Re: fixincl now run at every stage?

[Zack Weinberg <zack at wolery dot cumb dot org>]

On Tue, Feb 01, 2000 at 09:19:31PM +0100, Andi Kleen wrote:
> On Tue, Feb 01, 2000 at 07:10:02PM +0100, Zack Weinberg wrote:
> > No - you require basically every binary packager to go through their
> > includes and fix the damn bugs so gcc doesn't have to.
> > 
> > If octave's headers require fixing, that is a bug in octave.
> 
> I was not sure what exactly fixincl was fixing in them (and I have already
> deleted them, and I cannot bootstrap ATM because the fixincl stuff just
> runs amok). I'll try later again when the tree has settled.
> 
> They compile fine with egcs 1.1, because that is the compiler Octave was 
> built with.
> 
> So it would be only useful if fixincl would fix some breakage that
> the old gcc tolerated, but the new one doesn't.

fixinc tends to fix things that are not tolerated by the compiler in
standards conforming mode, but accepted otherwise.  Like #ifdef i386
vs. #ifdef __i386__.  In fact, in my tree, those are the only changes
it makes.

> In my experience (modulo harmless warnings like the X11 return types)

?

...
> Things like changing // comments to /* */ for C are IMHO just a nuisance and
> there should be some way to disable that.

Already scrapped, in my tree.

> Also fixing the original packages is the only way to handle packaged
> gcc versions. I'm afraid but fixincl is fundamentally incompatible
> to rpm, dpkg and friends. 

All you have to do is package only the files in
.../lib/gcc-lib/T/V/include that belong to gcc itself, and file bugs
against the packages whose headers got fixed.

zw