[Bug target/13722] [3.4/3.5 regression] [ia64] ICE in push_secondary_reload

["zack at codesourcery dot com" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From zack at codesourcery dot com  2004-01-22 08:22 -------
Subject: Re:  [3.4/3.5 regression] [ia64] ICE in
 push_secondary_reload

"wilson at specifixinc dot com" <gcc-bugzilla@gcc.gnu.org> writes:

> The simple fact is that you have broken the IA-64 compiler a number of 
> times, and you have not always been responsive to fixing the problems 
> you have caused.  This is very annoying.  There are many people using 
> the IA-64 port, and you inconvience them everytime you break it.
>
> I have important work to do also, but instead I am forced to do your 
> work for you, and I am not happy about this.
>
> If you aren't going to try to do IA-64 work properly, then you could at 
> least post patches for review instead of just blindly checking them in 
> and waiting to see what breaks.  Or you could voluntarily revert patches 
> when it is proven that they don't work.  This is much more friendly to 
> the rest of us, particularly those of us using the IA-64 gcc port for 
> real work.

The fact of the matter is that I *have* always tested these changes on
an ia64 machine.  It runs HP/UX, which (a) is big-endian, and (b) does
not (to my knowledge) have an Ada port yet.  Therefore testing on
ia64-linux is, unfortunately, going to show problems that my testing
cannot.  It is not feasible for me to change the operating system on
the machine.

The implication of what you are saying above is that you don't think
anyone without access to an ia64-linux box should be allowed to touch
the ia64 back end, which I think is bad policy.

It is certainly the case that I don't know as much about coding
machine descriptions as some people do, and if you would like, I will
stop checking backend patches in immediately, to give others a chance
to find problems - but my experience is that everyone ignores me when
I ask for a second opinion on my patches, so I don't expect this to
help much.

I *have* managed to reproduce the original bootstrap failure with an
i686-linux->ia64-hpux Ada cross compiler, for the record.

> It has always worked this way.  I know this is lame, but one should 
> never underestimate the difficulty of trying to change how reload works. 
>   It is much easier to change all md file than to change reload.  I 
> think there is an important reason why it works this way, but it 
> probably isn't possible to figure it out without spending an 
> unreasonable amount of time messing with the code.

RTH and I talked this over on IRC and came up with a clever plan,
which I intend to implement tonight and post for review, assuming it
works, tomorrow morning.

zw


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13722