Keep underscores in DW_AT_name of DW_TAG_labels

[dimitry]

Summary

Do not remove leading underscores from DW_AT_names of DW_TAG_labels.

Bug: https://bugs.llvm.org/show_bug.cgi?id=37470

[JDevlieghere]

This still breaks the assembler scenario @aprantl mentioned in D46878. Can we differentiate between the two in MC so we can fix PR37470 without that side-effect?

[aprantl]

This behavior goes back to r146262:

commit 94c2e85bea1ab1b837a4c055ccc83d5cd32dd027
Author: Kevin Enderby <[e-mail]>
Date:   Fri Dec 9 18:09:40 2011 +0000

    The second part of support for generating dwarf for assembly source files.  
This
    generates the dwarf Compile Unit DIE and a dwarf subprogram DIE for each
    non-temporary label.
    
    The next part will be to get the clang driver to enable this when assembling
    a .s file.  rdar://9275556
    
    
    git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146262

which added

+  // The dwarf subprogram's name does not have the symbol name's leading
+  // underbar if any.
+  StringRef Name = Symbol->getName();
+  if (Name.startswith("_"))
+    Name = Name.substr(1, Name.size()-1);
+

[aprantl]

My *guess* is that the underscore trimming was added to allow setting breakpoints by name on labels in assembler output generated by a compiler, but I'm not sure. Perhaps @echristo knows more?

[aprantl]

(The radar doesn't mention this at all unfortunately)

[echristo]

In D47104#1219760, @aprantl wrote:

My *guess* is that the underscore trimming was added to allow setting breakpoints by name on labels in assembler output generated by a compiler, but I'm not sure. Perhaps @echristo knows more?

I don't sadly. I believe the intention was to make it so that breakpoints would be for the "obvious" symbol name rather than the "mangled with an underscore" that you need for the platform. So... whatever makes that easiest? I'd probably think a linkage name would be more helpful than mangling/demangling the actual symbol name here.

Thoughts?

include/llvm/MC/MCDwarf.h

public:
   //
   static void Emit(MCStreamer *MCOS);
 };
 
 // When generating dwarf for assembly source files this is the info that is
 // needed to be gathered for each symbol that will have a dwarf label.
 class MCGenDwarfLabelEntry {
 private:
-  // Name of the symbol without a leading underbar, if any.
+  // Name of the symbol.
   StringRef Name;
   // The dwarf file number this symbol is in.
   unsigned FileNumber;
   // The line number this symbol is at.
   unsigned LineNumber;
   // The low_pc for the dwarf label is taken from this symbol.
   MCSymbol *Label;
 

lib/MC/MCDwarf.cpp

static void EmitGenDwarfInfo(MCStreamer *MCOS,
 
   // Loop on saved info for dwarf labels and create the DIEs for them.
   const std::vector<MCGenDwarfLabelEntry> &Entries =
       MCOS->getContext().getMCGenDwarfLabelEntries();
   for (const auto &Entry : Entries) {
     // The DW_TAG_label DIE abbrev (2).
     MCOS->EmitULEB128IntValue(2);
 
-    // AT_name, of the label without any leading underbar.
+    // AT_name, of the label.
     MCOS->EmitBytes(Entry.getName());
     MCOS->EmitIntValue(0, 1); // NULL byte to terminate the string.
 
     // AT_decl_file, index into the file table.
     MCOS->EmitIntValue(Entry.getFileNumber(), 4);
 
     // AT_decl_line, source line number.
     MCOS->EmitIntValue(Entry.getLineNumber(), 4);
void MCGenDwarfLabelEntry::Make(MCSymbol *Symbol, MCStreamer *MCOS,
   if (Symbol->isTemporary())
     return;
   MCContext &context = MCOS->getContext();
   // We won't create dwarf labels for symbols in sections that we are not
   // generating debug info for.
   if (!context.getGenDwarfSectionSyms().count(MCOS->getCurrentSectionOnly()))
     return;
 
-  // The dwarf label's name does not have the symbol name's leading
-  // underbar if any.
   StringRef Name = Symbol->getName();
-  if (Name.startswith("_"))
-    Name = Name.substr(1, Name.size()-1);
 
   // Get the dwarf file number to be used for the dwarf label.
   unsigned FileNumber = context.getGenDwarfFileNumber();
 
   // Finding the line number is the expensive part which is why we just don't
   // pass it in as for some symbols we won't create a dwarf label.
   unsigned CurBuffer = SrcMgr.FindBufferContainingLoc(Loc);
   unsigned LineNumber = SrcMgr.FindLineNumber(Loc, CurBuffer);

test/MC/MachO/gen-dwarf.s

 // CHECK:    DW_AT_high_pc (0x0000000000000008)
 // We don't check the file name as it is a temp directory
 // CHECK:    DW_AT_name
 // We don't check the DW_AT_comp_dir which is the current working directory
 // CHECK:    DW_AT_producer ("llvm-mc (based on {{.*}})")
 // CHECK:    DW_AT_language (DW_LANG_Mips_Assembler)
 
 // CHECK:    DW_TAG_label
-// CHECK:      DW_AT_name ("bar")
+// CHECK:      DW_AT_name ("_bar")
 // CHECK:      DW_AT_decl_file ([[FILE:".*gen-dwarf.s"]])
 // CHECK:      DW_AT_decl_line (5)
 // CHECK:      DW_AT_low_pc (0x0000000000000000)
 // CHECK:      DW_AT_prototyped (0x00)
 
 // CHECK:      DW_TAG_unspecified_parameters
 
 // CHECK:      NULL
 
 // CHECK:    DW_TAG_label
-// CHECK:      DW_AT_name ("foo")
+// CHECK:      DW_AT_name ("_foo")
 // CHECK:      DW_AT_decl_file ([[FILE]])
 // CHECK:      DW_AT_decl_line (9)
 // CHECK:      DW_AT_low_pc (0x0000000000000007)
 // CHECK:      DW_AT_prototyped (0x00)
 
 // CHECK:      DW_TAG_unspecified_parameters
 
 // CHECK:      NULL
 
 // CHECK:    DW_TAG_label
-// CHECK:      DW_AT_name ("baz")
+// CHECK:      DW_AT_name ("_baz")
 // CHECK:      DW_AT_decl_file ([[FILE]])
 // CHECK:      DW_AT_decl_line (10)
 // CHECK:      DW_AT_low_pc (0x0000000000000007)
 // CHECK:      DW_AT_prototyped (0x00)
 
 // CHECK:      DW_TAG_unspecified_parameters
 
 // CHECK:      NULL