> I Am Dumb.  Check CVS history, but I think I changed that just a
> couple of weeks ago; I audited all the sniffers looking for what ought
> to use the unwound PC and what ought to use the unwound block address.
> Here, I'm pretty sure I made the wrong choice.

Yep, I remembered. However, I also thought that your choice made sense.
I really think it does, but given:

> I would recommend you revert my changes to this function and
> mips_insn32_frame_sniffer instead.

And:

> >      It seems to me that the above check is only an optimization,
> >      and I've spotted at least one instance where I cannot see an
> >      obvious guaranty that the address has not been decremented
> >      by one of the _in_block functions... So the decision I made
> >      was to remove that check.
> 
> No, it's not just an optimization.  Especially with limited debug
> info, it's important.
        ^^^^^^^^^^^^^^
I decided to revert your changes for now.

> >   2. One minor: There was a confusion in the unwinder between
> >      the return address and the address of the instruction calling us.
> >      So I replaced frame_pc_unwind calls by their associated
> >      frame_unwind_address_in_block.
> 
> This half looks right.

Thanks.

So here is what I ended up checkin in:

2007-03-07  Joel Brobecker  <brobecker@adacore.com>

        * mips-tdep.c (mips_insn16_frame_cache, mips_insn32_frame_sniffer):
        Revert the previous change that had some unexpected side-effects
        on mips32.
        (mips_insn16_frame_cache, mips_insn32_frame_cache): Use the proper
        function to get the address of the calling instruction.

Re-tested on mips-irix, just to be sure. Same results as before
(meaning about 500 less FAILs).

I'm also sad to report that I have been told to put off work on
mips-irix for a while. That was a personal initiative on my side,
but I'm lacking time at work, so this was the first task that got cut.
I hope someone else can find the time to bring it back to life...

Thanks again, Daniel.
-- 
Joel