[PATCH] D40549: [ELF] - Add support for --just-symbols flag.

[Rafael Avila de Espindola via llvm-commits]

George Rimar <grimar at accesssoftek.com> writes:

>>> +    StringRef Name = check(Sym.getName(this->StringTable), toString(this));
>>> +    // We do not want to automatically resolve undefined symbols here, so
>>> +    // leaving them as is, assuming they must be defined somewhere else.
>>> +    if (Sym.st_shndx == SHN_UNDEF) {
>>> +      this->Symbols.push_back(Symtab->addUndefined<ELFT>(
>>> +          Name, Sym.getBinding(), Sym.st_other, Sym.getType(),
>>> +          /*CanOmitFromDynSym=*/false, this));
>>> +      continue;
>>> +    }
>>
>>bfd seems to ignore undefined symbols, no?
>
> No, if you have symbols.o with undefined symbol:
> .globl test
> test:
>   callq undef at PLT
>
> and link it using --just-symbols=symbols.o with stub main.o:
> _start:
>  nop
>
> Symbol will remain undefined:
> ld.bfd main.o --just-symbols=symbols.o -o out
> readelf -a out
>
> Symbol table '.symtab' contains 10 entries:
>    Num:    Value          Size Type    Bind   Vis      Ndx Name
> ....
>      6: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT  UND undef

I can't reproduce this with 2.27-28.fc26. In any case, it is probably
better to start without .o files support. With that we don't have to
worry about undefined symbols.

Cheers,
Rafael