[PATCH] Fix newlib-vs-tic80 clash that breaks enable-targets=all on newlib hosts

["Dave Korn" <dave dot korn at artimi dot com>]

    Hi everyone,

  I couldn't build all targets on cygwin the other day:

http://sourceware.org/ml/binutils/2005-04/msg00772.html

  Well, include/coff/tic80.h has this chunk...

/* Names of "special" sections.  */
#define _TEXT	".text"
#define _DATA	".data"
#define _BSS	".bss"
#define _CINIT	".cinit"
#define _CONST	".const"
#define _SWITCH	".switch"
#define _STACK	".stack"
#define _SYSMEM	".sysmem"

... that defines some system-specific names for linker sections.  Now I'm
building on a newlib-based host (cygwin), and newlib has a host header
called _ansi.h (which is included from stdio.h, which is pulled in by
sysdep.h), which includes these definitions:

#ifdef _HAVE_STDC
#define	_PTR		void *
#define	_AND		,
#define	_NOARGS		void
#define	_CONST		const
#define	_VOLATILE	volatile
[etc etc.]

  and so the build fails in bfd/coff-tic80.c, with a macro redefinition
error:

In file included from /usr/build/binutils-src/src/bfd/coff-tic80.c:31:
/usr/build/binutils-src/src/include/coff/tic80.h:71:1: "_CONST" redefined
In file included from /usr/include/stdio.h:29,
                 from /usr/build/binutils-src/src/bfd/sysdep.h:33,
                 from /usr/build/binutils-src/src/bfd/coff-tic80.c:29:
/usr/include/_ansi.h:30:1: this is the location of the previous definition


  Nick suggested killing the definition in tic80.h, but I'm not sure,
because it's a user-visible include file, and it might impact people's code.
_CONST belongs to the system namespace.  tic80.h is effectively a tic system
header; it's within its rights to define _CONST if that's how it's defined
in the native system library.  And newlib's _ansi.h is also a system header;
it too is within its rights to define _CONST.

  I imagine this hasn't come up before because nobody has been building
tic-targeted binutils on newlib-based hosts.  Being a small embedded target,
tic itself is probably a newlib target (I haven't checked), and I'm also
guessing that it's not self-hosting so nobody's been building bfd on it or
they would have come across this problem.  So it may well be that binutils
itself is in fact the only application code anywhere that references this
header.  But OTOH, if there is anyone out there using it, they will have
already come across this problem and arranged their own local solution for
which symbol they wanted defined.

  So I figured it was most minimally-invasive to take care of the clash in
the one and only source file that actually references both these headers,
coff-tic80.c.  Here's the patch I used to do so.  It's not tested yet
because there's another clash later on in the build, but it makes the build
succeed and it's pretty trivial; I'd call it 'obvious'.  I'll take a look at
the TARGET_FORMAT trouble next.

2005-05-07  Dave Korn <dave.korn@artimi.com>

bfd/
	* coff-tic80.c:  Undefine _CONST after system headers to prevent
	clash with tic80-specific definition in include/coff/tic80.h.


    cheers, 
      DaveK
-- 
Can't think of a witty .sigline today....

Attachment: newlib-vs-tic80-patch.diff
Description: Binary data