Mail Archives: djgpp-workers/2000/07/25/06:28:02
On Tue, 25 Jul 2000, Laurynas Biveinis wrote:
> > Anyway, as I see it, we have one more chance to get this done as we
> > think it should: we could submit patches to the configury to do what
> > we want, i.e. to use system headers at build time and not to install
> > GCC's headers (except stdarg.h and varargs.h) at "make install" time.
>
> I'm afraid this kind of patch semantically is too similar to overriding
> USER_H (set of headers to be installed), to be considered.
I'm not sure. In the couple of Mike Stump's messages that did make
sense, he said that stddef.h and assert.h can go. He even offered help
in doing so. And Zack said that limits.h causes more trouble than it's
worth and ``should die''.
So at least for these headers, some of the maintainers might favor
our view and agree to the patches which remove them (for all systems, not
just for us, so USER_H should not be involved).
If that succeeds, it will leave us with stdbool.h, iso646.h, which are
mostly harmless, and possibly float.h/math.h (I couldn't figure out if
these are or aren't installed, different people said different things).
We could deal with those when we get there.
> IMHO the most reasonable (I don't say 'the best') solution is to set
> defaults for *_t types in GCC sources, what does Mark's patch, and adopt
> header sentinels.
I'd want to see each of the GCC's headers that will get such treatment,
before I could form an opinion. Up to now, we only discussed stddef.h in
detail.
But please don't feel obliged to do anything of what I say, if I'm the
only one who says that. Right now, what I'd like most is to forget
about this episode as quickly as I can...
- Raw text -