[DWARFYAML] Make the address size of compilation units optional.

[Higuoxing]

Summary

This patch makes the 'AddrSize' field optional. If the address size is
missing, yaml2obj will infer it from the object file.

[grimar]

Perhaps, you should be able to simplify some other test cases too with this change? I.e. to remove the 'AddrSize' field?

[Higuoxing]

In D85805#2212570, @grimar wrote:

Perhaps, you should be able to simplify some other test cases too with this change? I.e. to remove the 'AddrSize' field?

Thanks! I will do it. Shall I simplify them in this change?

[grimar]

In D85805#2212576, @Higuoxing wrote:

In D85805#2212570, @grimar wrote:

Perhaps, you should be able to simplify some other test cases too with this change? I.e. to remove the 'AddrSize' field?

Thanks! I will do it. Shall I simplify them in this change?

Yes, I think it is fine to do it in this change - when we introduce some functionality that simplifies things it is often reasonable to
apply it to the same patch to demostrate how much it is usable/useful.

[jhenderson]

Code change looks good. Thanks. I'll hold off approving until you've made any test changes too.

[Higuoxing]

Simplify some tests to get rid of 'AddrSize' fields.

[jhenderson]

LGTM.

llvm/include/llvm/ObjectYAML/DWARFYAML.h

 struct DWARFContext {
   bool IsGNUPubSec = false;
 };
 
 struct Unit {
   dwarf::DwarfFormat Format;
   Optional<yaml::Hex64> Length;
   uint16_t Version;
-  uint8_t AddrSize;
+  Optional<uint8_t> AddrSize;
   llvm::dwarf::UnitType Type; // Added in DWARF 5
   yaml::Hex64 AbbrOffset;
   std::vector<Entry> Entries;
 };
 
 struct File {
   StringRef Name;
   uint64_t DirIdx;

llvm/lib/ObjectYAML/DWARFEmitter.cpp

for (; FormVal != Entry.Values.end() && AbbrForm != Abbrev.Attributes.end();
     } while (Indirect);
   }
 
   return OS.tell() - EntryBegin;
 }
 
 Error DWARFYAML::emitDebugInfo(raw_ostream &OS, const DWARFYAML::Data &DI) {
   for (const DWARFYAML::Unit &Unit : DI.CompileUnits) {
-    dwarf::FormParams Params = {Unit.Version, Unit.AddrSize, Unit.Format};
+    uint8_t AddrSize;
+    if (Unit.AddrSize)
+      AddrSize = *Unit.AddrSize;
+    else
+      AddrSize = DI.Is64BitAddrSize ? 8 : 4;
+    dwarf::FormParams Params = {Unit.Version, AddrSize, Unit.Format};
     uint64_t Length = 3; // sizeof(version) + sizeof(address_size)
     Length += Unit.Version >= 5 ? 1 : 0;       // sizeof(unit_type)
     Length += Params.getDwarfOffsetByteSize(); // sizeof(debug_abbrev_offset)
 
     // Since the length of the current compilation unit is undetermined yet, we
     // firstly write the content of the compilation unit to a buffer to
     // calculate it and then serialize the buffer content to the actual output
     // stream.
for (const DWARFYAML::Unit &Unit : DI.CompileUnits) {
     // the actual length.
     if (Unit.Length)
       Length = *Unit.Length;
 
     writeInitialLength(Unit.Format, Length, OS, DI.IsLittleEndian);
     writeInteger((uint16_t)Unit.Version, OS, DI.IsLittleEndian);
     if (Unit.Version >= 5) {
       writeInteger((uint8_t)Unit.Type, OS, DI.IsLittleEndian);
-      writeInteger((uint8_t)Unit.AddrSize, OS, DI.IsLittleEndian);
+      writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian);
       writeDWARFOffset(Unit.AbbrOffset, Unit.Format, OS, DI.IsLittleEndian);
     } else {
       writeDWARFOffset(Unit.AbbrOffset, Unit.Format, OS, DI.IsLittleEndian);
-      writeInteger((uint8_t)Unit.AddrSize, OS, DI.IsLittleEndian);
+      writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian);
     }
 
     OS.write(EntryBuffer.data(), EntryBuffer.size());
   }
 
   return Error::success();
 }
 

llvm/lib/ObjectYAML/DWARFYAML.cpp

 
 void MappingTraits<DWARFYAML::Unit>::mapping(IO &IO, DWARFYAML::Unit &Unit) {
   IO.mapOptional("Format", Unit.Format, dwarf::DWARF32);
   IO.mapOptional("Length", Unit.Length);
   IO.mapRequired("Version", Unit.Version);
   if (Unit.Version >= 5)
     IO.mapRequired("UnitType", Unit.Type);
   IO.mapRequired("AbbrOffset", Unit.AbbrOffset);
-  IO.mapRequired("AddrSize", Unit.AddrSize);
+  IO.mapOptional("AddrSize", Unit.AddrSize);
   IO.mapOptional("Entries", Unit.Entries);
 }
 
 void MappingTraits<DWARFYAML::Entry>::mapping(IO &IO, DWARFYAML::Entry &Entry) {
   IO.mapRequired("AbbrCode", Entry.AbbrCode);
   IO.mapOptional("Values", Entry.Values);
 }
 

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

- Code:       1
       Tag:        DW_TAG_compile_unit
       Children:   DW_CHILDREN_no
       Attributes: []
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:    []
 
 ## e) Test that yaml2obj emits an error message when both the "Content" and the
 ## "debug_info" entry are specified at the same time.
 
 # RUN: not yaml2obj --docnum=5 %s 2>&1 | FileCheck %s --check-prefix=AMBIGUOUS-CONTENT
 
 --- !ELF
- Code:       1
       Tag:        DW_TAG_compile_unit
       Children:   DW_CHILDREN_no
       Attributes: []
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:    []
 
 ## f) Test that all the properties can be overridden by the section header when
 ## the "debug_info" entry doesn't exist.
 
 # RUN: yaml2obj --docnum=6 %s -o %t6.o
 # RUN: llvm-readelf --sections %t6.o | FileCheck %s --check-prefix=OVERRIDDEN
 
- Code:       1
       Tag:        DW_TAG_compile_unit
       Children:   DW_CHILDREN_no
       Attributes: []
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:
         - AbbrCode: 0
           Values:   []
 
 ## h) Test that yaml2obj doesn't crash when the 'Entries' of a compilation unit is empty.
 
 # RUN: yaml2obj --docnum=8 %s -o %t8.o
 # RUN: llvm-readelf --hex-dump=.debug_info %t8.o | \
 # RUN:   FileCheck %s --check-prefix=EMPTY-ENTRIES --match-full-lines
 
 #       EMPTY-ENTRIES: Hex dump of section '.debug_info':
 #  EMPTY-ENTRIES-NEXT: 0x00000000 34120000 05000204 34120000 4.......4...
 # EMPTY-ENTRIES-EMPTY:
 ##                                ^- 'Entries' is empty
 
 --- !ELF
 FileHeader:
-  Class:   ELFCLASS64
+  Class:   ELFCLASS32
   Data:    ELFDATA2LSB
   Type:    ET_EXEC
   Machine: EM_X86_64
 DWARF:
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:    []
 
 ## h) Test that yaml2obj emits values in the DIE according to the abbreviation whose
 ## index is equal to the value in 'AbbrevCode'.
 
 # RUN: yaml2obj --docnum=9 %s -o %t9.o
 # RUN: llvm-readelf --hex-dump=.debug_info %t9.o | \
 # RUN:   FileCheck %s --check-prefix=FORM --match-full-lines

 ##                             ^-                            abbreviation code (ULEB128)
 ##                                ^-------                   Form: DW_FORM_data4 (4-byte)
 ##                                         ^---              Form: DW_FORM_data2 (2-byte)
 ##                                             ^-            abbreviation code (ULEB128)
 # FORM-EMPTY:
 
 --- !ELF
 FileHeader:
-  Class:   ELFCLASS64
+  Class:   ELFCLASS32
   Data:    ELFDATA2LSB
   Type:    ET_EXEC
   Machine: EM_X86_64
 DWARF:
   debug_abbrev:
     - Tag:        DW_TAG_compile_unit
       Children:   DW_CHILDREN_yes
       Attributes:
         - Attribute: DW_AT_low_pc
           Form:      DW_FORM_data4
         - Attribute: DW_AT_high_pc
           Form:      DW_FORM_data2
     - Tag:        DW_TAG_subprogram
       Children:   DW_CHILDREN_no
       Attributes:
         - Attribute: DW_AT_low_pc
           Form:      DW_FORM_data2
         - Attribute: DW_AT_high_pc
           Form:      DW_FORM_data4
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:
         ## Test that yaml2obj emits values when the abbrev code is specified.
         - AbbrCode: 2
           Values:
             - Value: 0x1234
             - Value: 0x87654321
         ## Test that yaml2obj emits values when the abbrev code is specified to
         ## be lower than the first abbrev.
FileHeader:
   Type:    ET_EXEC
   Machine: EM_X86_64
 DWARF:
   debug_info:
     - Length:     0x1234
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   4
       Entries:
         - AbbrCode: 1
           Values:
             - Value: 0x1234
 
 ## j) Test that yaml2obj emits the correct DWARF64 unit headers.
 
 ## DWARFv5 unit header.
FileHeader:
   Machine: EM_X86_64
 DWARF:
   debug_info:
     - Format:     DWARF64
       Length:     0x0c
       Version:    5
       UnitType:   DW_UT_type
       AbbrOffset: 0x1234
-      AddrSize:   8
       Entries:    []
 
 ## DWARFv4 unit header.
 
 # RUN: yaml2obj --docnum=12 %s -o %t12.o
 # RUN: llvm-readelf --hex-dump=.debug_info %t12.o | \
 # RUN:   FileCheck %s --check-prefix=DWARFV4-HEADER
 
FileHeader:
   Type:    ET_EXEC
   Machine: EM_X86_64
 DWARF:
   debug_info:
     - Format:     DWARF64
       Length:     0x0c
       Version:    4
       AbbrOffset: 0x1234
-      AddrSize:   8
       Entries:    []
 
 ## k) Test that yaml2obj is able to emit a correct length for compilation units.
 
 # RUN: yaml2obj --docnum=13 %s -o %t13.o
 # RUN: llvm-readelf --hex-dump=.debug_info %t13.o | \
 # RUN:   FileCheck %s --check-prefix=INFER-LENGTH
 
- Code: 2
           Form:      DW_FORM_addr
         - Attribute: DW_AT_high_pc
           Form:      DW_FORM_data4
         - Attribute: DW_AT_name
           Form:      DW_FORM_strp
   debug_info:
     - Version:    4
       AbbrOffset: 0x00
-      AddrSize:   0x08
       Entries:
         - AbbrCode: 1
           Values:
             - Value: 0x00   ## DW_AT_producer [DW_FORM_strp]
             - Value: 0x0c   ## DW_LANG_C99, DW_AT_language [DW_FORM_data2]
             - Value: 0x16   ## DW_AT_name [DW_FORM_strp]
             - Value: 0x00   ## DW_AT_stmt_list [DW_FORM_sec_offset]
             - Value: 0x1e   ## DW_AT_comp_dir [DW_FORM_strp]
FileHeader:
   Class:   ELFCLASS64
   Data:    ELFDATA2LSB
   Type:    ET_EXEC
   Machine: EM_X86_64
 DWARF:
   debug_info:
     - Version:    4
       AbbrOffset: 0x00
-      AddrSize:   0x08
       Entries:
         - AbbrCode: 1
         - AbbrCode: 0
         - AbbrCode: 2
+
+## m) Test that yaml2obj is able to infer the address size from the object file.
+
+# RUN: yaml2obj --docnum=15 -DBITS=64 %s -o %t15.64-bit.o
+# RUN: llvm-readelf --hex-dump=.debug_info %t15.64-bit.o | \
+# RUN:   FileCheck %s --check-prefix=ADDRSIZE -DADDRSIZE=08
+
+#      ADDRSIZE: Hex dump of section '.debug_info':
+# ADDRSIZE-NEXT: 0x00000000 07000000 04000000 0000[[ADDRSIZE]]            ...........
+
+# RUN: yaml2obj --docnum=15 -DBITS=32 %s -o %t15.32-bit.o
+# RUN: llvm-readelf --hex-dump=.debug_info %t15.32-bit.o | \
+# RUN:   FileCheck %s --check-prefix=ADDRSIZE -DADDRSIZE=04
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS[[BITS]]
+  Data:    ELFDATA2LSB
+  Type:    ET_EXEC
+  Machine: EM_X86_64
+DWARF:
+  debug_info:
+    - Version:    4
+      AbbrOffset: 0x00