[DWARFYAML] Make the debug_str section optional.

[Higuoxing]

Summary

This patch makes the debug_str section optional. When the debug_str
section exists but doesn't contain anything, yaml2obj will emit a
section header for it.

[grimar]

llvm/lib/ObjectYAML/DWARFEmitter.cpp

89

I think you don't really need this assert, because Optional<> implementation is based on
OptionalStorage which has it already:

T &getValue() LLVM_LVALUE_FUNCTION noexcept {
  assert(hasVal);
  return value;
}

90

I'd avoid using auto.

llvm/test/tools/yaml2obj/ELF/DWARF/debug-str.yaml

35

Seems -DSIZE is unused? (you are not using SHDRS-DEFAULT here).

llvm/tools/obj2yaml/dwarf2yaml.cpp

51

You can do something like:

Y.DebugStrings.emplace();

55

And then just

Y.DebugStrings->push_back(SymbolPair.first);

[Higuoxing]

Address comments.

llvm/lib/ObjectYAML/DWARFEmitter.cpp

89

Thanks. I use assert() in other places as well. I will propose a change to remove them.

[grimar]

llvm/test/tools/yaml2obj/ELF/DWARF/debug-str.yaml

221

It is a bit strange to see that the .debug_str section emitted has zero size?
I believe that ELF string tables are expected to have at least one null byte
(https://refspecs.linuxbase.org/elf/gabi4+/ch4.strtab.html)

With that we have an empty string at the offset 0 and a valid null-terminated string table.

It is probably unrelated to this patch though.

[Higuoxing]

This comment was removed by Higuoxing.

[Higuoxing]

Thanks for reviewing! Shall I add a test case that the debug_str string table has an empty string?

llvm/test/tools/yaml2obj/ELF/DWARF/debug-str.yaml

221

Sorry, my comments is deleted by an accident.

I think having a debug_str section who has zero size is fine here.

The debug_str string table doesn't start with an empty string. I think having an empty string might not apply here.

> llvm-dwarfdump --debug-str a.test 
a.test:        file format ELF64-x86-64

.debug_str contents:
0x00000000: "clang version 10.0.1 "
0x00000016: "/home/v/x/llvm/playground/a.c"
0x00000034: "/home/v/x/llvm/playground"
0x0000004e: "main"

Besides, we are able to create a debug_str string table with an empty string with the following YAML description.

debug_str:
  - '' ## Empty string.

[grimar]

In D86860#2247281, @Higuoxing wrote:

Shall I add a test case that the debug_str string table has an empty string?

I think it would be nice to have.

[Higuoxing]

Add test case (c).

[grimar]

LGTM, thanks!

[Higuoxing]

Rebase.

llvm/include/llvm/ObjectYAML/DWARFYAML.h

template <typename EntryType> struct ListTable {
   Optional<std::vector<yaml::Hex64>> Offsets;
   std::vector<ListEntries<EntryType>> Lists;
 };
 
 struct Data {
   bool IsLittleEndian;
   bool Is64BitAddrSize;
   std::vector<AbbrevTable> DebugAbbrev;
-  std::vector<StringRef> DebugStrings;
+  Optional<std::vector<StringRef>> DebugStrings;
   Optional<std::vector<StringOffsetsTable>> DebugStrOffsets;
   Optional<std::vector<ARange>> DebugAranges;
   std::vector<Ranges> DebugRanges;
   std::vector<AddrTableEntry> DebugAddr;
   Optional<PubSection> PubNames;
   Optional<PubSection> PubTypes;
 
   Optional<PubSection> GNUPubNames;

llvm/lib/ObjectYAML/DWARFEmitter.cpp

 
 static void writeDWARFOffset(uint64_t Offset, dwarf::DwarfFormat Format,
                              raw_ostream &OS, bool IsLittleEndian) {
   cantFail(writeVariableSizedInteger(Offset, Format == dwarf::DWARF64 ? 8 : 4,
                                      OS, IsLittleEndian));
 }
 
 Error DWARFYAML::emitDebugStr(raw_ostream &OS, const DWARFYAML::Data &DI) {
-  for (auto Str : DI.DebugStrings) {
+  for (StringRef Str : *DI.DebugStrings) {
     OS.write(Str.data(), Str.size());
     OS.write('\0');
   }
 
   return Error::success();
 }
 
 StringRef DWARFYAML::Data::getAbbrevTableContentByIndex(uint64_t Index) const {
   assert(Index < DebugAbbrev.size() &&

llvm/lib/ObjectYAML/DWARFYAML.cpp

 namespace llvm {
 
 bool DWARFYAML::Data::isEmpty() const {
   return getNonEmptySectionNames().empty();
 }
 
 SetVector<StringRef> DWARFYAML::Data::getNonEmptySectionNames() const {
   SetVector<StringRef> SecNames;
-  if (!DebugStrings.empty())
+  if (DebugStrings)
     SecNames.insert("debug_str");
   if (DebugAranges)
     SecNames.insert("debug_aranges");
   if (!DebugRanges.empty())
     SecNames.insert("debug_ranges");
   if (!DebugLines.empty())
     SecNames.insert("debug_line");
   if (!DebugAddr.empty())

llvm/test/ObjectYAML/MachO/DWARF-debug_str.yaml

-# RUN: yaml2obj %s | obj2yaml | FileCheck %s
+## a) Test that obj2yaml is able to dump the __debug_str section.
+
+# RUN: yaml2obj --docnum=1 %s | obj2yaml | FileCheck %s
 
 --- !mach-o
 FileHeader:      
   magic:           0xFEEDFACF
   cputype:         0x01000007
   cpusubtype:      0x00000003
   filetype:        0x0000000A
   ncmds:           6

 #CHECK:     - hello_world.c
 #CHECK:     - '/Users/cbieneman/dev/open-source/llvm-build-rel'
 #CHECK:     - main
 #CHECK:     - argc
 #CHECK:     - argv
 #CHECK:     - int
 #CHECK:     - char
 
+## b) Test dumping an empty __debug_str section.
+
+# RUN: yaml2obj --docnum=2 -DCONTENT='[]' %s | obj2yaml | FileCheck %s --check-prefix=EMPTY-CONTENT
+
+#      EMPTY-CONTENT: Sections:
+# EMPTY-CONTENT-NEXT:   - sectname:  __debug_str
+# EMPTY-CONTENT-NEXT:     segname:   __DWARF
+# EMPTY-CONTENT-NEXT:     addr:      0x0000000000000000
+# EMPTY-CONTENT-NEXT:     size:      0
+# EMPTY-CONTENT-NEXT:     offset:    0x00000210
+# EMPTY-CONTENT-NEXT:     align:     0
+# EMPTY-CONTENT-NEXT:     reloff:    0x00000000
+# EMPTY-CONTENT-NEXT:     nreloc:    0
+# EMPTY-CONTENT-NEXT:     flags:     0x00000000
+# EMPTY-CONTENT-NEXT:     reserved1: 0x00000000
+# EMPTY-CONTENT-NEXT:     reserved2: 0x00000000
+# EMPTY-CONTENT-NEXT:     reserved3: 0x00000000
+# EMPTY-CONTENT-NEXT: DWARF:
+# EMPTY-CONTENT-NEXT:   debug_str: []
+
+--- !mach-o
+FileHeader:
+  magic:      0xFEEDFACF
+  cputype:    0x01000007
+  cpusubtype: 0x00000003
+  filetype:   0x0000000A
+  ncmds:      1
+  sizeofcmds: 232
+  flags:      0x00000000
+  reserved:   0x00000000
+LoadCommands:
+  - cmd:      LC_SEGMENT_64
+    cmdsize:  152
+    segname:  __DWARF
+    vmaddr:   0x00
+    vmsize:   0x00
+    fileoff:  0x00
+    filesize: 0x00
+    maxprot:  0
+    initprot: 0
+    nsects:   1
+    flags:    0
+    Sections:
+      - sectname:  __debug_str
+        segname:   __DWARF
+        addr:      0x00
+        size:      [[SIZE=0]]
+        offset:    0x210
+        align:     0
+        reloff:    0x00000000
+        nreloc:    0
+        flags:     0x00000000
+        reserved1: 0x00000000
+        reserved2: 0x00000000
+        reserved3: 0x00000000
+DWARF:
+  debug_str:
+    [[CONTENT]]
+
+## c) Test generating and dumping a __debug_str section who only has an empty string.
+
+# RUN: yaml2obj --docnum=2 -DCONTENT='[ "" ]' -DSIZE=1 %s | obj2yaml | FileCheck %s --check-prefix=EMPTY-STRING
+
+#      EMPTY-STRING: DWARF:
+# EMPTY-STRING-NEXT:   debug_str:
+# EMPTY-STRING-NEXT:     - ''
+# EMPTY-STRING-NEXT: ...

llvm/test/tools/yaml2obj/ELF/DWARF/debug-str.yaml

 ## Test that yaml2obj emits .debug_str section.
 
 ## a) Generate the .debug_str section from the "DWARF" entry.
 
 # RUN: yaml2obj --docnum=1 %s -o %t1.o
 # RUN: llvm-readelf --string-dump=.debug_str %t1.o | FileCheck %s --check-prefix=DWARF-DEFAULT
 
 #      DWARF-DEFAULT: String dump of section '.debug_str':
 # DWARF-DEFAULT-NEXT: [     0] a
 # DWARF-DEFAULT-NEXT: [     2] b
 # DWARF-DEFAULT-NEXT: [     4] c
 
 ## Check the default sh_type, sh_entsize, sh_info, sh_flags and sh_addralign of the
 ## .debug_str section header.
 
-# RUN: llvm-readelf -S %t1.o | FileCheck %s --check-prefix=SHDRS-DEFAULT
+# RUN: llvm-readelf -S %t1.o | FileCheck -DSIZE=000006 %s --check-prefix=SHDRS-DEFAULT
 
-# SHDRS-DEFAULT: Name       Type     Address          Off    Size   ES Flg Lk Inf Al
-# SHDRS-DEFAULT: .debug_str PROGBITS 0000000000000000 000040 000006 01 MS  0  0   1
+# SHDRS-DEFAULT: Name       Type     Address          Off    Size     ES Flg Lk Inf Al
+# SHDRS-DEFAULT: .debug_str PROGBITS 0000000000000000 000040 [[SIZE]] 01 MS  0  0   1
 
 --- !ELF
 FileHeader:
   Class: ELFCLASS64
   Data:  ELFDATA2LSB
   Type:  ET_EXEC
 DWARF:
   debug_str:
     - a
     - b
     - c
 
 ## b) Generate the .debug_str section from the raw section content.
 
 # RUN: yaml2obj --docnum=2 %s -o %t2.o
 # RUN: llvm-readelf --string-dump=.debug_str %t2.o | FileCheck %s --check-prefix=DWARF-DEFAULT
 # RUN: llvm-readelf -S %t2.o | FileCheck %s --check-prefix=SHDRS
 
 # SHDRS: Name       Type     Address          Off    Size   ES Flg Lk Inf Al
 # SHDRS: .debug_str PROGBITS 0000000000000000 000040 000006 01 MS  0  0   0
 
 --- !ELF
 FileHeader:
   Class: ELFCLASS64
Sections:
   - Name: .debug_str
     Type: SHT_DYNAMIC
     Entries:
       - Tag:   DT_NULL
         Value: 0
 DWARF:
   debug_str:
     - a
+
+## i) Test that the .debug_str section header is emitted if the "debug_str"
+## entry is empty.
+
+# RUN: yaml2obj --docnum=10 %s -o %t10.o
+# RUN: llvm-readelf -S %t10.o | FileCheck -DSIZE=000000 %s --check-prefix=SHDRS-DEFAULT
+
+--- !ELF
+FileHeader:
+  Class: ELFCLASS64
+  Data:  ELFDATA2LSB
+  Type:  ET_EXEC
+DWARF:
+  debug_str: []

llvm/tools/obj2yaml/dwarf2yaml.cpp

for (auto AbbrvDeclSet : *AbbrevSetPtr) {
         Y.DebugAbbrev.back().Table.push_back(Abbrv);
       }
     }
   }
 }
 
 void dumpDebugStrings(DWARFContext &DCtx, DWARFYAML::Data &Y) {
   StringRef RemainingTable = DCtx.getDWARFObj().getStrSection();
+  Y.DebugStrings.emplace();
   while (RemainingTable.size() > 0) {
     auto SymbolPair = RemainingTable.split('\0');
     RemainingTable = SymbolPair.second;
-    Y.DebugStrings.push_back(SymbolPair.first);
+    Y.DebugStrings->push_back(SymbolPair.first);
   }
 }
 
 Error dumpDebugARanges(DWARFContext &DCtx, DWARFYAML::Data &Y) {
   DWARFDataExtractor ArangesData(DCtx.getDWARFObj().getArangesSection(),
                                  DCtx.isLittleEndian(), 0);
   uint64_t Offset = 0;
   DWARFDebugArangeSet Set;