[GC] Don't crash when printing gc'd sections with no name and no associated file

[davide]

Summary

Accidentally found while debugging an unrelated lto issue.

[davide]

To be more precise, before this code was crashing because getFile() returned a nullptr on which lld was trying to call getName().

[davide]

C->getFile() == nullptr
true
F->getName()
"LLD-INTERNAL-combined-lto-object"
    [Raw View]: {Data=0x0000000144483f18 "LLD-INTERNAL-combined-lto-object" Length=32 }

It comes from the combined LTO object.
The input section name is empty anyway so I'm not entirely sure what buys us in this particular circumstance.
In general, anyway, yes, I think <internal> is a better diagnostic if we don't have the filename available. Do you agree?

[ruiu]

How can a section from combined LTO object have no backing file? I guessed its getFile() should return a combined LTO object file.

[davide]

I analyzed this a little bit further.
I think this happen only when we try to print the "special" section &InputSection<ELFT>::Discarded (the debugger confirms).
This has no name associated, and I don't think it's backed by a file. I think it's not entirely unreasonable to skip it while reporting sections which have been gc'ed. What do you think?

[ruiu]

LGTM

lld/trunk/ELF/Writer.cpp

for (StringRef V :
     StringRef Prefix = V.drop_back();
     if (Name.startswith(V) || Name == Prefix)
       return Prefix;
   }
   return Name;
 }
 
 template <class ELFT> void elf::reportDiscarded(InputSectionBase<ELFT> *IS) {
-  if (!Config->PrintGcSections || !IS || IS->Live)
+  if (!Config->PrintGcSections || !IS || IS == &InputSection<ELFT>::Discarded ||
+      IS->Live)
     return;
   errs() << "removing unused section from '" << IS->Name << "' in file '"
          << IS->getFile()->getName() << "'\n";
 }
 
 template <class ELFT> static bool needsInterpSection() {
   return !Symtab<ELFT>::X->getSharedFiles().empty() &&
          !Config->DynamicLinker.empty() &&

lld/trunk/test/ELF/lto/dynsym.ll

 ; REQUIRES: x86
 ; RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux -o %t.o %p/Inputs/dynsym.s
 ; RUN: ld.lld -m elf_x86_64 %t.o -o %t.so -shared
 ; RUN: llvm-as %s -o %t2.o
 ; RUN: ld.lld -m elf_x86_64 %t2.o %t.so -o %t
 ; RUN: llvm-readobj -dyn-symbols %t | FileCheck %s
 
+; Check that we don't crash when gc'ing sections and printing the result.
+; RUN: ld.lld -m elf_x86_64 %t2.o %t.so --gc-sections --print-gc-sections \
+; RUN:   -o %t
+; RUN: llvm-readobj -dyn-symbols %t | FileCheck %s
+
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define void @_start() {
   call void @foo()
   ret void
 }