The nearby comment states "Avoid gnulib's test for pthread_sigmask", but
the method used doesn't seem to work.

What I see is that gl_INIT is called, which calls
gl_FUNC_PTHREAD_SIGMASK.

The method in the attached patch is the same as used for avoiding other
gnulib tests.

If the test is really to be avoided, then I would think it to be better
to remove m4/pthread_sigmask.m4 so that gl_INIT doesn't include the test
in the first place. It doesn't look like any of the code uses the result
of it.