[PATCH] COFF: Handle references from LTO object to lazy symbols correctly.

[Rui Ueyama]

LGTM

Thank you for the explanation. My concern was that before this patch, adding a combined compiled bitcode file is really a final operation in the linker, but after this patch, a new call of addMember() can read arbitrary number of new files. That would increase complexity. That being said, this change seems reasonable. It's probably better than maintaining a list of symbols to be linked unconditionally for LTO.


================
Comment at: COFF/SymbolTable.cpp:314
@@ +313,3 @@
+    if (auto *B = dyn_cast<Lazy>(Sym->Body)) {
+      unsigned NumBitcodeFiles = BitcodeFiles.size();
+      auto EC = addMemberFile(B);
----------------
size_t

================
Comment at: COFF/SymbolTable.cpp:317
@@ +316,3 @@
+      if (EC)
+        return EC;
+      if (BitcodeFiles.size() != NumBitcodeFiles) {
----------------
  if (auto EC = ...)
    return EC

================
Comment at: COFF/SymbolTable.cpp:329
@@ +328,3 @@
+  if (reportRemainingUndefines())
+    return make_error_code(LLDError::BrokenFile);
+
----------------
Can we call reportRemainingUndefines after Symtab.addCombinedLTOObject in the Driver to eliminate this call from here?

http://reviews.llvm.org/D10332

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/