[MC][DebugInfo] Consider the target when deciding what's a path separator

This patch teaches MCContext and DWARF linetable headers about path
separators, or at least lets them store them, so that we can switch out
the path separator at runtime. The use-case for this is targets where the
development environment isn't the same as the compilers native environment,
which can lead to differences in the computed DWARF directory tables in
.debug_line.

Author: jmorse

llvm/include/llvm/MC/MCContext.h

@@ -335,6 +335,8 @@ class MCContext {
 
   MCTargetOptions const *TargetOptions;
 
+  llvm::sys::path::Style PathStyle = llvm::sys::path::Style::native;
+
   bool HadError = false;
 
   void reportCommon(SMLoc Loc,
@@ -721,14 +723,16 @@ class MCContext {
 
   void setDwarfCompileUnitID(unsigned CUIndex) { DwarfCompileUnitID = CUIndex; }
 
+  llvm::sys::path::Style getPathStyle() const { return PathStyle; }
+
   /// Specifies the "root" file and directory of the compilation unit.
   /// These are "file 0" and "directory 0" in DWARF v5.
   void setMCLineTableRootFile(unsigned CUID, StringRef CompilationDir,
                               StringRef Filename,
                               std::optional<MD5::MD5Result> Checksum,
                               std::optional<StringRef> Source) {
     getMCDwarfLineTable(CUID).setRootFile(CompilationDir, Filename, Checksum,
-                                          Source);
+                                          Source, PathStyle);
   }
 
   /// Reports whether MD5 checksum usage is consistent (all-or-none).

llvm/include/llvm/MC/MCDwarf.h

@@ -21,6 +21,7 @@
 #include "llvm/MC/StringTableBuilder.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/MD5.h"
+#include "llvm/Support/Path.h"
 #include "llvm/Support/SMLoc.h"
 #include "llvm/Support/StringSaver.h"
 #include <cassert>
@@ -277,6 +278,7 @@ struct MCDwarfLineTableHeader {
   StringMap<unsigned> SourceIdMap;
   std::string CompilationDir;
   MCDwarfFile RootFile;
+  llvm::sys::path::Style PathStyle = llvm::sys::path::Style::native;
   bool HasAnySource = false;
 
 private:
@@ -311,7 +313,9 @@ struct MCDwarfLineTableHeader {
 
   void setRootFile(StringRef Directory, StringRef FileName,
                    std::optional<MD5::MD5Result> Checksum,
-                   std::optional<StringRef> Source) {
+                   std::optional<StringRef> Source,
+                   llvm::sys::path::Style S = llvm::sys::path::Style::native) {
+    PathStyle = S;
     CompilationDir = std::string(Directory);
     RootFile.Name = std::string(FileName);
     RootFile.DirIndex = 0;
@@ -342,10 +346,11 @@ class MCDwarfDwoLineTable {
 public:
   void maybeSetRootFile(StringRef Directory, StringRef FileName,
                         std::optional<MD5::MD5Result> Checksum,
-                        std::optional<StringRef> Source) {
+                        std::optional<StringRef> Source,
+                        llvm::sys::path::Style S) {
     if (!Header.RootFile.Name.empty())
       return;
-    Header.setRootFile(Directory, FileName, Checksum, Source);
+    Header.setRootFile(Directory, FileName, Checksum, Source, S);
   }
 
   unsigned getFile(StringRef Directory, StringRef FileName,
@@ -394,7 +399,9 @@ class MCDwarfLineTable {
 
   void setRootFile(StringRef Directory, StringRef FileName,
                    std::optional<MD5::MD5Result> Checksum,
-                   std::optional<StringRef> Source) {
+                   std::optional<StringRef> Source,
+                   llvm::sys::path::Style S) {
+    Header.PathStyle = S;
     Header.CompilationDir = std::string(Directory);
     Header.RootFile.Name = std::string(FileName);
     Header.RootFile.DirIndex = 0;

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

@@ -3429,7 +3429,8 @@ MCDwarfDwoLineTable *DwarfDebug::getDwoLineTable(const DwarfCompileUnit &CU) {
   const DICompileUnit *DIUnit = CU.getCUNode();
   SplitTypeUnitFileTable.maybeSetRootFile(
       DIUnit->getDirectory(), DIUnit->getFilename(),
-      getMD5AsBytes(DIUnit->getFile()), DIUnit->getSource());
+      getMD5AsBytes(DIUnit->getFile()), DIUnit->getSource(),
+      Asm->OutStreamer->getContext().getPathStyle());
   return &SplitTypeUnitFileTable;
 }
 

llvm/lib/MC/MCContext.cpp

@@ -77,6 +77,9 @@ MCContext::MCContext(const Triple &TheTriple, const MCAsmInfo *mai,
   SaveTempLabels = TargetOptions && TargetOptions->MCSaveTempLabels;
   SecureLogFile = TargetOptions ? TargetOptions->AsSecureLogFile : "";
 
+  if (TheTriple.isPS())
+    PathStyle = llvm::sys::path::Style::windows;
+
   if (SrcMgr && SrcMgr->getNumBuffers())
     MainFileName = std::string(SrcMgr->getMemoryBuffer(SrcMgr->getMainFileID())
                                    ->getBufferIdentifier());
@@ -970,12 +973,12 @@ void MCContext::setGenDwarfRootFile(StringRef InputFileName, StringRef Buffer) {
   if (FileNameBuf.empty() || FileNameBuf == "-")
     FileNameBuf = "<stdin>";
   if (!getMainFileName().empty() && FileNameBuf != getMainFileName()) {
-    llvm::sys::path::remove_filename(FileNameBuf);
-    llvm::sys::path::append(FileNameBuf, getMainFileName());
+    llvm::sys::path::remove_filename(FileNameBuf, PathStyle);
+    llvm::sys::path::append(FileNameBuf, PathStyle, getMainFileName());
   }
   StringRef FileName = FileNameBuf;
   if (FileName.consume_front(getCompilationDir()))
-    if (llvm::sys::path::is_separator(FileName.front()))
+    if (llvm::sys::path::is_separator(FileName.front(), PathStyle))
       FileName = FileName.drop_front();
   assert(!FileName.empty());
   setMCLineTableRootFile(

llvm/lib/MC/MCDwarf.cpp

@@ -664,9 +664,9 @@ MCDwarfLineTableHeader::tryGetFile(StringRef &Directory, StringRef &FileName,
 
   if (Directory.empty()) {
     // Separate the directory part from the basename of the FileName.
-    StringRef tFileName = sys::path::filename(FileName);
+    StringRef tFileName = sys::path::filename(FileName, PathStyle);
     if (!tFileName.empty()) {
-      Directory = sys::path::parent_path(FileName);
+      Directory = sys::path::parent_path(FileName, PathStyle);
       if (!Directory.empty())
         FileName = tFileName;
     }
@@ -939,7 +939,8 @@ static void EmitGenDwarfAranges(MCStreamer *MCOS,
 static void EmitGenDwarfInfo(MCStreamer *MCOS,
                              const MCSymbol *AbbrevSectionSymbol,
                              const MCSymbol *LineSectionSymbol,
-                             const MCSymbol *RangesSymbol) {
+                             const MCSymbol *RangesSymbol,
+                             llvm::sys::path::Style Style) {
   MCContext &context = MCOS->getContext();
 
   MCOS->switchSection(context.getObjectFileInfo()->getDwarfInfoSection());
@@ -1037,7 +1038,7 @@ static void EmitGenDwarfInfo(MCStreamer *MCOS,
   const SmallVectorImpl<std::string> &MCDwarfDirs = context.getMCDwarfDirs();
   if (MCDwarfDirs.size() > 0) {
     MCOS->emitBytes(MCDwarfDirs[0]);
-    MCOS->emitBytes(sys::path::get_separator());
+    MCOS->emitBytes(sys::path::get_separator(Style));
   }
   const SmallVectorImpl<MCDwarfFile> &MCDwarfFiles = context.getMCDwarfFiles();
   // MCDwarfFiles might be empty if we have an empty source file.
@@ -1225,7 +1226,8 @@ void MCGenDwarfInfo::Emit(MCStreamer *MCOS) {
   EmitGenDwarfAbbrev(MCOS);
 
   // Output the data for .debug_info section.
-  EmitGenDwarfInfo(MCOS, AbbrevSectionSymbol, LineSectionSymbol, RangesSymbol);
+  EmitGenDwarfInfo(MCOS, AbbrevSectionSymbol, LineSectionSymbol, RangesSymbol,
+                   context.getPathStyle());
 }
 
 //

llvm/test/DebugInfo/dir-table-path-separators.ll

@@ -0,0 +1,73 @@
+; RUN: llc %s -o - -filetype=obj -mtriple x86_64-pc-linux-gnu | llvm-dwarfdump - --debug-line | FileCheck %s --check-prefix=LINUX
+; RUN: llc %s -o - -filetype=obj -mtriple x86_64-sie-ps5 | llvm-dwarfdump - --debug-line | FileCheck %s --check-prefix=PS5
+;
+; UNSUPPORTED: system-windows
+;
+; Check that the DWARF-printing MC backend is willing to consider Windows '\'
+; characters as path separators so that it can build the directory index table.
+; On Linux, the Windows path separators below would been seen as part of the
+; filename, and so wouldn't be combined into a directory entry. Wheras on
+; Windows (or a target masquerading as Windows) they should be combined into a
+; "foo\bar" directory.
+;
+; LINUX: include_directories[  0] = "C:\\foobar"
+; LINUX: file_names[  0]:
+; LINUX:           name: "foo\\bar\\test.cpp"
+; LINUX:      dir_index: 0
+; LINUX: file_names[  1]:
+; LINUX:           name: "foo\\bar\\bar.cpp"
+; LINUX:      dir_index: 0
+; LINUX: file_names[  2]:
+; LINUX:           name: "foo\\bar\\baz.cpp"
+; LINUX:      dir_index: 0
+;
+; PS5:      include_directories[  0] = "C:\\foobar"
+; PS5-NEXT: include_directories[  1] = "foo\\bar"
+; PS5-NEXT: file_names[  0]:
+; PS5-NEXT:           name: "foo\\bar\\test.cpp"
+; PS5-NEXT:      dir_index: 0
+; PS5-NEXT: file_names[  1]:
+; PS5-NEXT:           name: "bar.cpp"
+; PS5-NEXT:      dir_index: 1
+; PS5-NEXT: file_names[  2]:
+; PS5-NEXT:           name: "baz.cpp"
+; PS5-NEXT:      dir_index: 1
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+
+define dso_local noundef i32 @_Z3foov() local_unnamed_addr !dbg !9 {
+  ret i32 0, !dbg !14
+}
+
+define dso_local noundef i32 @_Z3barv() local_unnamed_addr !dbg !15 {
+  ret i32 0, !dbg !17
+}
+
+define dso_local noundef i32 @main() local_unnamed_addr !dbg !18 {
+  ret i32 0, !dbg !19
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3, !4, !5, !6, !7}
+!llvm.ident = !{!8}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "foo\\bar\\test.cpp", directory: "C:\\foobar")
+!2 = !{i32 7, !"Dwarf Version", i32 5}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = !{i32 1, !"wchar_size", i32 4}
+!5 = !{i32 8, !"PIC Level", i32 2}
+!6 = !{i32 7, !"PIE Level", i32 2}
+!7 = !{i32 7, !"uwtable", i32 2}
+!8 = !{!"clang"}
+!9 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !10, file: !10, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!10 = !DIFile(filename: "foo\\bar\\bar.cpp", directory: "C:\\foobar")
+!11 = !DISubroutineType(types: !12)
+!12 = !{!13}
+!13 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!14 = !DILocation(line: 2, column: 3, scope: !9)
+!15 = distinct !DISubprogram(name: "bar", linkageName: "_Z3barv", scope: !16, file: !16, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!16 = !DIFile(filename: "foo\\bar\\baz.cpp", directory: "C:\\foobar")
+!17 = !DILocation(line: 2, column: 3, scope: !15)
+!18 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 3, type: !11, scopeLine: 3, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!19 = !DILocation(line: 4, column: 3, scope: !18)