Give same-named members unique timestamps on Darwin in llvm-ar.

[jyknight]

Summary

This change ensures that the (membername,timestamp) tuple uniquely
identifies an entry in an archive for format=darwin, in deterministic
mode (which is the default).

That, then, enables lldb and dsymutil to locate the appropriate object
within the archive.

[JDevlieghere]

llvm/lib/Object/ArchiveWriter.cpp

467

Spurious space in lldb path.

473

You could omit the braces as per the style guide.

501

I wonder if we should include the original timestamp in the calculation? I don't know if it's worth the extra work of keeping track of these artificial ones, as we'd need to ensure they don't crash with real timestamps.

[jyknight]

llvm/lib/Object/ArchiveWriter.cpp

501

This code is only used in Deterministic mode, so there are no original timestamps to worry about.

[JDevlieghere]

LGTM, Thanks!

[jyknight]

Just noticed I forgot to commit this. Will do so now. :)

llvm/lib/Object/ArchiveWriter.cpp

 #include "llvm/Support/EndianStream.h"
 #include "llvm/Support/Errc.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/ToolOutputFile.h"
 #include "llvm/Support/raw_ostream.h"
 
+#include <map>
+
 #if !defined(_MSC_VER) && !defined(__MINGW32__)
 #include <unistd.h>
 #else
 #include <io.h>
 #endif
 
 using namespace llvm;
 
if (Thin) {
     Out << M.MemberName;
   Out << "/\n";
 }
 
 static void printMemberHeader(raw_ostream &Out, uint64_t Pos,
                               raw_ostream &StringTable,
                               object::Archive::Kind Kind, bool Thin,
                               StringRef ArcName, const NewArchiveMember &M,
+                              sys::TimePoint<std::chrono::seconds> ModTime,
                               unsigned Size) {
+
   if (isBSDLike(Kind))
-    return printBSDMemberHeader(Out, Pos, M.MemberName, M.ModTime, M.UID, M.GID,
+    return printBSDMemberHeader(Out, Pos, M.MemberName, ModTime, M.UID, M.GID,
                                 M.Perms, Size);
   if (!useStringTable(Thin, M.MemberName))
-    return printGNUSmallMemberHeader(Out, M.MemberName, M.ModTime, M.UID, M.GID,
+    return printGNUSmallMemberHeader(Out, M.MemberName, ModTime, M.UID, M.GID,
                                      M.Perms, Size);
   Out << '/';
   uint64_t NamePos = StringTable.tell();
   addToStringTable(StringTable, ArcName, M, Thin);
   printWithSpacePadding(Out, NamePos, 15);
-  printRestOfMemberHeader(Out, M.ModTime, M.UID, M.GID, M.Perms, Size);
+  printRestOfMemberHeader(Out, ModTime, M.UID, M.GID, M.Perms, Size);
 }
 
 namespace {
 struct MemberData {
   std::vector<unsigned> Symbols;
   std::string Header;
   StringRef Data;
   StringRef Padding;
for (const object::BasicSymbolRef &S : Obj.symbols()) {
     SymNames << '\0';
   }
   return Ret;
 }
 
 static Expected<std::vector<MemberData>>
 computeMemberData(raw_ostream &StringTable, raw_ostream &SymNames,
                   object::Archive::Kind Kind, bool Thin, StringRef ArcName,
-                  ArrayRef<NewArchiveMember> NewMembers) {
+                  bool Deterministic, ArrayRef<NewArchiveMember> NewMembers) {
   static char PaddingData[8] = {'\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'};
 
   // This ignores the symbol table, but we only need the value mod 8 and the
   // symbol table is aligned to be a multiple of 8 bytes
   uint64_t Pos = 0;
 
   std::vector<MemberData> Ret;
   bool HasObject = false;
+
+  // UniqueTimestamps is a special case to improve debugging on Darwin:
+  //
+  // The Darwin linker does not link debug info into the final
+  // binary. Instead, it emits entries of type N_OSO in in the output
+  // binary's symbol table, containing references to the linked-in
+  // object files. Using that reference, the debugger can read the
+  // debug data directly from the object files. Alternatively, an
+  // invocation of 'dsymutil' will link the debug data from the object
+  // files into a dSYM bundle, which can be loaded by the debugger,
+  // instead of the object files.
+  //
+  // For an object file, the N_OSO entries contain the absolute path
+  // path to the file, and the file's timestamp. For an object
+  // included in an archive, the path is formatted like
+  // "/absolute/path/to/archive.a(member.o)", and the timestamp is the
+  // archive member's timestamp, rather than the archive's timestamp.
+  //
+  // However, this doesn't always uniquely identify an object within
+  // an archive -- an archive file can have multiple entries with the
+  // same filename. (This will happen commonly if the original object
+  // files started in different directories.) The only way they get
+  // distinguished, then, is via the timestamp. But this process is
+  // unable to find the correct object file in the archive when there
+  // are two files of the same name and timestamp.
+  //
+  // Additionally, timestamp==0 is treated specially, and causes the
+  // timestamp to be ignored as a match criteria.
+  //
+  // That will "usually" work out okay when creating an archive not in
+  // deterministic timestamp mode, because the objects will probably
+  // have been created at different timestamps.
+  //
+  // To ameliorate this problem, in deterministic archive mode (which
+  // is the default), on Darwin we will emit a unique non-zero
+  // timestamp for each entry with a duplicated name. This is still
+  // deterministic: the only thing affecting that timestamp is the
+  // order of the files in the resultant archive.
+  //
+  // See also the functions that handle the lookup:
+  // in lldb/ ObjectContainerBSDArchive::Archive::FindObject()
+  // in llvm/tools/dsymutil: BinaryHolder::GetArchiveMemberBuffers().
+  bool UniqueTimestamps = Deterministic && (Kind == object::Archive::K_DARWIN ||
+                                            object::Archive::K_DARWIN64);
+  std::map<StringRef, unsigned> FilenameCount;
+  if (UniqueTimestamps) {
+    for (const NewArchiveMember &M : NewMembers) {
+      FilenameCount[M.MemberName]++;
+    }
+    for (auto &Entry : FilenameCount) {
+      Entry.second = Entry.second > 1 ? 1 : 0;
+    }
+  }
+
   for (const NewArchiveMember &M : NewMembers) {
     std::string Header;
     raw_string_ostream Out(Header);
 
     MemoryBufferRef Buf = M.Buf->getMemBufferRef();
     StringRef Data = Thin ? "" : Buf.getBuffer();
 
     // ld64 expects the members to be 8-byte aligned for 64-bit content and at
     // least 4-byte aligned for 32-bit content.  Opt for the larger encoding
     // uniformly.  This matches the behaviour with cctools and ensures that ld64
     // is happy with archives that we generate.
     unsigned MemberPadding = Kind == object::Archive::K_DARWIN
                                  ? OffsetToAlignment(Data.size(), 8)
                                  : 0;
     unsigned TailPadding = OffsetToAlignment(Data.size() + MemberPadding, 2);
     StringRef Padding = StringRef(PaddingData, MemberPadding + TailPadding);
 
-    printMemberHeader(Out, Pos, StringTable, Kind, Thin, ArcName, M,
+    sys::TimePoint<std::chrono::seconds> ModTime;
+    if (UniqueTimestamps)
+      // Increment timestamp for each file of a given name.
+      ModTime = sys::toTimePoint(FilenameCount[M.MemberName]++);
+    else
+      ModTime = M.ModTime;
+    printMemberHeader(Out, Pos, StringTable, Kind, Thin, ArcName, M, ModTime,
                       Buf.getBufferSize() + MemberPadding);
     Out.flush();
 
     Expected<std::vector<unsigned>> Symbols =
         getSymbols(Buf, SymNames, HasObject);
     if (auto E = Symbols.takeError())
       return std::move(E);
 
Error llvm::writeArchive(StringRef ArcName,
                          std::unique_ptr<MemoryBuffer> OldArchiveBuf) {
   assert((!Thin || !isBSDLike(Kind)) && "Only the gnu format has a thin mode");
 
   SmallString<0> SymNamesBuf;
   raw_svector_ostream SymNames(SymNamesBuf);
   SmallString<0> StringTableBuf;
   raw_svector_ostream StringTable(StringTableBuf);
 
-  Expected<std::vector<MemberData>> DataOrErr =
-      computeMemberData(StringTable, SymNames, Kind, Thin, ArcName, NewMembers);
+  Expected<std::vector<MemberData>> DataOrErr = computeMemberData(
+      StringTable, SymNames, Kind, Thin, ArcName, Deterministic, NewMembers);
   if (Error E = DataOrErr.takeError())
     return E;
   std::vector<MemberData> &Data = *DataOrErr;
 
   if (!StringTableBuf.empty())
     Data.insert(Data.begin(), computeStringTable(StringTableBuf));
 
   // We would like to detect if we need to switch to a 64-bit symbol table.

llvm/test/Object/archive-darwin-duplicates.test

+Test that darwin archives are created with distinct (name,timestamp) pairs.
+
+The two members with the duplicated name "test.o" must have unique
+non-zero timestamps, while baz.o, being unique, remains a zero
+timestamp.
+
+RUN: mkdir -p %T/sub1
+RUN: printf test > %T/test.o
+RUN: printf sub1/test > %T/sub1/test.o
+RUN: printf baz > %T/baz.o
+
+RUN: rm -f %t.a
+RUN: llvm-ar -format=darwin rcs %t.a %T/sub1/test.o %T/test.o %T/baz.o
+RUN: FileCheck -strict-whitespace %s < %t.a
+
+CHECK:#1/12           1           0     0     644     28        `
+CHECK-NEXT:test.o
+CHECK:#1/12           2           0     0     644     20        `
+CHECK-NEXT:test.o
+CHECK:#1/12           0           0     0     644     20        `
+CHECK-NEXT:baz.o