[llvm-symbolizer] Make the output with `-output-style=GNU` closer to addr2line's

[ikudrin]

Summary

This is an extract from D60067. It addresses two differences in the output of llvm-symbolizer and GNU's addr2line:

• llvm-symbolizer prints an empty line after the report of each address.
• With "-f -i=0", llvm-symbolizer replaces the name of an inlined function with the name from the symbol table, i. e., the top caller function in the inlining chain. addr2line preserves the name of the inlined function.

[jhenderson]

test/tools/llvm-symbolizer/output-style-column.test

1

This test is getting too big to follow. I wonder if it would be better to add llvm-symbolizer --output-style=GNU/LLVM cases to the existing tests for -i/-p?

[ikudrin]

test/tools/llvm-symbolizer/output-style-column.test

1

It was my first intention but it looked even worse.

[ikudrin]

• Simplified tests.

[jhenderson]

LGTM, but I wouldn't mind a second pair of eyes on the new logic (@rupprecht?), as I'm not really familiar with how the DILineInfo class works.

test/tools/llvm-symbolizer/output-style-empty-line.test

19

Put a {{$}} at the end of this line, if you don't want it matching the column output.

[ikudrin]

test/tools/llvm-symbolizer/output-style-empty-line.test

19

I am trying to follow the rule that the test should check only one specific thing, if possible. This very test checks only the presence or absence of an empty line. There is another test, 'output-style-column.test', which checks for the column.

[jhenderson]

test/tools/llvm-symbolizer/output-style-empty-line.test

19

Okay, fair enough.

[rupprecht]

test/tools/llvm-symbolizer/output-style-inlined.test

18

Additionally, GNU-NOT: main?

[jhenderson]

test/tools/llvm-symbolizer/output-style-inlined.test

18

Better would be to use --implicit-check-not=main (to avoid ordering issues).

[ikudrin]

• Added --implicit-check-not=main and --implicit-check-not=inctwo to the checks in output-style-inlined.test. I added the similar option to the other two checks so that all the checks in the test follow the same pattern.

test/tools/llvm-symbolizer/output-style-empty-line.test

+This test checks that with --output-style=GNU the tool does not print an empty
+line after the report for an address. The current behavior is preserved for
+--output-style=LLVM or if the option is omitted.
+
+RUN: llvm-symbolizer -e %p/Inputs/addr.exe < %p/Inputs/addr.inp \
+RUN:   | FileCheck %s --check-prefix=LLVM
+
+RUN: llvm-symbolizer --output-style=LLVM -e %p/Inputs/addr.exe < %p/Inputs/addr.inp \
+RUN:   | FileCheck %s --check-prefix=LLVM
+
+RUN: llvm-symbolizer --output-style=GNU -e %p/Inputs/addr.exe < %p/Inputs/addr.inp \
+RUN:   | FileCheck %s --check-prefix=GNU
+
+LLVM: x.c:14:0
+LLVM-EMPTY:
+LLVM-NEXT: some text2
+
+GNU: x.c:14
+GNU-NEXT: some text2

test/tools/llvm-symbolizer/output-style-inlined.test

+This test checks that when inlined frames are not shown (-i=0) and the output
+style is set to GNU (--output-style=GNU) the name of an inlined function is not
+replaced with the name of the top caller function. At the same time, the current
+behavior of llvm-symbolizer is preserved with --output-style=LLVM or when
+the option is not specified.
+
+RUN: llvm-symbolizer -i=0 -e %p/Inputs/addr.exe 0x40054d \
+RUN:   | FileCheck %s --check-prefix=LLVM --implicit-check-not=inctwo
+
+RUN: llvm-symbolizer --output-style=LLVM -i=0 -e %p/Inputs/addr.exe 0x40054d \
+RUN:   | FileCheck %s --check-prefix=LLVM --implicit-check-not=inctwo
+
+RUN: llvm-symbolizer --output-style=GNU -i=0 -e %p/Inputs/addr.exe 0x40054d \
+RUN:   | FileCheck %s --check-prefix=GNU --implicit-check-not=main
+
+LLVM: main
+GNU: inctwo

tools/llvm-symbolizer/llvm-symbolizer.cpp

if (IsData) {
     auto ResOrErr = Symbolizer.symbolizeData(
         ModuleName, {Offset, object::SectionedAddress::UndefSection});
     Printer << (error(ResOrErr) ? DIGlobal() : ResOrErr.get());
   } else if (ClPrintInlining) {
     auto ResOrErr = Symbolizer.symbolizeInlinedCode(
         ModuleName, {Offset, object::SectionedAddress::UndefSection},
         ClDwpName);
     Printer << (error(ResOrErr) ? DIInliningInfo() : ResOrErr.get());
+  } else if (ClOutputStyle == DIPrinter::OutputStyle::GNU) {
+    // With ClPrintFunctions == FunctionNameKind::LinkageName (default)
+    // and ClUseSymbolTable == true (also default), Symbolizer.symbolizeCode()
+    // may override the name of an inlined function with the name of the topmost
+    // caller function in the inlining chain. This contradicts the existing
+    // behavior of addr2line. Symbolizer.symbolizeInlinedCode() overrides only
+    // the topmost function, which suits our needs better.
+    auto ResOrErr = Symbolizer.symbolizeInlinedCode(
+        ModuleName, {Offset, object::SectionedAddress::UndefSection},
+        ClDwpName);
+    Printer << (error(ResOrErr) ? DILineInfo() : ResOrErr.get().getFrame(0));
   } else {
     auto ResOrErr = Symbolizer.symbolizeCode(
         ModuleName, {Offset, object::SectionedAddress::UndefSection},
         ClDwpName);
     Printer << (error(ResOrErr) ? DILineInfo() : ResOrErr.get());
   }
+  if (ClOutputStyle == DIPrinter::OutputStyle::LLVM)
     outs() << "\n";
   outs().flush();
 }
 
 int main(int argc, char **argv) {
   InitLLVM X(argc, argv);
 
   llvm::sys::InitializeCOMRAII COM(llvm::sys::COMThreadingMode::MultiThreaded);
   cl::ParseCommandLineOptions(argc, argv, "llvm-symbolizer\n");