_block(w);
+	int i = 0;
+	struct reftable_block_stats *bstats = NULL;
+	if (err < 0)
+		return err;
+
+	while (w->index_len > threshold) {
+		struct reftable_index_record *idx = NULL;
+		int idx_len = 0;
+
+		max_level++;
+		index_start = w->next;
+		writer_reinit_block_writer(w, BLOCK_TYPE_INDEX);
+
+		idx = w->index;
+		idx_len = w->index_len;
+
+		w->index = NULL;
+		w->index_len = 0;
+		w->index_cap = 0;
+		for (i = 0; i < idx_len; i++) {
+			struct reftable_record rec = { NULL };
+			reftable_record_from_index(&rec, idx + i);
+			if (block_writer_add(w->block_writer, &rec) == 0) {
+				continue;
+			}
+
+			err = writer_flush_block(w);
+			if (err < 0)
+				return err;
+
+			writer_reinit_block_writer(w, BLOCK_TYPE_INDEX);
+
+			err = block_writer_add(w->block_writer, &rec);
+			if (err != 0) {
+				/* write into fresh block should always succeed
+				 */
+				abort();
+			}
+		}
+		for (i = 0; i < idx_len; i++) {
+			strbuf_release(&idx[i].last_key);
+		}
+		reftable_free(idx);
+	}
+
+	writer_clear_index(w);
+
+	err = writer_flush_block(w);
+	if (err < 0)
+		return err;
+
+	bstats = writer_reftable_block_stats(w, typ);
+	bstats->index_blocks = w->stats.idx_stats.blocks - before_blocks;
+	bstats->index_offset = index_start;
+	bstats->max_index_level = max_level;
+
+	/* Reinit lastKey, as the next section can start with any key. */
+	w->last_key.len = 0;
+
+	return 0;
+}
+
+struct common_prefix_arg {
+	struct strbuf *last;
+	int max;
+};
+
+static void update_common(void *void_arg, void *key)
+{
+	struct common_prefix_arg *arg = void_arg;
+	struct obj_index_tree_node *entry = key;
+	if (arg->last) {
+		int n = common_prefix_size(&entry->hash, arg->last);
+		if (n > arg->max) {
+			arg->max = n;
+		}
+	}
+	arg->last = &entry->hash;
+}
+
+struct write_record_arg {
+	struct reftable_writer *w;
+	int err;
+};
+
+static void write_object_record(void *void_arg, void *key)
+{
+	struct write_record_arg *arg = void_arg;
+	struct obj_index_tree_node *entry = key;
+	struct reftable_obj_record obj_rec = {
+		.hash_prefix = (uint8_t *)entry->hash.buf,
+		.hash_prefix_len = arg->w->stats.object_id_len,
+		.offsets = entry->offsets,
+		.offset_len = entry->offset_len,
+	};
+	struct reftable_record rec = { NULL };
+	if (arg->err < 0)
+		goto done;
+
+	reftable_record_from_obj(&rec, &obj_rec);
+	arg->err = block_writer_add(arg->w->block_writer, &rec);
+	if (arg->err == 0)
+		goto done;
+
+	arg->err = writer_flush_block(arg->w);
+	if (arg->err < 0)
+		goto done;
+
+	writer_reinit_block_writer(arg->w, BLOCK_TYPE_OBJ);
+	arg->err = block_writer_add(arg->w->block_writer, &rec);
+	if (arg->err == 0)
+		goto done;
+	obj_rec.offset_len = 0;
+	arg->err = block_writer_add(arg->w->block_writer, &rec);
+
+	/* Should be able to write into a fresh block. */
+	assert(arg->err == 0);
+
+done:;
+}
+
+static void object_record_free(void *void_arg, void *key)
+{
+	struct obj_index_tree_node *entry = key;
+
+	FREE_AND_NULL(entry->offsets);
+	strbuf_release(&entry->hash);
+	reftable_free(entry);
+}
+
+static int writer_dump_object_index(struct reftable_writer *w)
+{
+	struct write_record_arg closure = { .w = w };
+	struct common_prefix_arg common = { NULL };
+	if (w->obj_index_tree) {
+		infix_walk(w->obj_index_tree, &update_common, &common);
+	}
+	w->stats.object_id_len = common.max + 1;
+
+	writer_reinit_block_writer(w, BLOCK_TYPE_OBJ);
+
+	if (w->obj_index_tree) {
+		infix_walk(w->obj_index_tree, &write_object_record, &closure);
+	}
+
+	if (closure.err < 0)
+		return closure.err;
+	return writer_finish_section(w);
+}
+
+static int writer_finish_public_section(struct reftable_writer *w)
+{
+	uint8_t typ = 0;
+	int err = 0;
+
+	if (w->block_writer == NULL)
+		return 0;
+
+	typ = block_writer_type(w->block_writer);
+	err = writer_finish_section(w);
+	if (err < 0)
+		return err;
+	if (typ == BLOCK_TYPE_REF && !w->opts.skip_index_objects &&
+	    w->stats.ref_stats.index_blocks > 0) {
+		err = writer_dump_object_index(w);
+		if (err < 0)
+			return err;
+	}
+
+	if (w->obj_index_tree) {
+		infix_walk(w->obj_index_tree, &object_record_free, NULL);
+		tree_free(w->obj_index_tree);
+		w->obj_index_tree = NULL;
+	}
+
+	w->block_writer = NULL;
+	return 0;
+}
+
+int reftable_writer_close(struct reftable_writer *w)
+{
+	uint8_t footer[72];
+	uint8_t *p = footer;
+	int err = writer_finish_public_section(w);
+	int empty_table = w->next == 0;
+	if (err != 0)
+		goto done;
+	w->pending_padding = 0;
+	if (empty_table) {
+		/* Empty tables need a header anyway. */
+		uint8_t header[28];
+		int n = writer_write_header(w, header);
+		err = padded_write(w, header, n, 0);
+		if (err < 0)
+			goto done;
+	}
+
+	p += writer_write_header(w, footer);
+	put_be64(p, w->stats.ref_stats.index_offset);
+	p += 8;
+	put_be64(p, (w->stats.obj_stats.offset) << 5 | w->stats.object_id_len);
+	p += 8;
+	put_be64(p, w->stats.obj_stats.index_offset);
+	p += 8;
+
+	put_be64(p, w->stats.log_stats.offset);
+	p += 8;
+	put_be64(p, w->stats.log_stats.index_offset);
+	p += 8;
+
+	put_be32(p, crc32(0, footer, p - footer));
+	p += 4;
+
+	err = padded_write(w, footer, footer_size(writer_version(w)), 0);
+	if (err < 0)
+		goto done;
+
+	if (empty_table) {
+		err = REFTABLE_EMPTY_TABLE_ERROR;
+		goto done;
+	}
+
+done:
+	/* free up memory. */
+	block_writer_release(&w->block_writer_data);
+	writer_clear_index(w);
+	strbuf_release(&w->last_key);
+	return err;
+}
+
+static void writer_clear_index(struct reftable_writer *w)
+{
+	int i = 0;
+	for (i = 0; i < w->index_len; i++) {
+		strbuf_release(&w->index[i].last_key);
+	}
+
+	FREE_AND_NULL(w->index);
+	w->index_len = 0;
+	w->index_cap = 0;
+}
+
+static const int debug = 0;
+
+static int writer_flush_nonempty_block(struct reftable_writer *w)
+{
+	uint8_t typ = block_writer_type(w->block_writer);
+	struct reftable_block_stats *bstats =
+		writer_reftable_block_stats(w, typ);
+	uint64_t block_typ_off = (bstats->blocks == 0) ? w->next : 0;
+	int raw_bytes = block_writer_finish(w->block_writer);
+	int padding = 0;
+	int err = 0;
+	struct reftable_index_record ir = { .last_key = STRBUF_INIT };
+	if (raw_bytes < 0)
+		return raw_bytes;
+
+	if (!w->opts.unpadded && typ != BLOCK_TYPE_LOG) {
+		padding = w->opts.block_size - raw_bytes;
+	}
+
+	if (block_typ_off > 0) {
+		bstats->offset = block_typ_off;
+	}
+
+	bstats->entries += w->block_writer->entries;
+	bstats->restarts += w->block_writer->restart_len;
+	bstats->blocks++;
+	w->stats.blocks++;
+
+	if (debug) {
+		fprintf(stderr, "block %c off %" PRIu64 " sz %d (%d)\n", typ,
+			w->next, raw_bytes,
+			get_be24(w->block + w->block_writer->header_off + 1));
+	}
+
+	if (w->next == 0) {
+		writer_write_header(w, w->block);
+	}
+
+	err = padded_write(w, w->block, raw_bytes, padding);
+	if (err < 0)
+		return err;
+
+	if (w->index_cap == w->index_len) {
+		w->index_cap = 2 * w->index_cap + 1;
+		w->index = reftable_realloc(
+			w->index,
+			sizeof(struct reftable_index_record) * w->index_cap);
+	}
+
+	ir.offset = w->next;
+	strbuf_reset(&ir.last_key);
+	strbuf_addbuf(&ir.last_key, &w->block_writer->last_key);
+	w->index[w->index_len] = ir;
+
+	w->index_len++;
+	w->next += padding + raw_bytes;
+	w->block_writer = NULL;
+	return 0;
+}
+
+static int writer_flush_block(struct reftable_writer *w)
+{
+	if (w->block_writer == NULL)
+		return 0;
+	if (w->block_writer->entries == 0)
+		return 0;
+	return writer_flush_nonempty_block(w);
+}
+
+const struct reftable_stats *writer_stats(struct reftable_writer *w)
+{
+	return &w->stats;
+}
diff --git a/reftable/writer.h b/reftable/writer.h
new file mode 100644
index 0000000..09b8867
--- /dev/null
+++ b/reftable/writer.h
@@ -0,0 +1,50 @@
+/*
+Copyright 2020 Google LLC
+
+Use of this source code is governed by a BSD-style
+license that can be found in the LICENSE file or at
+https://developers.google.com/open-source/licenses/bsd
+*/
+
+#ifndef WRITER_H
+#define WRITER_H
+
+#include "basics.h"
+#include "block.h"
+#include "tree.h"
+#include "reftable-writer.h"
+
+struct reftable_writer {
+	ssize_t (*write)(void *, const void *, size_t);
+	void *write_arg;
+	int pending_padding;
+	struct strbuf last_key;
+
+	/* offset of next block to write. */
+	uint64_t next;
+	uint64_t min_update_index, max_update_index;
+	struct reftable_write_options opts;
+
+	/* memory buffer for writing */
+	uint8_t *block;
+
+	/* writer for the current section. NULL or points to
+	 * block_writer_data */
+	struct block_writer *block_writer;
+
+	struct block_writer block_writer_data;
+
+	/* pending index records for the current section */
+	struct reftable_index_record *index;
+	size_t index_len;
+	size_t index_cap;
+
+	/*
+	 * tree for use with tsearch; used to populate the 'o' inverse OID
+	 * map */
+	struct tree_node *obj_index_tree;
+
+	struct reftable_stats stats;
+};
+
+#endif
diff --git a/t/helper/test-reftable.c b/t/helper/test-reftable.c
new file mode 100644
index 0000000..26b03d7
--- /dev/null
+++ b/t/helper/test-reftable.c
@@ -0,0 +1,21 @@
+#include "reftable/reftable-tests.h"
+#include "test-tool.h"
+
+int cmd__reftable(int argc, const char **argv)
+{
+	basics_test_main(argc, argv);
+	block_test_main(argc, argv);
+	merged_test_main(argc, argv);
+	pq_test_main(argc, argv);
+	record_test_main(argc, argv);
+	refname_test_main(argc, argv);
+	readwrite_test_main(argc, argv);
+	stack_test_main(argc, argv);
+	tree_test_main(argc, argv);
+	return 0;
+}
+
+int cmd__dump_reftable(int argc, const char **argv)
+{
+	return reftable_dump_main(argc, (char *const *)argv);
+}
diff --git a/t/helper/test-tool.c b/t/helper/test-tool.c
index 3ce5585..338a57b 100644
--- a/t/helper/test-tool.c
+++ b/t/helper/test-tool.c
@@ -53,13 +53,15 @@ static struct test_cmd cmds[] = {
 	{ "pcre2-config", cmd__pcre2_config },
 	{ "pkt-line", cmd__pkt_line },
 	{ "prio-queue", cmd__prio_queue },
-	{ "proc-receive", cmd__proc_receive},
+	{ "proc-receive", cmd__proc_receive },
 	{ "progress", cmd__progress },
 	{ "reach", cmd__reach },
 	{ "read-cache", cmd__read_cache },
 	{ "read-graph", cmd__read_graph },
 	{ "read-midx", cmd__read_midx },
 	{ "ref-store", cmd__ref_store },
+	{ "reftable", cmd__reftable },
+	{ "dump-reftable", cmd__dump_reftable },
 	{ "regex", cmd__regex },
 	{ "repository", cmd__repository },
 	{ "revision-walking", cmd__revision_walking },
diff --git a/t/helper/test-tool.h b/t/helper/test-tool.h
index 9f0f522..48cee1f 100644
--- a/t/helper/test-tool.h
+++ b/t/helper/test-tool.h
@@ -19,6 +19,7 @@ int cmd__dump_cache_tree(int argc, const char **argv);
 int cmd__dump_fsmonitor(int argc, const char **argv);
 int cmd__dump_split_index(int argc, const char **argv);
 int cmd__dump_untracked_cache(int argc, const char **argv);
+int cmd__dump_reftable(int argc, const char **argv);
 int cmd__example_decorate(int argc, const char **argv);
 int cmd__fast_rebase(int argc, const char **argv);
 int cmd__genrandom(int argc, const char **argv);
@@ -49,6 +50,7 @@ int cmd__read_cache(int argc, const char **argv);
 int cmd__read_graph(int argc, const char **argv);
 int cmd__read_midx(int argc, const char **argv);
 int cmd__ref_store(int argc, const char **argv);
+int cmd__reftable(int argc, const char **argv);
 int cmd__regex(int argc, const char **argv);
 int cmd__repository(int argc, const char **argv);
 int cmd__revision_walking(int argc, const char **argv);
diff --git a/t/t0032-reftable-unittest.sh b/t/t0032-reftable-unittest.sh
new file mode 100755
index 0000000..0ed1497
--- /dev/null
+++ b/t/t0032-reftable-unittest.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+#
+# Copyright (c) 2020 Google LLC
+#
+
+test_description='reftable unittests'
+
+. ./test-lib.sh
+
+test_expect_success 'unittests' '
+	TMPDIR=$(pwd) && export TMPDIR &&
+	test-tool reftable
+'
+
+test_done