RFC: weakref GCC attribute and .weakref assembly directive

[Nick Clifton]

Hi Alex,

> Here's the revised patch.  Tested on amd64-linux-gnu and all of the
> above.  Ok to install?

Almost. :-)  I reran the GAS testsuite with your patch applied, but a 
few targets still have new regressions because of the new tests:

   alpha-osf4.0:
   mips-ecoff:
     Fail because ".weak" is an unrecognised pseudo op
       for the ECOFF file format.
     Recommendation: Either import obj_coff_weak() from
       tc-coff.c to tc-ecoff.c, or else mark the tests as
       XFAIL because the ECOFF format does not (yet)
       support weak symbols.

   arm-aout:
   ns32k-netbsd:
   tic30-aout:
     Fail because the "strong undefined syms" test does not
       detect any undefined symbols.  This is because such
       symbols are not supported by the aout file format -
       they are turned into absolute symbols instead.
     Recommendation:  Mark this test as xfail for any aout
       format target.

   mmix-mmixware:
     Fails because its target specific code cannot handle the
       "<symbol> = <expression>" syntax.
     Recommendation: Mark the test as xfail and prompt
       Hans-Peter to fix the mmix backend.

   pdp11-aout:
     Fails because it does not support the .long directive.
     Recommendation:  Mark the new tests as XFAIL for this
       target.

   sh64-linux-gnu:
     Fails with an internal error, cause unknown.
     Recommendation: Investigate cause of internal error.  It
      looks like it will be another "target does not support
      pseudo op foo" problem, but this should be confirmed
      before marking the tests as XFAILs.

   sh-nto:
     Fails with a seg-fault.
     Recommendation:  Investigate cause of seg-fault.  It will
      almost certainly be SH-specific, but it would be good to
      prove this and even better to fix it.

   tic4x-coff:
   tic54x-coff:
     Fail because they require all pseudo ops to be prefixed
      by whitespace.  (ie they cannot start in column 1).
     Recommendation: Fix the tests by adding a tab character
      before the start of all the pseudo ops in the new
      assembler source files.

The problems with the SH toolchains are the most serious, but it would 
great if you could fix all of the above.

Cheers
   Nick