Hi Tim,
Thanks for the more verbose configure output.
I misinterpreted the config.log output; I now see how it was produced even
without --disable-threads.
The logic in threadlib.m4 is as follows:
- If no preference is given (no --enable-threads={posix|solaris|pth|windows}
options), the macro attempts to use POSIX threads if possible. This makes
sense in general, when you think of systems like Haiku or such.
- Since you have mingw's pthread.h installed, the logic goes this route.
The two errors that you encounter:
* The error
> >> In file included from /usr/share/mingw-w64/include/signal.h:10,
> >> from ./signal.h:52,
> >> from pthread_sigmask.c:20:
> >> pthread_sigmask.c:34:1: error: expected identifier or ‘(’ before numeric
> >> constant
> >> pthread_sigmask (int how, const sigset_t *new_mask, sigset_t *old_mask)
> >> ^~~~~~~~~~~~~~~
is caused by
#define pthread_sigmask(H, S1, S2) 0
in pthread_signal.h. You could probably easily avoid the error by
inserting a '#undef pthread_sigmask' at the appropriate place.
* The error
libtests.a(localename.o):localename.c:(.text+0x10): undefined reference to
`pthread_mutex_lock'
is apparently because -lpthread or -lpthread are not occurring on the
link command line. These link options should be transmitted through
LIBTHREAD, LTLIBTHREAD in config.status. But I don't see these variables
in the config.status that you sent me.
config.status contains a setting of LTLIBPTHREAD, which comes from this
line in gnutls-3.6.7/configure:
AC_LIB_HAVE_LINKFLAGS(pthread,, [#include <pthread.h>], [pthread_mutex_lock
(0);])
So this is just a different autoconf test. But it doesn't explain why
LIBTHREAD, LTLIBTHREAD are missing.
Btw, there are two(!) pthreads implementations for mingw.
- The one that ships with mingw in Debian is initially '(C) 2010 Lockless
Inc.'
and is under a BSD license.
- The one at https://sourceware.org/pthreads-win32/ is by Ross Johnson and
is under LGPL.
Another difference between the two is explained in
https://en.wikipedia.org/wiki/POSIX_Threads#POSIX_Threads_for_Windows
I'm not going to work on this. But if you have time, these would be some
improvements:
- Test a gnulib testdir in an environment where you have the mingw-pthreads
installed. Probably you'll hit the pthread_sigmask problem when you do this.
Possibly also some test failures.
- Build your preferred packages
once with --enable-threads=windows,
once with --enable-threads=posix and the mingw pthreads,
once with --enable-threads=posix and the pthreads-win32 package,
and compare/analyze/report the results.
Bruno
