[w32] Build failure on master (pthread_sigmask.h)

[w32] Build failure on master (pthread_sigmask.h)

[Richard Copley]

[-- Attachment #1: Type: text/plain, Size: 894 bytes --]

Emacs master fails to build in MinGW-W64 (MSYS2).

The cause seems to be that since the configuration changes in commit
"Update from Gnulib" of Wed May 29 16:33:51 2019 -0700, we now include
pthread_signal.h (<
https://sourceforge.net/p/mingw-w64/mingw-w64/ci/master/tree/mingw-w64-libraries/winpthreads/include/pthread_signal.h>)
which does this:

#define pthread_sigmask(H, S1, S2) 0

The result:

gcc  -c -mtune=generic   -DUSE_CRT_DLL=1 -I /c/projects/emacs/nt/inc
-mtune=generic  -MMD -MF deps/pthread_sigmask.d -MP -fno-common -O2 -I.
-I../src -I. -I./../src  pthread_sigmask.c
In file included from
C:/msys64/mingw64/x86_64-w64-mingw32/include/signal.h:10,
                 from pthread_sigmask.c:20:
pthread_sigmask.c:34:1: error: expected identifier or '(' before numeric
constant
   34 | pthread_sigmask (int how, const sigset_t *new_mask, sigset_t
*old_mask)
      | ^~~~~~~~~~~~~~~

[-- Attachment #2: Type: text/html, Size: 1230 bytes --]

Re: [w32] Build failure on master (pthread_sigmask.h)

[Eli Zaretskii]

> From: Richard Copley <rcopley@gmail.com>
> Date: Thu, 30 May 2019 14:04:26 +0100
> 
> Emacs master fails to build in MinGW-W64 (MSYS2).
> 
> The cause seems to be that since the configuration changes in commit "Update from Gnulib" of Wed May 29
> 16:33:51 2019 -0700, we now include pthread_signal.h
> (<https://sourceforge.net/p/mingw-w64/mingw-w64/ci/master/tree/mingw-w64-libraries/winpthreads/include/pthread_signal.h>)
> which does this:
> 
> #define pthread_sigmask(H, S1, S2) 0
> 
> The result:
> 
> gcc  -c -mtune=generic   -DUSE_CRT_DLL=1 -I /c/projects/emacs/nt/inc -mtune=generic  -MMD -MF
> deps/pthread_sigmask.d -MP -fno-common -O2 -I. -I../src -I. -I./../src  pthread_sigmask.c
> In file included from C:/msys64/mingw64/x86_64-w64-mingw32/include/signal.h:10,
>                  from pthread_sigmask.c:20:
> pthread_sigmask.c:34:1: error: expected identifier or '(' before numeric constant
>    34 | pthread_sigmask (int how, const sigset_t *new_mask, sigset_t *old_mask)
>       | ^~~~~~~~~~~~~~~

This doesn't happen to me, but I don't use MinGW64.

Can you spot the part of the changes which caused us to include
pthread_signal.h?  We shouldn't include any part of pthreads in the
Windows build.

Thanks.

Re: [w32] Build failure on master (pthread_sigmask.h)

[Richard Copley]

[-- Attachment #1: Type: text/plain, Size: 1498 bytes --]

On Thu, 30 May 2019 at 15:12, Eli Zaretskii <eliz@gnu.org> wrote:

> > From: Richard Copley <rcopley@gmail.com>
> > Date: Thu, 30 May 2019 14:04:26 +0100
> >
> > Emacs master fails to build in MinGW-W64 (MSYS2).
> >
> > The cause seems to be that since the configuration changes in commit
> "Update from Gnulib" of Wed May 29
> > 16:33:51 2019 -0700, we now include pthread_signal.h
> > (<
> https://sourceforge.net/p/mingw-w64/mingw-w64/ci/master/tree/mingw-w64-libraries/winpthreads/include/pthread_signal.h
> >)
> > which does this:
> >
> > #define pthread_sigmask(H, S1, S2) 0
> >
> > The result:
> >
> > gcc  -c -mtune=generic   -DUSE_CRT_DLL=1 -I /c/projects/emacs/nt/inc
> -mtune=generic  -MMD -MF
> > deps/pthread_sigmask.d -MP -fno-common -O2 -I. -I../src -I. -I./../src
> pthread_sigmask.c
> > In file included from
> C:/msys64/mingw64/x86_64-w64-mingw32/include/signal.h:10,
> >                  from pthread_sigmask.c:20:
> > pthread_sigmask.c:34:1: error: expected identifier or '(' before numeric
> constant
> >    34 | pthread_sigmask (int how, const sigset_t *new_mask, sigset_t
> *old_mask)
> >       | ^~~~~~~~~~~~~~~
>
> This doesn't happen to me, but I don't use MinGW64.
>
> Can you spot the part of the changes which caused us to include
> pthread_signal.h?  We shouldn't include any part of pthreads in the
> Windows build.
>

Not beyond the obvious (it's the pthread_sigmask.m4 changes), no. Sorry.
Prior to this change, lib\pthread_sigmask.c wasn't compiled at all AFAICS.

[-- Attachment #2: Type: text/html, Size: 2272 bytes --]

Re: [w32] Build failure on master (pthread_sigmask.h)

[Eli Zaretskii]

> From: Richard Copley <rcopley@gmail.com>
> Date: Thu, 30 May 2019 16:54:13 +0100
> Cc: Emacs Development <emacs-devel@gnu.org>
> 
>  > gcc  -c -mtune=generic   -DUSE_CRT_DLL=1 -I /c/projects/emacs/nt/inc -mtune=generic  -MMD -MF
>  > deps/pthread_sigmask.d -MP -fno-common -O2 -I. -I../src -I. -I./../src  pthread_sigmask.c
>  > In file included from C:/msys64/mingw64/x86_64-w64-mingw32/include/signal.h:10,
>  >                  from pthread_sigmask.c:20:
>  > pthread_sigmask.c:34:1: error: expected identifier or '(' before numeric constant
>  >    34 | pthread_sigmask (int how, const sigset_t *new_mask, sigset_t *old_mask)
>  >       | ^~~~~~~~~~~~~~~
> 
>  This doesn't happen to me, but I don't use MinGW64.
> 
>  Can you spot the part of the changes which caused us to include
>  pthread_signal.h?  We shouldn't include any part of pthreads in the
>  Windows build.
> 
> Not beyond the obvious (it's the pthread_sigmask.m4 changes), no. Sorry.
> Prior to this change, lib\pthread_sigmask.c wasn't compiled at all AFAICS. 

We need to augment nt/mingw-cfg.site to bypass pthread_sigmask, there
are already some settings there to that effect, we need to adapt them
to this Gnulib change.  Maybe try adding this line to mingw-cfg.site:

  gl_cv_func_pthread_sigmask_macro=no

If that doesn't help, please show the relevant parts of your
config.log.  Anything that matches "pthread".

(On my box the problem doesn't happen because I don't have the
pthread.h header polluting my system.)

Re: [w32] Build failure on master (pthread_sigmask.h)

[Richard Copley]

[-- Attachment #1: Type: text/plain, Size: 784 bytes --]

On Thu, 30 May 2019 at 19:43, Eli Zaretskii <eliz@gnu.org> wrote:

    We need to augment nt/mingw-cfg.site to bypass pthread_sigmask, there
    are already some settings there to that effect, we need to adapt them
    to this Gnulib change.  Maybe try adding this line to mingw-cfg.site:

      gl_cv_func_pthread_sigmask_macro=no

Yes, that works. (The build succeeds and no pthread_sigmask.o is generated.)
I'm impressed, thanks.

    If that doesn't help, please show the relevant parts of your
    config.log.  Anything that matches "pthread".

    (On my box the problem doesn't happen because I don't have the
    pthread.h header polluting my system.)

One package's pollution is another's prerequisite. Easily eradicated:

pacman -R ${MINGW_PACKAGE_PREFIX}-libwinpthread-git

[-- Attachment #2: Type: text/html, Size: 982 bytes --]

Re: [w32] Build failure on master (pthread_sigmask.h)

[Eli Zaretskii]

> From: Richard Copley <rcopley@gmail.com>
> Date: Thu, 30 May 2019 20:39:30 +0100
> Cc: Emacs Development <emacs-devel@gnu.org>
> 
>       gl_cv_func_pthread_sigmask_macro=no
> 
> Yes, that works. (The build succeeds and no pthread_sigmask.o is generated.)

Thanks for testing, pushed.

> I'm impressed, thanks.

There's no magic: I looked into my config.log and saw that the above
variable was set to "no" because I didn't have pthreads.h.

>     (On my box the problem doesn't happen because I don't have the
>     pthread.h header polluting my system.)
> 
> One package's pollution is another's prerequisite. Easily eradicated:
> 
> pacman -R ${MINGW_PACKAGE_PREFIX}-libwinpthread-git

Right.  FWIW, I've decided long ago that the trouble produced by
having pthreads headers and import libraries installed was way greater
than any advantages it gives.  The Windows ports of pthreads are
simply not reliable enough, IME.