Re: [tree-ssa] Don't return but abort() in tree-ssa-ccp on non-GIMPLEwith checking enabled

[Steven Bosscher <s dot bosscher at student dot tudelft dot nl>]

law@redhat.com wrote:

> In message <3EECF3A6.4060406@student.tudelft.nl>, Steven Bosscher writes:
> >This is a multi-part message in MIME format.
> >--------------010002090904000804070908
> >Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> >Content-Transfer-Encoding: 7bit
> >
> >We should never ever see non-GIMPLE at this point, so the return truely 
> >is lame.  But what's worse
> >is that this is the kind of thing that causes bugs that disappear with 
> >checking enabled but cause ICEs
> >with checking disabled.  Shouldn't _all_ checks in #if ENABLE_CHECKING 
> >abort if they fail?
> >
> >Bootstrapped C/C++ on i686-pc-cygwin.  Not really obvious because I 
> >don't know the history of
> >this check, so... OK?
> Err, what about an ASM operand or a backend builtin?  I believe we have
> those marked as non-gimple right now.  Have you checked those explicitly?
>
> jeff
>
>  
Huh?  No but I thought that everything in the tree optimizers is GIMPLE.

But even then, don't you think it is strange that we would return for 
non-GIMPLE with checking enabled, but that with checking disabled we 
would just ignore the "non-GIMPLE-ness" of the statement and just go on?

Anyway, Diego said he will kill the check in an upcoming patch.

Gr.
Steven