[PATCH] D37771: llvm-dwarfdump: automatically dump both regular and .dwo variant of sections

[Adrian Prantl via Phabricator via llvm-commits]

aprantl added inline comments.


================
Comment at: lib/DebugInfo/DWARF/DWARFContext.cpp:367-370
+    if (!DObj->getStringOffsetSection().Data.empty())
+      dumpStringOffsetsSection(
+          OS, "debug_str_offsets", *DObj, DObj->getStringOffsetSection(),
+          DObj->getStringSection(), isLittleEndian(), getMaxVersion());
----------------
dblaikie wrote:
> Should this condition be sunk into dumpStringOffsetsSection?
No, it shouldn't be there at all to mirror all the other cases.


================
Comment at: test/DebugInfo/X86/generate-odr-hash.ll:130
+; FISSION: .debug_line.dwo contents:
+; CHECK-NOT: .debug_line.dwo contents:
 ; FISSION: Line table prologue
----------------
dblaikie wrote:
> Also probably "CHECK-NOT: contents:"
No, it looks like there's another section in between in some of the FileCheck runs using the same CHECK prefix.


================
Comment at: test/DebugInfo/X86/live-debug-variables.ll:30
 ; CHECK-NOT: :
+; CHECK:      .debug_loc.dwo contents:
 
----------------
dblaikie wrote:
> Should we print the header if there's nothing in the section? I've know we do it for some sections and not for others - and kind of think maybe we shouldn't do it for any sections.
Currently we print the section header when the section was explicitly requested (either through --debug-loc or --all). We only print the .dwo section header when the non-dwo variant was requested and it is non-empty. My selfish thinking behind this decision was that we don't want the .dwo sections to show up at all on platforms that don't use DWOs.


https://reviews.llvm.org/D37771