Guidelines for building fixincludes for targets?

[Zack Weinberg]

Joe Buck wrote:
> On Fri, Oct 22, 2004 at 05:30:07PM -0400, Andrew Pinski wrote:
> > Almost all targets today need fixincludes, including linux.
> 
> It would seem that any fixincludes support needed on a glibc-based
> platform is a bug in glibc, right?

glibc doesn't provide all of the headers in /usr/include.

On a wholly Free system, yes, theoretically every fixincludes edit is
a bug in some other piece of software that we ought to be able to get
fixed.  I'm actually quite tempted by the notion of disabling
fixincludes altogether for such systems, and seeing how much breaks.

However, there have been real cases in the past where lots of people
had a broken header and it wasn't practical to make them all get a new
libc (libc upgrades are often very painful).  For instance, the bad
__FD_ZERO definitions (AAB_fd_zero_*) and the lvalue cast in obstack.h.

zw