Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

> revno: 109306
> committer: Glenn Morris <rgm@gnu.org>
> branch nick: trunk
> timestamp: Mon 2012-07-30 17:20:43 -0400
> message:
>   Use an include file in configure.ac's AH_BOTTOM

This change brings a little trouble on Windows.

1) The new conf_pos.h unconditionally includes alloca.h, which does
not work on Windows.
2) It unconditionally defines AMPERSAND_FULL_NAME, which again, is
undef on Windows.

To fix 2) it would be enough to #undef it on ms-w32.h, but alas,
post_conf.h is included in config.in / config.nt after
config_opsysfile.

Both problems can be "fixed" by adding suitable #ifdef WINDOWSNT /
#else / #endif to conf_post.h... except that, again, WINDOWSNT is
defined in ms-w32.h, so too late.

I can certainly move the definition of WINDOWSNT to the start of
config.nt, but that adds a difference that I'd like to avoid (and it's
ugly). Would it be possible to move AMPERSAND_FULL_NAME back to
config.in?

As for alloca.h, another possible fix is just to add an empty alloca.h
to nt/sys.

Thoughts?

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 2:10 PM, Juanma Barranquero <lekktu@gmail.com> wrote:

> Both problems can be "fixed" by adding suitable #ifdef WINDOWSNT /
> #else / #endif to conf_post.h... except that, again, WINDOWSNT is
> defined in ms-w32.h, so too late.

Sorry, this is a braino. src/s/ms-w32.h is included before
conf_post.h, so the definition of WINDOWSNT suffices.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Tue, 31 Jul 2012 14:10:05 +0200
> Cc: Emacs developers <emacs-devel@gnu.org>
> 
> > revno: 109306
> > committer: Glenn Morris <rgm@gnu.org>
> > branch nick: trunk
> > timestamp: Mon 2012-07-30 17:20:43 -0400
> > message:
> >   Use an include file in configure.ac's AH_BOTTOM
> 
> This change brings a little trouble on Windows.

What are the manifestations of this trouble?  I just built the current
trunk on Windows, and didn't see any problems.  And I don't think any
related changes were committed since you wrote the above.

What am I missing?

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 7:09 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> What are the manifestations of this trouble?  I just built the current
> trunk on Windows, and didn't see any problems.  And I don't think any
> related changes were committed since you wrote the above.

I didn't make myself clear. The trouble is trying to sync nt/config.nt
with autogen/config.in, i.e, removing code from nt/config.nt that is
already in src/conf_post.h. I have the goal of keeping these two
(config.nt and config.i) as similar as possible, and moving general
stuff to conf_post.h is a good idea, but it causes the conflicts I
explained in the previous message.

I've had no time to delve into it, but it seems like a problem
compiling the lib sources, not src/*.c.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Tue, 31 Jul 2012 19:15:37 +0200
> Cc: rgm@gnu.org, emacs-devel@gnu.org
> 
> On Tue, Jul 31, 2012 at 7:09 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
> > What are the manifestations of this trouble?  I just built the current
> > trunk on Windows, and didn't see any problems.  And I don't think any
> > related changes were committed since you wrote the above.
> 
> I didn't make myself clear. The trouble is trying to sync nt/config.nt
> with autogen/config.in, i.e, removing code from nt/config.nt that is
> already in src/conf_post.h.

So you are saying that replacing the stuff at the end of nt/config.nt
with inclusion of conf_post.h causes compilation failures?  I wonder
why, because conf_post.h looks like an identical copy of that stuff,
and comes at the same place.

The alloca.h stuff is already #ifdef'ed away on config.nt, so it's not
new.  Either ifdef it in conf_post.h, or provide alloca.h in nt/inc.

As for AMPERSAND_FULL_NAME, its unconditional definition is bogus IMO,
and needs to be autoconfiscated, or at the very least ifdef'ed away
for WINDOWSNT and MSDOS.  Undefining it on ms-w32.h is possible, but
not clean, IMO.

> I've had no time to delve into it, but it seems like a problem
> compiling the lib sources, not src/*.c.

If you bump into problems beyond the above, please bring them up.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

Btw, if the idea of conf_post.h is to be the only place of
system-dependent configury that isn't autoconfiscated, and if we want
to remove src/s/ altogether, would it be okay to move stuff from
msdos.h and ms-w32.h into conf_post.h?

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Eli Zaretskii wrote:

> Btw, if the idea of conf_post.h is to be the only place of
> system-dependent configury that isn't autoconfiscated

That wasn't particularly the intention, it was just necessary to work
around an autoconf issue;
http://lists.gnu.org/archive/html/autoconf/2006-10/msg00020.html

> , and if we want to remove src/s/ altogether, would it be okay to move
> stuff from msdos.h and ms-w32.h into conf_post.h?

I'm obviously hoping Someone (TM) will remove s/msdos.h and s/ms-w32.h,
either by moving them to the nt/ or msdos/ directory (or if all else
fails, to src/), or by folding them in to config.nt or some other
place...  No other platform uses config_opsysfile any more, so ideally
that would be removed altogether from conf_post.h.

ms-w32.h is kind of massive, so IMO it would be ugly for that to be in
conf_post.h; but since that platform uses an entirely different
config.nt, I don't see any reason for all that stuff to end up in
conf_post.h. I'd just have an extra section at the end of config.nt that
is not synced from config.in.


AMPERSAND_FULL_NAME and subprocesses obviously belong in configure.ac; I
moved them.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: lekktu@gmail.com,  emacs-devel@gnu.org
> Date: Tue, 31 Jul 2012 14:01:10 -0400
> 
> Eli Zaretskii wrote:
> 
> > Btw, if the idea of conf_post.h is to be the only place of
> > system-dependent configury that isn't autoconfiscated
> 
> That wasn't particularly the intention, it was just necessary to work
> around an autoconf issue;
> http://lists.gnu.org/archive/html/autoconf/2006-10/msg00020.html

OK, but then what's The Plan wrt this stuff?  Are we waiting for
Autoconf to be fixed?  Is this stuff going to be left in a separate
header for the observable future?  Something else?

I mean, it's hard to follow suit on non-Posix platforms, when the
maintainer's intent on Posix was not revealed (apologies if it was and
I missed that).

> > , and if we want to remove src/s/ altogether, would it be okay to move
> > stuff from msdos.h and ms-w32.h into conf_post.h?
> 
> I'm obviously hoping Someone (TM) will remove s/msdos.h and s/ms-w32.h,
> either by moving them to the nt/ or msdos/ directory (or if all else
> fails, to src/), or by folding them in to config.nt or some other
> place...

There's any number of possible ways to do this.  Which one is the
best, I cannot decide without knowing your goal with this stuff, which
is clearly platform-dependent.  The removal of things from src/s/*.h
is going on for some time, but the last thing I expected to see at its
end is a bunch platform-specific macros on a platform-independent
header.  It's confusing.

> AMPERSAND_FULL_NAME and subprocesses obviously belong in configure.ac; I
> moved them.

What about SYSTEM_PURESIZE_EXTRA?  conf_post.h defines that for NS and
Darwin; should I move the msdos.h definitions there as well?  If not,
why not?

Or what about NULL_DEVICE or SEPCHAR? should they be autoconfiscated
(in which case they can be removed from s/msdos.h)?

There are more questions like this, and they get in the way of making
the decision what to do with src/s/ms*.h.

TIA

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Eli Zaretskii wrote:

> Is this stuff going to be left in a separate header for the observable
> future?

I think so.

> What about SYSTEM_PURESIZE_EXTRA?  conf_post.h defines that for NS and
> Darwin; should I move the msdos.h definitions there as well?

I think it's up to you.

> Or what about NULL_DEVICE or SEPCHAR? should they be autoconfiscated

I think so (ditto USER_FULL_NAME, etc), then msdos could sed them.

AFAICS, however, unless you do some more complex sed stuff, you are
always going to need a separate msdos.h, due to 'define DOS_NT' etc. So
how much you want to remove from msdos.h is up to you. So probably the
config_opsysfile piece stays in conf_post.h just for the use of MS; not
a problem IMO.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

I'd be fine with the absolute minimum change:

mv s/msdos.h conf_msdos.h

and change config_opsysfile on msdos from "s/msdos.h" to "conf_msdos.h".

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 7:31 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> So you are saying that replacing the stuff at the end of nt/config.nt
> with inclusion of conf_post.h causes compilation failures?  I wonder
> why, because conf_post.h looks like an identical copy of that stuff,
> and comes at the same place.

Except for alloca.h.

> The alloca.h stuff is already #ifdef'ed away on config.nt, so it's not
> new.  Either ifdef it in conf_post.h, or provide alloca.h in nt/inc.

I'm glad we agree, as that what I proposed. But adding "#ifndef
WINDOWSNT / #endif" around the alloca.h include does not work while
compiling the lib/ sources.

With the attached patch, after "cd nt; configure ...; make" I get

In file included from ../src/config.h:1621:0,
                 from c-ctype.c:18:
../src/conf_post.h:32:20: fatal error: alloca.h: No such file or directory
compilation terminated.
make[1]: *** [oo-spd/i386/c-ctype.o] Error 1
make[1]: Leaving directory `C:/Devel/emacs/repo/trunk/lib'
make: *** [all-other-dirs-gmake] Error 2

    Juanma



=== modified file 'nt/config.nt'
--- nt/config.nt	2012-07-29 16:56:18 +0000
+++ nt/config.nt	2012-07-31 19:32:11 +0000
@@ -1257,6 +1257,9 @@
 /* Make process_send_signal work by "typing" a signal character on the pty. */
 #undef SIGNALS_VIA_CHARACTERS

+/* Define if AH_BOTTOM should include signal.h. */
+#undef SIGNAL_H_AHB
+
 /* Define to l, ll, u, ul, ull, etc., as suitable for constants of type
    'sig_atomic_t'. */
 #undef SIG_ATOMIC_T_SUFFIX
@@ -1306,6 +1309,9 @@
 /* Define to 1 if you can safely include both <sys/time.h> and <time.h>. */
 #define TIME_WITH_SYS_TIME 1

+/* Some platforms redefine this. */
+#undef TIOCSIGSEND
+
 /* Define to 1 if your <sys/time.h> declares `struct tm'. */
 #undef TM_IN_SYS_TIME

@@ -1353,6 +1359,9 @@
 /* Define if the system is compatible with System V. */
 #undef USG5

+/* Define if the system is compatible with System V Release 4. */
+#undef USG5_4
+
 /* Define for USG systems where it works to open a pty's tty in the parent
    process, then close and reopen it in the child. */
 #undef USG_SUBTTY_WORKS
@@ -1383,6 +1392,9 @@
 # endif
 #endif

+/* Some platforms redefine this. */
+#undef WRETCODE
+
 /* Define this to check for malloc buffer overrun. */
 #undef XMALLOC_OVERRUN_CHECK

@@ -1483,6 +1495,24 @@
    declarations. Define as empty for no equivalent. */
 #undef __restrict_arr

+/* Some platforms redefine this. */
+#undef _longjmp
+
+/* Some platforms redefine this. */
+#undef _setjmp
+
+/* Define AMPERSAND_FULL_NAME if you use the convention
+   that & in the full name stands for the login id.  */
+#undef AMPERSAND_FULL_NAME
+
+/* `subprocesses' should be defined if you want to
+   have code for asynchronous subprocesses
+   (as used in M-x compile and M-x shell).
+   Only MSDOS does not support this (it overrides
+   this in its config_opsysfile below).  */
+
+#define subprocesses
+
 /* Define to the used os dependent file. */
 #define config_opsysfile "s/ms-w32.h"

@@ -1585,104 +1615,10 @@
 /* Define as `fork' if `vfork' does not work. */
 #undef vfork

-
-/* Not valid on Windows.  */
-#if 0
-/* On AIX 3 this must be included before any other include file.  */
-#include <alloca.h>
-#if ! HAVE_ALLOCA
-# error "alloca not available on this machine"
-#endif
-#endif
-
-/* This silences a few compilation warnings on FreeBSD.  */
-#ifdef BSD_SYSTEM_AHB
-#undef BSD_SYSTEM_AHB
-#undef BSD_SYSTEM
-#if __FreeBSD__ == 1
-#define BSD_SYSTEM 199103
-#elif __FreeBSD__ == 2
-#define BSD_SYSTEM 199306
-#elif __FreeBSD__ >= 3
-#define BSD_SYSTEM 199506
-#endif
-#endif
-
-/* Define AMPERSAND_FULL_NAME if you use the convention
-   that & in the full name stands for the login id.  */
-#undef AMPERSAND_FULL_NAME
-
-/* `subprocesses' should be defined if you want to
-   have code for asynchronous subprocesses
-   (as used in M-x compile and M-x shell).
-   Only MSDOS does not support this (it overrides
-   this in its config_opsysfile below).  */
-
-#define subprocesses
-
-/* Include the os dependent file.  */
-#ifdef config_opsysfile
-# include config_opsysfile
-#endif
-
-/* Mac OS X / GNUstep need a bit more pure memory.  Of the existing knobs,
-   SYSTEM_PURESIZE_EXTRA seems like the least likely to cause problems.  */
-#ifdef HAVE_NS
-#if defined NS_IMPL_GNUSTEP
-#  define SYSTEM_PURESIZE_EXTRA 30000
-#elif defined DARWIN_OS
-#  define SYSTEM_PURESIZE_EXTRA 200000
-#endif
-#endif
-
-#ifdef emacs /* Don't do this for lib-src.  */
-/* Tell regex.c to use a type compatible with Emacs.  */
-#define RE_TRANSLATE_TYPE Lisp_Object
-#define RE_TRANSLATE(TBL, C) CHAR_TABLE_TRANSLATE (TBL, C)
-#ifdef make_number
-/* If make_number is a macro, use it.  */
-#define RE_TRANSLATE_P(TBL) (!EQ (TBL, make_number (0)))
-#else
-/* If make_number is a function, avoid it.  */
-#define RE_TRANSLATE_P(TBL) (!(INTEGERP (TBL) && XINT (TBL) == 0))
-#endif
-#endif
-
-#include <string.h>
-#include <stdlib.h>
-
-#if __GNUC__ >= 3  /* On GCC 3.0 we might get a warning.  */
-#define NO_INLINE __attribute__((noinline))
-#else
-#define NO_INLINE
-#endif
-
-#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1))
-#define EXTERNALLY_VISIBLE __attribute__((externally_visible))
-#else
-#define EXTERNALLY_VISIBLE
-#endif
-
-#if __GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 7)
-# define ATTRIBUTE_FORMAT(spec) __attribute__ ((__format__ spec))
-#else
-# define ATTRIBUTE_FORMAT(spec) /* empty */
-#endif
-
-#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)
-# define ATTRIBUTE_FORMAT_PRINTF(formatstring_parameter, first_argument) \
-   ATTRIBUTE_FORMAT ((__gnu_printf__, formatstring_parameter, first_argument))
-#else
-# define ATTRIBUTE_FORMAT_PRINTF(formatstring_parameter, first_argument) \
-   ATTRIBUTE_FORMAT ((__printf__, formatstring_parameter, first_argument))
-#endif
-
-#define ATTRIBUTE_CONST _GL_ATTRIBUTE_CONST
-
-/* Some versions of GNU/Linux define noinline in their headers.  */
-#ifdef noinline
-#undef noinline
-#endif
+/* Some platforms redefine this. */
+#undef wait3
+
+#include <conf_post.h>

 #endif /* EMACS_CONFIG_H */


=== modified file 'src/conf_post.h'
--- src/conf_post.h	2012-07-31 18:34:26 +0000
+++ src/conf_post.h	2012-07-31 19:33:28 +0000
@@ -27,11 +27,13 @@

 /* Code: */

+#ifndef WINDOWSNT
 /* On AIX 3 this must be included before any other include file.  */
 #include <alloca.h>
 #if ! HAVE_ALLOCA
 # error "alloca not available on this machine"
 #endif
+#endif

 #ifdef SIGNAL_H_AHB
 #undef SIGNAL_H_AHB

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 9:35 PM, Glenn Morris <rgm@gnu.org> wrote:

> I'd be fine with the absolute minimum change:
>
> mv s/msdos.h conf_msdos.h
>
> and change config_opsysfile on msdos from "s/msdos.h" to "conf_msdos.h".

Well, I have no objection to moving src/s/ms-w32.h to src/conf_w32.h
if you really want to get rid of src/s.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 7:31 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> Either ifdef it in conf_post.h, or provide alloca.h in nt/inc.

Hmmm.

The code in conf_post.h not just includes alloca.h, but checks that
HAVE_ALLOCA is defined, and we don't want that. So adding
nt/inc/alloca.h is not enough, unless we want to add "#undef
HAVE_ALLOCA" in ms-w32.h, which is piling ugliness upon ugliness.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: lekktu@gmail.com,  emacs-devel@gnu.org
> Date: Tue, 31 Jul 2012 15:28:27 -0400
> 
> Eli Zaretskii wrote:
> 
> > Is this stuff going to be left in a separate header for the observable
> > future?
> 
> I think so.
> 
> > What about SYSTEM_PURESIZE_EXTRA?  conf_post.h defines that for NS and
> > Darwin; should I move the msdos.h definitions there as well?
> 
> I think it's up to you.

Well, it always was ;-)  But I hoped you will have some guidance.

> > Or what about NULL_DEVICE or SEPCHAR? should they be autoconfiscated
> 
> I think so (ditto USER_FULL_NAME, etc), then msdos could sed them.

Exactly.

> AFAICS, however, unless you do some more complex sed stuff, you are
> always going to need a separate msdos.h, due to 'define DOS_NT' etc.

Sed has the 'i' command, you know ;-)

Like I said: there are several ways to go about this, but I'd prefer
one that is the cleanest, as being part of a cleanup provides more
motivation than being part of just rearranging the dirt ;-)

> So how much you want to remove from msdos.h is up to you.

If you want me, I can remove everything.  But then I don't want to see
NS-specific SYSTEM_PURESIZE_EXTRA, or any similar things, anywhere
near conf_post.h, either.  Because "quod licet Iovi non licet bovi" is
a game I prefer not to play, if I can avoid it.  (But I will if you
ask.)

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: lekktu@gmail.com,  emacs-devel@gnu.org
> Date: Tue, 31 Jul 2012 15:35:04 -0400
> 
> 
> I'd be fine with the absolute minimum change:
> 
> mv s/msdos.h conf_msdos.h
> 
> and change config_opsysfile on msdos from "s/msdos.h" to "conf_msdos.h".

That's rearranging the dirt in my book.  Why bother?  Do we care so
much about one more directory in the tree?

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Tue, 31 Jul 2012 21:58:52 +0200
> Cc: emacs-devel@gnu.org
> 
> On Tue, Jul 31, 2012 at 7:31 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
> > Either ifdef it in conf_post.h, or provide alloca.h in nt/inc.
> 
> Hmmm.
> 
> The code in conf_post.h not just includes alloca.h, but checks that
> HAVE_ALLOCA is defined, and we don't want that.

We could define HAVE_ALLOCA in nt/inc/alloca.h, could we not?

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 10:06 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> We could define HAVE_ALLOCA in nt/inc/alloca.h, could we not?

Well, of course.

I had assumed that defining it would have consecuences, but it seems
like HAVE_ALLOCA is not really used anywhere (regex.c uses
HAVE_ALLOCA_H).

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Eli Zaretskii wrote:

> That's rearranging the dirt in my book.  Why bother?  Do we care so
> much about one more directory in the tree?

I think it's ugly to have a directory with two platform-specific files
in, one specific to msdos, one specific to ms-windows, especially when
msdos and ms-windows both have their own dedicated directories (at
top-level) for the other platform-specific files that they need.

So personally I would move it to msdos/ rather than src/.
But if you don't want to bother, then don't bother.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Juanma Barranquero wrote:

> To fix 2) it would be enough to #undef it on ms-w32.h, but alas,
> post_conf.h is included in config.in / config.nt after
> config_opsysfile.

You are free to simply move the inclusion of config_opsysfile file to
the top of conf_post.h, since no non-MS platform uses config_opsysfile
any more, so it does not matter to anyone else at what point it gets
included.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Paul Eggert]

On 07/31/2012 12:35 PM, Glenn Morris wrote:
> I'd be fine with the absolute minimum change:
> 
> mv s/msdos.h conf_msdos.h

It'd be better to move it to the msdos/ directory,
since it's MS-DOS specific.  Similarly for the other
include file.  This will help localize the system-specific
information.

Something like this:

=== modified file 'admin/MAINTAINERS'
--- admin/MAINTAINERS	2012-05-22 16:20:27 +0000
+++ admin/MAINTAINERS	2012-07-31 20:31:34 +0000
@@ -34,7 +34,6 @@
 	    src/msdos.[ch]
 	    src/dosfns.[ch]
 	    src/w16select.c
-	    src/s/msdos.h
 	    lisp/term/internal.el
 	    lisp/term/pc-win.el
 	    lisp/dos-fns.el

=== modified file 'lib-src/make-docfile.c'
--- lib-src/make-docfile.c	2012-07-10 21:48:34 +0000
+++ lib-src/make-docfile.c	2012-07-31 20:31:34 +0000
@@ -80,7 +80,7 @@
 static void write_globals (void);
 
 #ifdef MSDOS
-/* s/msdos.h defines this as sys_chdir, but we're not linking with the
+/* msdos/msdos.h defines this as sys_chdir, but we're not linking with the
    file where that function is defined.  */
 #undef chdir
 #endif

=== modified file 'lib-src/makefile.w32-in'
--- lib-src/makefile.w32-in	2012-07-29 15:53:31 +0000
+++ lib-src/makefile.w32-in	2012-07-31 20:31:34 +0000
@@ -360,7 +360,7 @@
 NT_INC		= $(EMACS_ROOT)/nt/inc
 GNU_LIB		= $(EMACS_ROOT)/lib
 
-MS_W32_H       = $(SRC)/s/ms-w32.h \
+MS_W32_H       = $(SRC)/../nt/ms-w32.h \
 		 $(NT_INC)/sys/stat.h
 CONFIG_H       = $(SRC)/config.h \
 		 $(MS_W32_H)

=== modified file 'lib/makefile.w32-in'
--- lib/makefile.w32-in	2012-07-29 08:18:29 +0000
+++ lib/makefile.w32-in	2012-07-31 20:31:34 +0000
@@ -71,7 +71,7 @@
 
 C_CTYPE_H      = $(GNU_LIB)/c-ctype.h \
 		 $(NT_INC)/stdbool.h
-MS_W32_H       = $(SRC)/s/ms-w32.h \
+MS_W32_H       = $(SRC)/../nt/ms-w32.h \
 		 $(NT_INC)/sys/stat.h
 CONFIG_H       = $(SRC)/config.h \
 		 $(MS_W32_H)

=== modified file 'lisp/emacs-lisp/authors.el'
--- lisp/emacs-lisp/authors.el	2012-07-10 01:11:08 +0000
+++ lisp/emacs-lisp/authors.el	2012-07-31 20:31:34 +0000
@@ -578,6 +578,7 @@
     ("w32console.c" . "w32term.c")
     ("unexnt.c" . "unexw32.c")
     ("s/windowsnt.h" . "s/ms-w32.h")
+    ("s/ms-w32.h" . "nt/ms-w32.h")
     ("winnt.el" . "w32-fns.el")
     ("config.emacs" . "configure")
     ("configure.in" . "configure.ac")

=== modified file 'msdos/ChangeLog'
--- msdos/ChangeLog	2012-07-28 23:05:32 +0000
+++ msdos/ChangeLog	2012-07-31 20:31:34 +0000
@@ -1,3 +1,7 @@
+2012-07-31  Paul Eggert  <eggert@cs.ucla.edu>
+
+	* msdos.h: Rename from ../src/s/msdos.h.  All uses changed.
+
 2012-07-28  Paul Eggert  <eggert@cs.ucla.edu>
 
 	Use Gnulib stdalign module (Bug#9772, Bug#9960).

=== modified file 'msdos/mainmake.v2'
--- msdos/mainmake.v2	2012-06-30 15:32:51 +0000
+++ msdos/mainmake.v2	2012-07-31 20:31:34 +0000
@@ -162,7 +162,7 @@
 	../bin/etags --include=../lisp/TAGS \
 	 --regex='/[ 	]*DEFVAR_[A-Z_ 	(]+"\([^"]+\)"/' \
 	 $(CURDIR)/src/*.c $(CURDIR)/src/*.h \
-         $(CURDIR)/src/s/msdos.h
+         $(CURDIR)/src/../msdos/msdos.h
 	cd ..
 	./bin/etags --include=src/TAGS
 

=== renamed file 'src/s/msdos.h' => 'msdos/msdos.h'
=== modified file 'msdos/sed1v2.inp'
--- msdos/sed1v2.inp	2012-07-14 10:03:30 +0000
+++ msdos/sed1v2.inp	2012-07-31 20:31:34 +0000
@@ -129,7 +129,7 @@
 /^MKDEPDIR *=/s/@MKDEPDIR@//
 /^version *=/s/@[^@\n]*@//
 /^M_FILE *=/s!@M_FILE@!$(srcdir)/m/intel386.h!
-/^S_FILE *=/s!@S_FILE@!$(srcdir)/s/msdos.h!
+/^S_FILE *=/s!@S_FILE@!$(srcdir)/../msdos/msdos.h!
 /^@SET_MAKE@$/s/@SET_MAKE@//
 /^	[ 	]*\$(libsrc)\/make-docfile.*>.*\/DOC/s!make-docfile!make-docfile -o $(etc)/DOC!
 /^	[ 	]*\$(libsrc)\/make-docfile.*>.*gl-tmp/s!make-docfile!make-docfile -o gl-tmp!

=== modified file 'msdos/sed2v2.inp'
--- msdos/sed2v2.inp	2012-07-28 23:05:32 +0000
+++ msdos/sed2v2.inp	2012-07-31 20:31:34 +0000
@@ -90,7 +90,7 @@
 s/^#undef STACK_DIRECTION *$/#define STACK_DIRECTION -1/
 s/^#undef EMACS_CONFIGURATION *$/#define EMACS_CONFIGURATION "i386-pc-msdosdjgpp"/
 s/^#undef EMACS_CONFIG_OPTIONS *$/#define EMACS_CONFIG_OPTIONS "msdos"/
-s!^#undef config_opsysfile *$!#define config_opsysfile "s/msdos.h"!
+s!^#undef config_opsysfile *$!#define config_opsysfile "../msdos/msdos.h"!
 s!^#undef config_machfile *$!#define config_machfile "m/intel386.h"!
 s/^#undef PROTOTYPES *$/#define PROTOTYPES 1/
 s/^#undef POINTER_TYPE *$/#define POINTER_TYPE void/

=== modified file 'nt/ChangeLog'
--- nt/ChangeLog	2012-07-29 16:56:18 +0000
+++ nt/ChangeLog	2012-07-31 20:31:34 +0000
@@ -1,3 +1,7 @@
+2012-07-31  Paul Eggert  <eggert@cs.ucla.edu>
+
+	* ms-w32.h: Rename from ../src/s/ms-w32.h.  All uses changed.
+
 2012-07-29  Juanma Barranquero  <lekktu@gmail.com>
 
 	* config.nt: Sync with autogen/config.in.

=== modified file 'nt/config.nt'
--- nt/config.nt	2012-07-29 16:56:18 +0000
+++ nt/config.nt	2012-07-31 20:31:34 +0000
@@ -22,7 +22,7 @@
    This file is intentionally kept in sync with autogen/config.in to
    ease maintenance.  Please do not remove non-Windows related stuff
    unless strictly necessary.  Also, before adding anything here
-   consider whether src/s/ms-w32.h would be a better place; this is
+   consider whether nt/ms-w32.h would be a better place; this is
    particularly true for gcc vs. MSVC conditional defines, MinGW or
    MSVC specific code, and macros not already defined in config.in.  */
 
@@ -1484,7 +1484,7 @@
 #undef __restrict_arr
 
 /* Define to the used os dependent file. */
-#define config_opsysfile "s/ms-w32.h"
+#define config_opsysfile "../nt/ms-w32.h"
 
 /* A replacement for va_copy, if needed.  */
 #define gl_va_copy(a,b) ((a) = (b))

=== renamed file 'src/s/ms-w32.h' => 'nt/ms-w32.h'
=== modified file 'src/makefile.w32-in'
--- src/makefile.w32-in	2012-07-30 21:20:43 +0000
+++ src/makefile.w32-in	2012-07-31 20:31:34 +0000
@@ -339,7 +339,7 @@
 ##
 ## This works only with GNU Make.
 
-TAGS: $(OBJ0) $(OBJ1) $(OBJ2) $(CURDIR)/s/ms-w32.h
+TAGS: $(OBJ0) $(OBJ1) $(OBJ2) $(CURDIR)/../nt/ms-w32.h
 	$(MAKE) $(MFLAGS) TAGS-$(MAKETYPE)
 
 TAGS-LISP: $(OBJ0) $(OBJ1) $(OBJ2)
@@ -353,7 +353,7 @@
 	  $(OBJ1_c)
 	../lib-src/$(BLD)/etags.exe -a --regex=@../nt/emacs-src.tags \
 	  $(OBJ2_c) \
-	  $(CURDIR)/*.h $(CURDIR)/s/ms-w32.h
+	  $(CURDIR)/*.h $(CURDIR)/../nt/ms-w32.h
 
 TAGS-nmake:
 	echo This target is not supported with NMake
@@ -403,7 +403,7 @@
 		 $(GNU_LIB)/verify.h
 CODING_H       = $(SRC)/coding.h \
 		 $(SRC)/composite.h
-MS_W32_H       = $(SRC)/s/ms-w32.h \
+MS_W32_H       = $(SRC)/../nt/ms-w32.h \
 		 $(NT_INC)/sys/stat.h
 CONFIG_H       = $(SRC)/config.h \
 		 $(SRC)/conf_post.h \

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Though if I were you, I'd simply insert s/ms-w32.h into config.nt and
delete the former. Put a comment block in the relevant point in
conf_post.h if you like:

/* In config.nt, platform specific code starts here.  */

/* Platform specific code ends here.  */


Then config.nt is the same, but with the former contents of s/ms-w32.h
between the comments.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Paul Eggert]

On 07/31/2012 01:36 PM, Glenn Morris wrote:
> Then config.nt is the same, but with the former contents of s/ms-w32.h
> between the comments.

Even better.  That's simpler than my proposal.
Similarly for msdos.h, I assume.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Tue, Jul 31, 2012 at 10:36 PM, Glenn Morris <rgm@gnu.org> wrote:

> Though if I were you, I'd simply insert s/ms-w32.h into config.nt and
> delete the former. Put a comment block in the relevant point in
> conf_post.h if you like:
>
> /* In config.nt, platform specific code starts here.  */
>
> /* Platform specific code ends here.  */
>
>
> Then config.nt is the same, but with the former contents of s/ms-w32.h
> between the comments.

I'm not sure I understand.

Are you suggesting moving s/ms-32.h into nt/config.nt (this I get),
and then *not* including src/conf_post.h from nt/config.nt?

Because otherwise adding s/ms-w32.h at the end of nt/config.nt will
reverse the current order between conf_post.h and ms-w32.h.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

Juanma Barranquero wrote:

> Are you suggesting moving s/ms-32.h into nt/config.nt (this I get),

Yes.

> and then *not* including src/conf_post.h from nt/config.nt?

No.

> Because otherwise adding s/ms-w32.h at the end of nt/config.nt will
> reverse the current order between conf_post.h and ms-w32.h.

You have complete freedom to include ms-w32.h at whatever point you
like. It needs to be earlier than it is now if you want to define away
the alloca.h business cleanly. The position it ends up in has no
relevance to or effect on any non-Windows platform.


*** nt/config.nt	2012-07-29 16:56:18 +0000
--- nt/config.nt	2012-07-31 22:46:59 +0000
***************
*** 1483,1491 ****
     declarations. Define as empty for no equivalent. */
  #undef __restrict_arr
  
- /* Define to the used os dependent file. */
- #define config_opsysfile "s/ms-w32.h"
- 
  /* A replacement for va_copy, if needed.  */
  #define gl_va_copy(a,b) ((a) = (b))
  
--- 1483,1488 ----
***************
*** 1586,1593 ****
  #undef vfork
  
  
! /* Not valid on Windows.  */
! #if 0
  /* On AIX 3 this must be included before any other include file.  */
  #include <alloca.h>
  #if ! HAVE_ALLOCA
--- 1583,1592 ----
  #undef vfork
  
  
! /* Insert contents of s/ms-w32.h here.  .*/
! 
! 
! #ifndef WINDOWSNT
  /* On AIX 3 this must be included before any other include file.  */
  #include <alloca.h>
  #if ! HAVE_ALLOCA
***************
*** 1620,1630 ****
  
  #define subprocesses
  
- /* Include the os dependent file.  */
- #ifdef config_opsysfile
- # include config_opsysfile
- #endif
- 
  /* Mac OS X / GNUstep need a bit more pure memory.  Of the existing knobs,
     SYSTEM_PURESIZE_EXTRA seems like the least likely to cause problems.  */
  #ifdef HAVE_NS
--- 1619,1624 ----

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Wed, Aug 1, 2012 at 12:50 AM, Glenn Morris <rgm@gnu.org> wrote:

> You have complete freedom to include ms-w32.h at whatever point you
> like. It needs to be earlier than it is now if you want to define away
> the alloca.h business cleanly.

OK, that was the other alternative. It's quite ugly, and makes diffing
autogen/config.in and nt/config.nt messier. I much prefer to keep the
config_opsysfile stuff, or at the very least add #include
"../nt/ms-w32.h" so it's just one line.

> The position it ends up in has no
> relevance to or effect on any non-Windows platform.

I wasn't really worried about that.

    Juanma

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Glenn Morris]

I made the changes to conf_post.h to try to clarify what I meant.

Now I'm giving up on this.

Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109306: Use an include file in configure.ac's AH_BOTTOM

[Juanma Barranquero]

On Wed, Aug 1, 2012 at 12:50 AM, Glenn Morris <rgm@gnu.org> wrote:

> *** nt/config.nt        2012-07-29 16:56:18 +0000
> --- nt/config.nt        2012-07-31 22:46:59 +0000
> ***************
[...]
> ! /* Not valid on Windows.  */
> ! #if 0
>   /* On AIX 3 this must be included before any other include file.  */
>   #include <alloca.h>
>   #if ! HAVE_ALLOCA
> --- 1583,1592 ----
>   #undef vfork
>
>
> ! /* Insert contents of s/ms-w32.h here.  .*/
> !
> !
> ! #ifndef WINDOWSNT
>   /* On AIX 3 this must be included before any other include file.  */
>   #include <alloca.h>
>   #if ! HAVE_ALLOCA

How does this bit fix the problem of the #include <alloca.h> in
conf_post.h?  The #ifndef WINDOWSNT / #endif around that include
should be in conf_post.h, not config.nt.

    Juanma