[lld][WebAssembly] Match the ELF linker in transitioning away from archive indexes.

The ELF linker transitioned away from archive indexes in
https://reviews.llvm.org/D117284.

This paves the way for supporting `--start-lib`/`--end-lib` (See #77960)

The ELF linker unified library handling with `--start-lib`/`--end-lib`
handling in https://reviews.llvm.org/D119074.

Author: sbc100

lld/docs/ReleaseNotes.rst

@@ -50,5 +50,9 @@ MachO Improvements
 WebAssembly Improvements
 ------------------------
 
+* Indexes are no longer required on archive files.  Instead symbol information
+  is read from object files within the archive.  This matches the behaviour of
+  the ELF linker.
+
 Fixes
 #####

lld/test/wasm/archive-no-index.s

@@ -5,7 +5,7 @@
 # RUN: llvm-mc -filetype=obj -triple=x86_64-pc-linux -o %t.dir/elf.o %s
 # RUN: llvm-ar rcs %t.dir/libfoo.a %t.dir/elf.o
 # RUN: not wasm-ld %t.dir/libfoo.a -o /dev/null 2>&1 | FileCheck %s
-# CHECK: error: unknown file type: {{.*}}libfoo.a(elf.o)
+# CHECK: warning: {{.*}}libfoo.a: archive member 'elf.o' is neither Wasm object file nor LLVM bitcode
 
 .globl _start
 _start:

lld/test/wasm/bad-archive-member.s

@@ -272,9 +272,11 @@ void LinkerDriver::addFile(StringRef path) {
     if (fs::exists(importFile))
       readImportFile(importFile.str());
 
+    auto members = getArchiveMembers(mbref);
+
     // Handle -whole-archive.
     if (inWholeArchive) {
-      for (const auto &[m, offset] : getArchiveMembers(mbref)) {
+      for (const auto &[m, offset] : members) {
         auto *object = createObjectFile(m, path, offset);
         // Mark object as live; object members are normally not
         // live by default but -whole-archive is designed to treat
@@ -289,12 +291,15 @@ void LinkerDriver::addFile(StringRef path) {
     std::unique_ptr<Archive> file =
         CHECK(Archive::create(mbref), path + ": failed to parse archive");
 
-    if (!file->isEmpty() && !file->hasSymbolTable()) {
-      error(mbref.getBufferIdentifier() +
-            ": archive has no index; run ranlib to add one");
+    for (const auto &[m, offset] : members) {
+      auto magic = identify_magic(m.getBuffer());
+      if (magic == file_magic::wasm_object || magic == file_magic::bitcode)
+        files.push_back(createObjectFile(m, path, offset, true));
+      else
+        warn(path + ": archive member '" + m.getBufferIdentifier() +
+             "' is neither Wasm object file nor LLVM bitcode");
     }
 
-    files.push_back(make<ArchiveFile>(mbref));
     return;
   }
   case file_magic::bitcode:
@@ -732,16 +737,10 @@ static Symbol *handleUndefined(StringRef name, const char *option) {
 
 static void handleLibcall(StringRef name) {
   Symbol *sym = symtab->find(name);
-  if (!sym)
-    return;
-
-  if (auto *lazySym = dyn_cast<LazySymbol>(sym)) {
-    MemoryBufferRef mb = lazySym->getMemberBuffer();
-    if (isBitcode(mb)) {
-      if (!config->whyExtract.empty())
-        ctx.whyExtractRecords.emplace_back("<libcall>", sym->getFile(), *sym);
-      lazySym->extract();
-    }
+  if (sym && sym->isLazy() && isa<BitcodeFile>(sym->getFile())) {
+    if (!config->whyExtract.empty())
+      ctx.whyExtractRecords.emplace_back("<libcall>", sym->getFile(), *sym);
+    cast<LazySymbol>(sym)->extract();
   }
 }
 

lld/wasm/Driver.cpp

@@ -75,26 +75,19 @@ std::optional<MemoryBufferRef> readFile(StringRef path) {
 }
 
 InputFile *createObjectFile(MemoryBufferRef mb, StringRef archiveName,
-                            uint64_t offsetInArchive) {
+                            uint64_t offsetInArchive, bool lazy) {
   file_magic magic = identify_magic(mb.getBuffer());
   if (magic == file_magic::wasm_object) {
     std::unique_ptr<Binary> bin =
         CHECK(createBinary(mb), mb.getBufferIdentifier());
     auto *obj = cast<WasmObjectFile>(bin.get());
     if (obj->isSharedObject())
       return make<SharedFile>(mb);
-    return make<ObjFile>(mb, archiveName);
+    return make<ObjFile>(mb, archiveName, lazy);
   }
 
-  if (magic == file_magic::bitcode)
-    return make<BitcodeFile>(mb, archiveName, offsetInArchive);
-
-  std::string name = mb.getBufferIdentifier().str();
-  if (!archiveName.empty()) {
-    name = archiveName.str() + "(" + name + ")";
-  }
-
-  fatal("unknown file type: " + name);
+  assert(magic == file_magic::bitcode);
+  return make<BitcodeFile>(mb, archiveName, offsetInArchive, lazy);
 }
 
 // Relocations contain either symbol or type indices.  This function takes a
@@ -391,9 +384,30 @@ static bool shouldMerge(const WasmSegment &seg) {
   return true;
 }
 
-void ObjFile::parse(bool ignoreComdats) {
-  // Parse a memory buffer as a wasm file.
-  LLVM_DEBUG(dbgs() << "Parsing object: " << toString(this) << "\n");
+void ObjFile::parseLazy() {
+  LLVM_DEBUG(dbgs() << "ObjFile::parseLazy: " << toString(this) << "\n");
+  for (const SymbolRef &sym : wasmObj->symbols()) {
+    const WasmSymbol &wasmSym = wasmObj->getWasmSymbol(sym.getRawDataRefImpl());
+    if (!wasmSym.isDefined())
+      continue;
+    symtab->addLazy(wasmSym.Info.Name, this);
+    // addLazy() may trigger this->extract() if an existing symbol is an
+    // undefined symbol. If that happens, this function has served its purpose,
+    // and we can exit from the loop early.
+    if (!lazy)
+      break;
+  }
+}
+
+ObjFile::ObjFile(MemoryBufferRef m, StringRef archiveName, bool lazy)
+    : InputFile(ObjectKind, m) {
+  this->lazy = lazy;
+  this->archiveName = std::string(archiveName);
+
+  // If this isn't part of an archive, it's eagerly linked, so mark it live.
+  if (archiveName.empty())
+    markLive();
+
   std::unique_ptr<Binary> bin = CHECK(createBinary(mb), toString(this));
 
   auto *obj = dyn_cast<WasmObjectFile>(bin.get());
@@ -406,6 +420,11 @@ void ObjFile::parse(bool ignoreComdats) {
   wasmObj.reset(obj);
 
   checkArch(obj->getArch());
+}
+
+void ObjFile::parse(bool ignoreComdats) {
+  // Parse a memory buffer as a wasm file.
+  LLVM_DEBUG(dbgs() << "ObjFile::parse: " << toString(this) << "\n");
 
   // Build up a map of function indices to table indices for use when
   // verifying the existing table index relocations
@@ -717,43 +736,6 @@ void StubFile::parse() {
   }
 }
 
-void ArchiveFile::parse() {
-  // Parse a MemoryBufferRef as an archive file.
-  LLVM_DEBUG(dbgs() << "Parsing library: " << toString(this) << "\n");
-  file = CHECK(Archive::create(mb), toString(this));
-
-  // Read the symbol table to construct Lazy symbols.
-  int count = 0;
-  for (const Archive::Symbol &sym : file->symbols()) {
-    symtab->addLazy(this, &sym);
-    ++count;
-  }
-  LLVM_DEBUG(dbgs() << "Read " << count << " symbols\n");
-  (void) count;
-}
-
-void ArchiveFile::addMember(const Archive::Symbol *sym) {
-  const Archive::Child &c =
-      CHECK(sym->getMember(),
-            "could not get the member for symbol " + sym->getName());
-
-  // Don't try to load the same member twice (this can happen when members
-  // mutually reference each other).
-  if (!seen.insert(c.getChildOffset()).second)
-    return;
-
-  LLVM_DEBUG(dbgs() << "loading lazy: " << sym->getName() << "\n");
-  LLVM_DEBUG(dbgs() << "from archive: " << toString(this) << "\n");
-
-  MemoryBufferRef mb =
-      CHECK(c.getMemoryBufferRef(),
-            "could not get the buffer for the member defining symbol " +
-                sym->getName());
-
-  InputFile *obj = createObjectFile(mb, getName(), c.getChildOffset());
-  symtab->addFile(obj, sym->getName());
-}
-
 static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) {
   switch (gvVisibility) {
   case GlobalValue::DefaultVisibility:
@@ -790,8 +772,9 @@ static Symbol *createBitcodeSymbol(const std::vector<bool> &keptComdats,
 }
 
 BitcodeFile::BitcodeFile(MemoryBufferRef m, StringRef archiveName,
-                         uint64_t offsetInArchive)
+                         uint64_t offsetInArchive, bool lazy)
     : InputFile(BitcodeKind, m) {
+  this->lazy = lazy;
   this->archiveName = std::string(archiveName);
 
   std::string path = mb.getBufferIdentifier().str();
@@ -817,6 +800,20 @@ BitcodeFile::BitcodeFile(MemoryBufferRef m, StringRef archiveName,
 
 bool BitcodeFile::doneLTO = false;
 
+void BitcodeFile::parseLazy() {
+  for (auto [i, irSym] : llvm::enumerate(obj->symbols())) {
+    if (irSym.isUndefined())
+      continue;
+    StringRef name = saver().save(irSym.getName());
+    symtab->addLazy(name, this);
+    // addLazy() may trigger this->extract() if an existing symbol is an
+    // undefined symbol. If that happens, this function has served its purpose,
+    // and we can exit from the loop early.
+    if (!lazy)
+      break;
+  }
+}
+
 void BitcodeFile::parse(StringRef symName) {
   if (doneLTO) {
     error(toString(this) + ": attempt to add bitcode file after LTO (" + symName + ")");

lld/wasm/InputFiles.cpp

@@ -14,7 +14,6 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/LTO/LTO.h"
-#include "llvm/Object/Archive.h"
 #include "llvm/Object/Wasm.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/TargetParser/Triple.h"
@@ -45,7 +44,6 @@ class InputFile {
   enum Kind {
     ObjectKind,
     SharedKind,
-    ArchiveKind,
     BitcodeKind,
     StubKind,
   };
@@ -69,6 +67,11 @@ class InputFile {
   void markLive() { live = true; }
   bool isLive() const { return live; }
 
+  // True if this file is exists as in an archive file and has not yet been
+  // extracted.
+  // TODO(sbc): Use this to implement --start-lib/--end-lib.
+  bool lazy = false;
+
 protected:
   InputFile(Kind k, MemoryBufferRef m)
       : mb(m), fileKind(k), live(!config->gcSections) {}
@@ -85,35 +88,14 @@ class InputFile {
   bool live;
 };
 
-// .a file (ar archive)
-class ArchiveFile : public InputFile {
-public:
-  explicit ArchiveFile(MemoryBufferRef m) : InputFile(ArchiveKind, m) {}
-  static bool classof(const InputFile *f) { return f->kind() == ArchiveKind; }
-
-  void addMember(const llvm::object::Archive::Symbol *sym);
-
-  void parse();
-
-private:
-  std::unique_ptr<llvm::object::Archive> file;
-  llvm::DenseSet<uint64_t> seen;
-};
-
 // .o file (wasm object file)
 class ObjFile : public InputFile {
 public:
-  explicit ObjFile(MemoryBufferRef m, StringRef archiveName)
-      : InputFile(ObjectKind, m) {
-    this->archiveName = std::string(archiveName);
-
-    // If this isn't part of an archive, it's eagerly linked, so mark it live.
-    if (archiveName.empty())
-      markLive();
-  }
+  ObjFile(MemoryBufferRef m, StringRef archiveName, bool lazy = false);
   static bool classof(const InputFile *f) { return f->kind() == ObjectKind; }
 
   void parse(bool ignoreComdats = false);
+  void parseLazy();
 
   // Returns the underlying wasm file.
   const WasmObjectFile *getWasmObj() const { return wasmObj.get(); }
@@ -173,10 +155,11 @@ class SharedFile : public InputFile {
 class BitcodeFile : public InputFile {
 public:
   BitcodeFile(MemoryBufferRef m, StringRef archiveName,
-              uint64_t offsetInArchive);
+              uint64_t offsetInArchive, bool lazy);
   static bool classof(const InputFile *f) { return f->kind() == BitcodeKind; }
 
   void parse(StringRef symName);
+  void parseLazy();
   std::unique_ptr<llvm::lto::InputFile> obj;
 
   // Set to true once LTO is complete in order prevent further bitcode objects
@@ -196,14 +179,10 @@ class StubFile : public InputFile {
   llvm::DenseMap<StringRef, std::vector<StringRef>> symbolDependencies;
 };
 
-inline bool isBitcode(MemoryBufferRef mb) {
-  return identify_magic(mb.getBuffer()) == llvm::file_magic::bitcode;
-}
-
 // Will report a fatal() error if the input buffer is not a valid bitcode
 // or wasm object file.
 InputFile *createObjectFile(MemoryBufferRef mb, StringRef archiveName = "",
-                            uint64_t offsetInArchive = 0);
+                            uint64_t offsetInArchive = 0, bool lazy = false);
 
 // Opens a given file.
 std::optional<MemoryBufferRef> readFile(StringRef path);

lld/wasm/InputFiles.h

@@ -26,9 +26,13 @@ SymbolTable *symtab;
 void SymbolTable::addFile(InputFile *file, StringRef symName) {
   log("Processing: " + toString(file));
 
-  // .a file
-  if (auto *f = dyn_cast<ArchiveFile>(file)) {
-    f->parse();
+  // Lazy object file
+  if (file->lazy) {
+    if (auto *f = dyn_cast<BitcodeFile>(file)) {
+      f->parseLazy();
+    } else {
+      cast<ObjFile>(file)->parseLazy();
+    }
     return;
   }
 
@@ -737,16 +741,15 @@ TableSymbol *SymbolTable::resolveIndirectFunctionTable(bool required) {
   return nullptr;
 }
 
-void SymbolTable::addLazy(ArchiveFile *file, const Archive::Symbol *sym) {
-  LLVM_DEBUG(dbgs() << "addLazy: " << sym->getName() << "\n");
-  StringRef name = sym->getName();
+void SymbolTable::addLazy(StringRef name, InputFile *file) {
+  LLVM_DEBUG(dbgs() << "addLazy: " << name << "\n");
 
   Symbol *s;
   bool wasInserted;
   std::tie(s, wasInserted) = insertName(name);
 
   if (wasInserted) {
-    replaceSymbol<LazySymbol>(s, name, 0, file, *sym);
+    replaceSymbol<LazySymbol>(s, name, 0, file);
     return;
   }
 
@@ -763,15 +766,15 @@ void SymbolTable::addLazy(ArchiveFile *file, const Archive::Symbol *sym) {
     if (auto *f = dyn_cast<UndefinedFunction>(s))
       oldSig = f->signature;
     LLVM_DEBUG(dbgs() << "replacing existing weak undefined symbol\n");
-    auto newSym = replaceSymbol<LazySymbol>(s, name, WASM_SYMBOL_BINDING_WEAK,
-                                            file, *sym);
+    auto newSym =
+        replaceSymbol<LazySymbol>(s, name, WASM_SYMBOL_BINDING_WEAK, file);
     newSym->signature = oldSig;
     return;
   }
 
   LLVM_DEBUG(dbgs() << "replacing existing undefined\n");
   const InputFile *oldFile = s->getFile();
-  file->addMember(sym);
+  replaceSymbol<LazySymbol>(s, name, 0, file)->extract();
   if (!config->whyExtract.empty())
     ctx.whyExtractRecords.emplace_back(toString(oldFile), s->getFile(), *s);
 }