Re: Fix assertion failure on aliases of dynamic weak symbols

["H.J. Lu" <hjl at lucon dot org>]

On Mon, Jul 16, 2007 at 09:32:54PM -0700, Ian Lance Taylor wrote:
> "H.J. Lu" <hjl@lucon.org> writes:
> 
> > That piece of code was added by Richard on May 3, 1999 with
> > comments:
> 
> Richard added the entire binutils world on May 3, 1999.  I actually
> wrote that code originally, in the Cygnus repository.
> 
> >   /* Now set the weakdefs field correctly for all the weak defined
> >      symbols we found.  The only way to do this is to search all the
> >      symbols.  Since we only need the information for non functions in
> >      dynamic objects, that's the only time we actually put anything on
> >      the list WEAKS.  We need this information so that if a regular
> >      object refers to a symbol defined weakly in a dynamic object, the
> >      real symbol in the dynamic object is also put in the dynamic
> >      symbols; we also must arrange for both symbols to point to the
> >      same memory location.  We could handle the general case of symbol
> >      aliasing, but a general symbol alias can only be generated in
> >      assembler code, handling it correctly would be very time
> >      consuming, and other ELF linkers don't handle general aliasing
> >      either.  */
> > 
> > It no longer applies today since there is no difference between
> > weak def and strong def in a shared library. Richard, do you
> > remember if there is a testcase for your change? Can we remove
> > the whole u.weakdef thing now? It will make linker simpler and
> > faster.
> 
> The fact that there is no difference between weak defs and strong defs
> is irrelevant here.  What is happening here is that the library
> defines "environ" as a weak symbol.  "environ" is a weak alias for the
> real symbol named "__environ".  The program refers to "environ".
> Nothing outside the shared library refers to "__environ".  The shared
> library refers only to "__environ", never to "environ" (other than
> defining it as a weak alias).  If we only put "environ" in the dynamic
> symbol table, then the dynamic linker will update the GOT entry for
> "environ" so that all references to "environ" refer to the copy in the
> main program.  If we don't put "__environ" in the dynamic symbol
> table, then the GOT entry for "__environ" will continue to refer to
> the copy in the dynamic object's data segment.  The effect is that the
> dynamic object will set "__environ" at run time, and the program will
> happily look at the uninitialized variable "environ".

It sounds familar. I looked at the binutils archive. Unfortunately,
it starts around May 10, 1999 :-(. I will add a testcase for this.

I think there is a small problem. We are making a typed weak def an
alias of a linker-created strong def without type.  I don't think we
want to do it.  Should we require symbols with aliass should be typed?


H.J.