Re: mips and frv testsuite failures after plugin patch

[Dave Korn <dave dot korn dot cygwin at gmail dot com>]

On 25/10/2010 04:34, Alan Modra wrote:
> On Wed, Oct 20, 2010 at 04:08:23PM +0100, Dave Korn wrote:
>> On 18/10/2010 14:31, Tristan Gingold wrote:
>>> Does it also mean that you're OK if I create the release branch this week ?
>>   Yep.  I'm still waiting for someone to explicitly OK the one real bug fix at
> 
> Dave, your patch introduced a bunch of testsuite failures, some shown
> below.
> 
> frv-elf  +FAIL: FRV TLS relocs, static linking

> mips-linux  +FAIL: Dynamic executable with TLS

  Urgh.  Any other targets?  A quick grep suggests bfin and m68k might be
affected too.

> From a quick look at the symptoms and at the changes your patch
> introduced, I'd say it was simply due to opening the plugin bfd in
> plugin_get_ir_dummy_bfd via ldfile_try_open_bfd.  A number of the hash
> functions in elf32-frv.c and elfxx-mips.x use abfd->id, resulting in
> various bits and pieces coming out of a hash traversal in a different
> order after your patch.

  Groan.  Would it be suitable to update the testcases so they match the new
order of outputs?


  ... oooooor, I could make _bfd_delete_bfd decrement the _bfd_id_counter when
it notices we're deleting the last-created BFD?  Hmm, I like that idea.  It
seems a lot simpler.  There shouldn't be any problem with the notion of
speculatively allocating an ID and then recycling it immediately (at any rate,
before any other IDs are allocated).

    cheers,
      DaveK