Re: [RFA/RFC] fix problems with unwinder on mips-irix

[Andrew Cagney <cagney@gnu.org>]

> Andrew,
> 
> 
>>>> >2004-07-22  Joel Brobecker  <brobecker@gnat.com>
>>>> >
>>>> >        * mips-tdep.c (mips_mdebug_frame_cache): Minor reformatting.
>>>> >        (set_saved_reg_info): New function.
>>>> >        (mips32_heuristic_proc_desc): Compute the procedure descriptor
>>>> >        PROC_REG_OFFSET.
>>>> >
>>
>>> 
>>> It's time to cut our losses and split the mdebug and heuristic cases. 
>>> Lets work through this change after I've done that.
> 
> 
> I am sorry. I didn't have time to follow very well your recent changes
> to the mips unwinders. Is it time for me to come back to this issue,
> or do you think I should wait some more?

If you've the time, please do.

Two key things to know:

- with three unwinders handling three different cases previously handled 
by one, there's a lot of dead code paths.  For instance, 
mips32_heuristic_proc_desc is now only called by mips_insn32_frame_cache 
and hence can be inlined there, making it possible for your problem to 
be solved more locally.

- I'm interested in a brutal overhaul of an unwinder, not a small tweak :-)

>>> In the mean time, can I encourage you to look at the s390 and it's 
>>> stub-unwinder - the MIPS desperatly needs one of these (and it is 
>>> something that can be wipped up in isolation).  To see where its needed 
>>> do a STEPI into shlib trampoline code.
> 
> 
> I would love to, IRIX looks like a fun platform to play with, but I am
> a bit time-starved right now. Hopefully in september.

I've added a stub unwinder.

Andrew