codeGen: Produce local symbols for module-internal functionswip/local-symbols

It turns out that some important native debugging/profiling tools (e.g.
perf) rely only on symbol tables for function name resolution (as
opposed to using DWARF DIEs). However, previously GHC would emit
temporary symbols (e.g. `.La42b`) to identify module-internal
entities. Such symbols are dropped during linking and therefore not
visible to runtime tools (in addition to having rather un-helpful unique
names). For instance, `perf report` would often end up attributing all
cost to the libc `frame_dummy` symbol since Haskell code was no covered
by any proper symbol (see #17605).

We now rather follow the model of C compilers and emit
descriptively-named local symbols for module internal things. Since this
will increase object file size this behavior can be disabled with the
`-fno-expose-all-symbols` flag.

With this `perf record` can finally be used against Haskell executables.
Even more, with `-g3` `perf annotate` provides inline source code.

4 files changed, 26 insertions, 6 deletions

diff --git a/compiler/basicTypes/Name.hs b/compiler/basicTypes/Name.hs
index 2cbd50e..9798617 100644
--- a/compiler/basicTypes/Name.hs
+++ b/compiler/basicTypes/Name.hs
@@ -567,7 +567,10 @@ pprExternal sty uniq mod occ is_wired is_builtin
 
 pprInternal :: PprStyle -> Unique -> OccName -> SDoc
 pprInternal sty uniq occ
-  | codeStyle sty  = pprUniqueAlways uniq
+  | codeStyle sty  = sdocWithDynFlags $ \dflags ->
+                       if gopt Opt_ExposeAllSymbols dflags
+                         then char '_' <> ppr_z_occ_name occ <> ppr_underscore_unique uniq
+                         else pprUniqueAlways uniq
   | debugStyle sty = ppr_occ_name occ <> braces (hsep [pprNameSpaceBrief (occNameSpace occ),
                                                        pprUnique uniq])
   | dumpStyle sty  = ppr_occ_name occ <> ppr_underscore_unique uniq
diff --git a/compiler/cmm/CLabel.hs b/compiler/cmm/CLabel.hs
index 9452b5f..1a9e14c 100644
--- a/compiler/cmm/CLabel.hs
+++ b/compiler/cmm/CLabel.hs
@@ -1302,8 +1302,8 @@ pprCLbl (RtsLabel (RtsSlowFastTickyCtr pat))
 pprCLbl (ForeignLabel str _ _ _)
   = ftext str
 
-pprCLbl (IdLabel name _cafs flavor) =
-  internalNamePrefix name <> ppr name <> ppIdFlavor flavor
+pprCLbl (IdLabel name _cafs flavor)
+  = internalNamePrefix name <> ppr name <> ppIdFlavor flavor
 
 pprCLbl (CC_Label cc)           = ppr cc
 pprCLbl (CCS_Label ccs)         = ppr ccs
@@ -1347,8 +1347,8 @@ instance Outputable ForeignLabelSource where
         ForeignLabelInExternalPackage   -> parens $ text "external package"
 
 internalNamePrefix :: Name -> SDoc
-internalNamePrefix name = getPprStyle $ \ sty ->
-  if asmStyle sty && isRandomGenerated then
+internalNamePrefix name = sdocWithDynFlags $ \dflags -> getPprStyle $ \ sty ->
+  if asmStyle sty && isRandomGenerated && not (gopt Opt_ExposeAllSymbols dflags) then
     sdocWithPlatform $ \platform ->
       ptext (asmTempLabelPrefix platform)
   else
diff --git a/compiler/main/DynFlags.hs b/compiler/main/DynFlags.hs
index 94cee4a..65e5f77 100644
--- a/compiler/main/DynFlags.hs
+++ b/compiler/main/DynFlags.hs
@@ -656,6 +656,7 @@ data GeneralFlag
    -- forwards all -L flags to the collect2 command without using a
    -- response file and as such breaking apart.
    | Opt_SingleLibFolder
+   | Opt_ExposeAllSymbols
    | Opt_KeepCAFs
    | Opt_KeepGoing
    | Opt_ByteCode
@@ -4220,6 +4221,7 @@ fFlagsDeps = [
   flagSpec "error-spans"                      Opt_ErrorSpans,
   flagSpec "excess-precision"                 Opt_ExcessPrecision,
   flagSpec "expose-all-unfoldings"            Opt_ExposeAllUnfoldings,
+  flagSpec "expose-all-symbols"               Opt_ExposeAllSymbols,
   flagSpec "external-dynamic-refs"            Opt_ExternalDynamicRefs,
   flagSpec "external-interpreter"             Opt_ExternalInterpreter,
   flagSpec "flat-cache"                       Opt_FlatCache,
@@ -4589,7 +4591,8 @@ defaultFlags settings
       Opt_RPath,
       Opt_SharedImplib,
       Opt_SimplPreInlining,
-      Opt_VersionMacros
+      Opt_VersionMacros,
+      Opt_ExposeAllSymbols
     ]
 
     ++ [f | (ns,f) <- optLevelFlags, 0 `elem` ns]
diff --git a/docs/users_guide/phases.rst b/docs/users_guide/phases.rst
index d0afe80..705d16c 100644
--- a/docs/users_guide/phases.rst
+++ b/docs/users_guide/phases.rst
@@ -652,6 +652,20 @@ Options affecting code generation
     and ``-dynhisuf`` are the counterparts of ``-o``, ``-osuf``, and
     ``-hisuf`` respectively, but applying to the dynamic compilation.
 
+.. ghc-flag:: -fexpose-all-symbols
+    :shortdesc: Produce symbols for all functions, including internal functions.
+    :type: dynamic
+    :category: codegen
+    :default: on
+
+    By default, GHC emits verbose symbol tables which include local symbols for
+    module-internal functions. These can be useful for tools like :ref:`perf
+    <https://perf.wiki.kernel.org/>` but increase object file sizes.
+
+    :ghc-flag:`-fno-expose-all-symbols` suppresses all non-global symbol table
+    entries, resulting in smaller object file sizes at the expense of
+    debuggability.
+
 .. _options-linker:
 
 Options affecting linking