Re: obstack fails to compile on OS X 10.7

[David Aguilar <davvid@gmail.com>]

On Sun, Aug 28, 2011 at 10:08:46PM +0200, Fredrik Kuivinen wrote:
> On Sat, Aug 27, 2011 at 03:14:43AM -0700, David Aguilar wrote:
> > On Sat, Aug 27, 2011 at 02:21:40AM -0400, Brian Gernhardt wrote:
> > > Some of the errors look like things I could track down, but some just confuse me.  If anyone else could take a look into this, it would be much appreciated.
> > > 
> > > ~~ Brian G.
> > > 
> > > gcc -o compat/obstack.o -c -MF compat/.depend/obstack.o.d -MMD -MP  -Wall -Wdeclaration-after-statement -Werror -Wno-deprecated-declarations -I. -DUSE_ST_TIMESPEC  -DSHA1_HEADER='"block-sha1/sha1.h"'  -DNO_MEMMEM  compat/obstack.c
> > > In file included from compat/obstack.c:30:
> > > compat/obstack.h:190: error: __block attribute can be specified on variables only
> > > compat/obstack.c:70: error: expected specifier-qualifier-list before ‘uintmax_t’
> > > compat/obstack.c:111:24: error: exitfail.h: No such file or directory
> > > cc1: warnings being treated as errors
> > > compat/obstack.c: In function ‘print_and_abort’:
> > > compat/obstack.c:436: warning: implicit declaration of function ‘gettext’
> > > compat/obstack.c:436: warning: incompatible implicit declaration of built-in function ‘gettext’
> > > compat/obstack.c:438: error: ‘exit_failure’ undeclared (first use in this function)
> > > compat/obstack.c:438: error: (Each undeclared identifier is reported only once
> > > compat/obstack.c:438: error: for each function it appears in.)
> > > compat/obstack.c:439: warning: ‘noreturn’ function does return
> > > make: *** [compat/obstack.o] Error 1
> > > 
> > > $ gcc --version
> > > i686-apple-darwin11-llvm-gcc-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)
> > 
> > I ran into the same thing.
> > 
> > This fixes it for me, but we might want to rearrange the
> > #includes a bit.  I think this needs more work.. including
> > compat/obstack.h from kwset.c seems wrong.
> > Should we just include obstack.h in git-compat-util instead?
> > 
> > I suspect that more exotic platforms may have problems
> > with obstack.h as well.  This probably needs some testing
> > on SunOS, AIX, IRIX, etc.
> 
> 
> How about doing something a bit simpler instead and changing obstack.c
> to not make use of exit.h and exitfail.h? Then we don't have to update
> Makefile for all platforms needing NEEDS_OBSTACK and NEEDS_EXITFAIL.

I like this much better.  Less code is better code.

> I don't understand why the ELIDE_CODE check is not sufficient. Care to
> explain?

I can't say I know.  The intention of the check is to avoid
pulling in that section of code when it is already built-in to
the C library.  Maybe that check doesn't quite mean the same
thing when the file is used alone outside of its original
context?

> -void obstack_free (struct obstack *__obstack, void *__block);
> +void obstack_free (struct obstack *obstack, void *block);

Tiny nit: I know it's just a declaration but would it be
advisable to drop the variable names altogether here?
Having a pointer and a structure with the same "obstack"
name could be confusing.

This looks good otherwise.  I was a bit iffy about my patch
when I had to bring in the extra headers.
Doing without them is much better.

I tried your patch on top of my recent
"Improve compiler header dependency check" and it worked fine.

So...
Tested-by: David Aguilar <davvid@gmail.com>
-- 
					David