This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: MAINTAINERS policy question
- To: aj at suse dot de, mark at codesourcery dot com
- Subject: Re: MAINTAINERS policy question
- From: mike stump <mrs at windriver dot com>
- Date: Mon, 17 Sep 2001 17:38:41 -0700 (PDT)
- Cc: bernds at redhat dot com, gcc at gcc dot gnu dot org
> From: Andreas Jaeger <aj@suse.de>
> Date: Sun, 16 Sep 2001 20:14:57 +0200
> --- cvswrite.html.~1~ Thu Jul 26 16:10:30 2001
> +++ cvswrite.html Sun Sep 16 20:13:24 2001
> @@ -98,7 +98,13 @@
> primary responsibility for ports, front ends, or significant hunks
> of code in the compiler. These folks are allowed to make changes in
> code they maintain without going through any kind of approval
> - process.</dd>
> + process.
> +
> + <p>
> + Maintainers of a port maintain the files in config/<port>/,
> + the configure fragments for the port, documentation for the port and
> + test cases for features or bugs specific to this port. Port
> + maintainers do not have approval writes in other files. </dd>
I mostly agree as well.
The part about testcases and bugs specific to a port is a little to
general. Basically, a port maintainer has privs for all those bytes
that _only_ affect ports under his control, if they are normally done
in such a manner.
This would mean that a bug fix to libgcc2.c in:
#if defined (sony_news) && defined (SYSTYPE_BSD)
#include <stdio.h>
#include <sys/types.h>
#include <sys/param.h>
#include <syscall.h>
#include <machine/sysnews.h>
/* cacheflush function for NEWS-OS 4.2.
This function is called from trampoline-initialize code
defined in config/mips/mips.h. */
void
cacheflush (char *beg, int size, int flag)
{
if (syscall (SYS_sysnews, NEWS_CACHEFLUSH, beg, size, FLUSH_BCACHE))
{
perror ("cache_flush");
fflush (stderr);
abort ();
}
}
#endif /* sony_news */
would be OK by the sony_news/mips maintainer. My wording also makes
it clear that an FreeBSD maintainer can change the FreeBSD make file
fragments, but not the Linux parts of the x86 files. Adding a
#ifdef __mips
[ hack around ]
#endif
into MI code, would not be acceptable, unless signed off by a
maintainer of the MI code.
Testcases are potentially hard to decide. A testcase _should_ be done
in a fairly MI way. A port maintainer can miss ways in which a
testcase is machine dependent. Having someone else review the
testcase is reasonably a good idea. A port maintainer that can
generate a MI testcase I think should be allowed to just check it in
directly.
I am in favor of having things loosely defined, and leaving a little
wiggle room. Those people that are new/inexperienced should caution
to the conservative side, those that are very experienced should be
granted just a little bit more leeway in what parts they change.