[RFA/commit] include alloca.h if available.

[Pedro Alves]

On Wednesday 01 September 2010 05:38:10, Joel Brobecker wrote:
> > If it is just for malloc, it sounds like the include is no longer
> > necessary? Unless of course malloc.h is the only location where
> > alloca is defined on mingw32.
> I just checked, and from what I can tell, no alloca.h on MinGW.

Right.  malloc.h is there for alloca, and there is no alloca.h, AFAIK.

> So the question is: Do we want to group all includes of malloc.h
> in server.h? I don't see any strong reason to believe that one
> way or the other is better, so I can live with the status quo.
> But I'm happy to group them all, or even move the include of
> alloca.h inside the various .c file as needed (it seems odd, now
> that I think of it, to include alloca.h in server.h, and yet have
> the includes of malloc.h spread out in the .c files).

This is exactly what I was pointing at.  It'd be tidier to have
alloca.h and malloc.h together, either in server.h or spread out
in the .c files, but together.  Just for the principle that
headers should only contain includes of what is scrictly necessary
for the interface they describe, I'd prefer the .c files.  But
then again, server.h is a kitchen sink, and, already includes
other related system includes like stdlib.h and string.h, that
putting it there is certainly fine, and probably avoids more work
in the future (note how I needed to add malloc.h to mem-break.c
recently for mingw32).  Thus, I mildly prefer server.h as you
have done.  But this was just a note, not a request!

-- 
Pedro Alves