[DWARFv5] llvm-mc -dwarf-version does not imply -g.

[probinson]

Summary

Don't provide the assembler source as the "root file" unless the user
asked to have debug info for the assembler source (with -g).

If the source doesn't provide an explicit ".file 0" then (a) use the
compilation directory as directory #0, and (b) use the file #1 info
for file #0 also.

[aprantl]

So llvm-mc performs a literal translation by default and with -g it generates debug info for the assembler input? Seems reasonable.

llvm/lib/MC/MCDwarf.cpp

392 ↗

(On Diff #150855)

Could this be a const StringRef &?

[probinson]

In D48055#1129175, @aprantl wrote:

So llvm-mc performs a literal translation by default and with -g it generates debug info for the assembler input? Seems reasonable.

Yes, that has been the behavior for some time, imitating the GNU assembler interface.
I stumbled over this while working on PR37623, and it was simpler to do this part first.

llvm/lib/MC/MCDwarf.cpp

392 ↗

(On Diff #150855)

Indeed it can!

llvm/trunk/include/llvm/MC/MCDwarf.h

Emit(MCStreamer *MCOS, MCDwarfLineTableParams Params,
        Optional<MCDwarfLineStr> &LineStr) const;
   std::pair<MCSymbol *, MCSymbol *>
   Emit(MCStreamer *MCOS, MCDwarfLineTableParams Params,
        ArrayRef<char> SpecialOpcodeLengths,
        Optional<MCDwarfLineStr> &LineStr) const;
 
 private:
   void emitV2FileDirTables(MCStreamer *MCOS) const;
-  void emitV5FileDirTables(MCStreamer *MCOS,
-                           Optional<MCDwarfLineStr> &LineStr) const;
+  void emitV5FileDirTables(MCStreamer *MCOS, Optional<MCDwarfLineStr> &LineStr,
+                           StringRef CtxCompilationDir) const;
 };
 
 class MCDwarfDwoLineTable {
   MCDwarfLineTableHeader Header;
 
 public:
   void maybeSetRootFile(StringRef Directory, StringRef FileName,
                         MD5::MD5Result *Checksum, Optional<StringRef> Source) {

llvm/trunk/lib/MC/MCDwarf.cpp

else {
       MCOS->EmitBytes(
           DwarfFile.Source.getValueOr(StringRef())); // Source and...
       MCOS->EmitBytes(StringRef("\0", 1));           // its null terminator.
     }
   }
 }
 
 void MCDwarfLineTableHeader::emitV5FileDirTables(
-    MCStreamer *MCOS, Optional<MCDwarfLineStr> &LineStr) const {
+    MCStreamer *MCOS, Optional<MCDwarfLineStr> &LineStr,
+    StringRef CtxCompilationDir) const {
   // The directory format, which is just a list of the directory paths.  In a
   // non-split object, these are references to .debug_line_str; in a split
   // object, they are inline strings.
   MCOS->EmitIntValue(1, 1);
   MCOS->EmitULEB128IntValue(dwarf::DW_LNCT_path);
   MCOS->EmitULEB128IntValue(LineStr ? dwarf::DW_FORM_line_strp
                                     : dwarf::DW_FORM_string);
   MCOS->EmitULEB128IntValue(MCDwarfDirs.size() + 1);
+  // Try not to emit an empty compilation directory.
+  const StringRef &CompDir =
+      CompilationDir.empty() ? CtxCompilationDir : CompilationDir;
   if (LineStr) {
     // Record path strings, emit references here.
-    LineStr->emitRef(MCOS, CompilationDir);
+    LineStr->emitRef(MCOS, CompDir);
     for (auto &Dir : MCDwarfDirs)
       LineStr->emitRef(MCOS, Dir);
   } else {
-    // The list of directory paths.  CompilationDir comes first.
-    MCOS->EmitBytes(CompilationDir);
+    // The list of directory paths.  Compilation directory comes first.
+    MCOS->EmitBytes(CompDir);
     MCOS->EmitBytes(StringRef("\0", 1));
     for (auto &Dir : MCDwarfDirs) {
       MCOS->EmitBytes(Dir);                // The DirectoryName, and...
       MCOS->EmitBytes(StringRef("\0", 1)); // its null terminator.
     }
   }
 
   // The file format, which is the inline null-terminated filename and a
if (HasMD5) {
     MCOS->EmitULEB128IntValue(dwarf::DW_FORM_data16);
   }
   if (HasSource) {
     MCOS->EmitULEB128IntValue(dwarf::DW_LNCT_LLVM_source);
     MCOS->EmitULEB128IntValue(LineStr ? dwarf::DW_FORM_line_strp
                                       : dwarf::DW_FORM_string);
   }
   // Then the counted list of files. The root file is file #0, then emit the
-  // files as provide by .file directives.
+  // files as provide by .file directives.  To accommodate assembler source
+  // written for DWARF v4 but trying to emit v5, if we didn't see a root file
+  // explicitly, replicate file #1.
   MCOS->EmitULEB128IntValue(MCDwarfFiles.size());
-  emitOneV5FileEntry(MCOS, RootFile, HasMD5, HasSource, LineStr);
+  emitOneV5FileEntry(MCOS, RootFile.Name.empty() ? MCDwarfFiles[1] : RootFile,
+                     HasMD5, HasSource, LineStr);
   for (unsigned i = 1; i < MCDwarfFiles.size(); ++i)
     emitOneV5FileEntry(MCOS, MCDwarfFiles[i], HasMD5, HasSource, LineStr);
 }
 
 std::pair<MCSymbol *, MCSymbol *>
 MCDwarfLineTableHeader::Emit(MCStreamer *MCOS, MCDwarfLineTableParams Params,
                              ArrayRef<char> StandardOpcodeLengths,
                              Optional<MCDwarfLineStr> &LineStr) const {
MCDwarfLineTableHeader::Emit(MCStreamer *MCOS, MCDwarfLineTableParams Params,
 
   // Standard opcode lengths
   for (char Length : StandardOpcodeLengths)
     MCOS->EmitIntValue(Length, 1);
 
   // Put out the directory and file tables.  The formats vary depending on
   // the version.
   if (LineTableVersion >= 5)
-    emitV5FileDirTables(MCOS, LineStr);
+    emitV5FileDirTables(MCOS, LineStr, context.getCompilationDir());
   else
     emitV2FileDirTables(MCOS);
 
   // This is the end of the prologue, so set the value of the symbol at the
   // end of the prologue (that was used in a previous expression).
   MCOS->EmitLabel(ProEndSym);
 
   return std::make_pair(LineStartSym, LineEndSym);

llvm/trunk/test/MC/ELF/debug-file-options.s

 
 # CHECK: debug_line[0x00000000]
 # CHECK: version: 5
 # CHECK: include_directories[ 0] = .debug_line_str[0x[[DIR0:[0-9a-f]+]]] = "{{.+}}"
 # CHECK: include_directories[ 1] = .debug_line_str[0x[[DIR1:[0-9a-f]+]]] = "dir1"
 # CHECK: include_directories[ 2] = .debug_line_str[0x[[DIR2:[0-9a-f]+]]] = "dir2"
 # CHECK-NOT: include_directories
 # CHECK: file_names[ 0]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "{{.+}}"
-# CHECK-NEXT: dir_index: 0
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: dir_index: 1
 # CHECK-NEXT: md5_checksum:
-# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC:[0-9a-f]+]]] = ""
+# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC:[0-9a-f]+]]] = "void foo() {}"
 # CHECK: file_names[ 1]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE1:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0]]] = "foo"
 # CHECK-NEXT: dir_index: 1
 # CHECK-NEXT: md5_checksum: ee87e05688663173cd6043a3a15bba6e
-# CHECK-NEXT: source: .debug_line_str[0x[[FILE1SRC:[0-9a-f]+]]] = "void foo() {}"
+# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC]]] = "void foo() {}"
 # CHECK: file_names[ 2]:
 # CHECK-NEXT: name: .debug_line_str[0x[[FILE2:[0-9a-f]+]]] = "bar"
 # CHECK-NEXT: dir_index: 2
 # CHECK-NEXT: md5_checksum: 816225a0c90ca8948b70eb58be4d522f
 # CHECK-NEXT: source: .debug_line_str[0x[[FILE2SRC:[0-9a-f]+]]] = "void bar() {}"
 
 # CHECK: .debug_line_str contents:
 # CHECK-NEXT: 0x[[DIR0]]: "{{.+}}"
 # CHECK-NEXT: 0x[[DIR1]]: "dir1"
 # CHECK-NEXT: 0x[[DIR2]]: "dir2"
-# CHECK-NEXT: 0x[[FILE0]]: "{{.+}}"
-# CHECK-NEXT: 0x[[FILE0SRC]]: ""
-# CHECK-NEXT: 0x[[FILE1]]: "foo"
-# CHECK-NEXT: 0x[[FILE1SRC]]: "void foo() {}"
+# CHECK-NEXT: 0x[[FILE0]]: "foo"
+# CHECK-NEXT: 0x[[FILE0SRC]]: "void foo() {}"
 # CHECK-NEXT: 0x[[FILE2]]: "bar"
 # CHECK-NEXT: 0x[[FILE2SRC]]: "void bar() {}"

llvm/trunk/test/MC/ELF/debug-md5.s

 // RUN: llvm-mc -triple x86_64-unknown-unknown -dwarf-version 5 -fdebug-compilation-dir=/tmp -filetype=obj %s -o - | llvm-dwarfdump --debug-line --debug-line-str -v - | FileCheck %s
 
         .file 1 "dir1/foo"   md5 0x00112233445566778899aabbccddeeff
         .file 2 "dir2" "bar" md5 0xffeeddccbbaa99887766554433221100
         .loc 1 1 0
         nop
         .loc 2 1 0
         nop
 
 # CHECK: debug_line[0x00000000]
 # CHECK: version: 5
 # CHECK: include_directories[ 0] = .debug_line_str[0x[[DIR0:[0-9a-f]+]]] = "/tmp"
 # CHECK: include_directories[ 1] = .debug_line_str[0x[[DIR1:[0-9a-f]+]]] = "dir1"
 # CHECK: include_directories[ 2] = .debug_line_str[0x[[DIR2:[0-9a-f]+]]] = "dir2"
 # CHECK-NOT: include_directories
 # CHECK: file_names[ 0]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "{{.+}}"
-# CHECK-NEXT: dir_index: 0
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: dir_index: 1
 # CHECK: file_names[ 1]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE1:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0]]] = "foo"
 # CHECK-NEXT: dir_index: 1
 # CHECK-NEXT: md5_checksum: 00112233445566778899aabbccddeeff
 # CHECK: file_names[ 2]:
 # CHECK-NEXT: name: .debug_line_str[0x[[FILE2:[0-9a-f]+]]] = "bar"
 # CHECK-NEXT: dir_index: 2
 # CHECK-NEXT: md5_checksum: ffeeddccbbaa99887766554433221100
 
 # CHECK: .debug_line_str contents:
 # CHECK-NEXT: 0x[[DIR0]]: "/tmp"
 # CHECK-NEXT: 0x[[DIR1]]: "dir1"
 # CHECK-NEXT: 0x[[DIR2]]: "dir2"
-# CHECK-NEXT: 0x[[FILE0]]: "{{.+}}"
-# CHECK-NEXT: 0x[[FILE1]]: "foo"
+# CHECK-NEXT: 0x[[FILE0]]: "foo"
 # CHECK-NEXT: 0x[[FILE2]]: "bar"

llvm/trunk/test/MC/ELF/debug-source.s

 // RUN: llvm-mc -triple x86_64-unknown-unknown -dwarf-version 5 -filetype=obj %s -o -| llvm-dwarfdump --debug-line --debug-line-str -v - | FileCheck %s
 
         .file 1 "dir1/foo"   source "void foo() {}"
         .file 2 "dir2" "bar" source "void bar()\n{\n}"
         .loc 1 1 0
         nop
         .loc 2 1 0
         nop
 
 # CHECK: debug_line[0x00000000]
 # CHECK: version: 5
 # CHECK: include_directories[ 0] = .debug_line_str[0x[[DIR0:[0-9a-f]+]]] = "{{.+}}"
 # CHECK: include_directories[ 1] = .debug_line_str[0x[[DIR1:[0-9a-f]+]]] = "dir1"
 # CHECK: include_directories[ 2] = .debug_line_str[0x[[DIR2:[0-9a-f]+]]] = "dir2"
 # CHECK-NOT: include_directories
 # CHECK: file_names[ 0]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "{{.+}}"
-# CHECK-NEXT: dir_index: 0
-# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC:[0-9a-f]+]]] = ""
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: dir_index: 1
+# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC:[0-9a-f]+]]] = "void foo() {}"
 # CHECK: file_names[ 1]:
-# CHECK-NEXT: name: .debug_line_str[0x[[FILE1:[0-9a-f]+]]] = "foo"
+# CHECK-NEXT: name: .debug_line_str[0x[[FILE0]]] = "foo"
 # CHECK-NEXT: dir_index: 1
-# CHECK-NEXT: source: .debug_line_str[0x[[FILE1SRC:[0-9a-f]+]]] = "void foo() {}"
+# CHECK-NEXT: source: .debug_line_str[0x[[FILE0SRC]]] = "void foo() {}"
 # CHECK: file_names[ 2]:
 # CHECK-NEXT: name: .debug_line_str[0x[[FILE2:[0-9a-f]+]]] = "bar"
 # CHECK-NEXT: dir_index: 2
 # CHECK-NEXT: source: .debug_line_str[0x[[FILE2SRC:[0-9a-f]+]]] = "void bar()\n{\n}"
 
 # CHECK: .debug_line_str contents:
 # CHECK-NEXT: 0x[[DIR0]]: "{{.+}}"
 # CHECK-NEXT: 0x[[DIR1]]: "dir1"
 # CHECK-NEXT: 0x[[DIR2]]: "dir2"
-# CHECK-NEXT: 0x[[FILE0]]: "{{.+}}"
-# CHECK-NEXT: 0x[[FILE0SRC]]: ""
-# CHECK-NEXT: 0x[[FILE1]]: "foo"
-# CHECK-NEXT: 0x[[FILE1SRC]]: "void foo() {}"
+# CHECK-NEXT: 0x[[FILE0]]: "foo"
+# CHECK-NEXT: 0x[[FILE0SRC]]: "void foo() {}"
 # CHECK-NEXT: 0x[[FILE2]]: "bar"
 # CHECK-NEXT: 0x[[FILE2SRC]]: "void bar()\n{\n}"

llvm/trunk/test/MC/ELF/dwarf-file0.s

 # RUN: llvm-mc -dwarf-version 4 %s -filetype=obj -o - | llvm-dwarfdump -debug-line - | FileCheck %s --check-prefixes=CHECK,CHECK-4
 # RUN: llvm-mc -dwarf-version 5 %s -filetype=obj -o - | llvm-dwarfdump -debug-line - | FileCheck %s --check-prefixes=CHECK,CHECK-5
 # REQUIRES: default_triple
 # Darwin is stuck on DWARF v2.
 # XFAIL: darwin
         .file 0 "root.cpp"
         .file 1 "header.h"
         .file 2 "root.cpp"
-# CHECK-5:     include_directories[ 0] = ""
+# CHECK-5:     include_directories[ 0] = "{{.+}}"
 # CHECK-4-NOT: include_directories
 # CHECK-4-NOT: file_names[ 0]
 # CHECK-5:     file_names[ 0]:
 # CHECK-5-NEXT: name: "root.cpp"
 # CHECK-5-NEXT: dir_index: 0
 # CHECK:       file_names[ 1]:
 # CHECK-NEXT:  name: "header.h"
 # CHECK-NEXT:  dir_index: 0
 # CHECK:       file_names[ 2]:
 # CHECK-NEXT:  name: "root.cpp"
 # CHECK-NEXT:  dir_index: 0

llvm/trunk/tools/llvm-mc/llvm-mc.cpp

int main(int argc, char **argv) {
   else {
     // If no compilation dir is set, try to use the current directory.
     SmallString<128> CWD;
     if (!sys::fs::current_path(CWD))
       Ctx.setCompilationDir(CWD);
   }
   if (!MainFileName.empty())
     Ctx.setMainFileName(MainFileName);
-  if (DwarfVersion >= 5) {
+  if (GenDwarfForAssembly && DwarfVersion >= 5) {
     // DWARF v5 needs the root file as well as the compilation directory.
     // If we find a '.file 0' directive that will supersede these values.
     MD5 Hash;
     MD5::MD5Result *Cksum =
         (MD5::MD5Result *)Ctx.allocate(sizeof(MD5::MD5Result), 1);
     Hash.update(Buffer->getBuffer());
     Hash.final(*Cksum);
     Ctx.setMCLineTableRootFile(