On Mon, 20 Feb 2012 14:17:03 -0500
Rich Felker <dalias@aerifal.cx> wrote:
 
> Doesn't 3.81 already have some important bugs you need to patch
> anyway? If so you could add ar.h at the same time. This won't be
> needed for musl once I get ar.h added, but it would make the resulting
> build more portable anyway.

It's running OK without patches, but I haven't pushed too hard.  As mentioned, I just copied ar.h to /usr/include.
 
> > By the way, I've put together a sys/cdefs.h header that gets a lot
> > of stuff to compile; ~95% of it is backwards-compatability macros.
> > If you want, I can submit it; I'd add
> I wouldn't mind seeing it, and as long as it's clean I'll probably add
> it...
> 
> > #include'ing <sys/cdefs.h> is semi-portable, though (NetBSD, GNU
> > libc)--and the only option if you need to support some old systems
> > as well as c99.
> 
> I'm confused how it would be necessary for this. The __STDC__ and
> __STDC_VERSION__ macros give you all you need to make the appropriate
> definitions yourself if you really need to support pre-standardized C.
> If you just need to support C89 and C99, you'd rarely have to test
> anything anyway; just use the intersection of the 2 languages (which
> is equivalent to C89, except that you can't use C99 keywords like
> inline, restrict, _Bool, etc.).

It looks like it's meant to have one header allow the same source to build 
with anything from K&R to C99 (in theory).

> As such, I still see apps that include sys/cdefs.h as buggy.
> Rich

I view it like features.h: nonsense, but if it's a choice between getting it building in a minute or an hour, I'd rather get it building sooner.  Once I know how broken it is, I can decide whether to clean it up, forget it, or delete it.  I'd rather not waste my time on completely broken code.