bug#32452: 26.1; gnutls_try_handshake maxes out cpu retrying when server is a bit busy

[Paul Eggert]

On 3/5/22 09:11, Lars Ingebrigtsen wrote:
>         $(AM_V_at)$(CC) -shared $(CPPFLAGS) $(MODULE_CFLAGS) $(LDFLAGS) \
>           -o $@ $< $(LIBGMP) \
> +         ../lib/libgnu.a \

Unfortunately that won't work in general; the comment for mod-test.so 
says "In the compilation command, we can't use any object or archive 
file as source because those are not compiled with -fPIC" but as things 
stood mod-test.so was being linked with libgnu.a.

Anyway, mod-test.so should be decoupled from Emacs as much as possible - 
that's the whole point of modules after all - so the mod-test.so rule 
shouldn't make assumptions about how Emacs internals use Gnulib. I 
installed the attached patches to lessen the coupling between 
mod-test.so and Emacs internals and I hope this fixes the problem. It 
may need to be adjusted a tad for platforms that lack 
nanosleep/clock_gettime/libgmp whatever, but the basic idea should work 
on these platforms with only minor twiddling.

Even with these patches, mod-test.so uses "config.h"; but at least it 
doesn't use Gnulib code.

0001-Do-not-include-attribute.h-from-config.h.patch
Description: Text Data

0002-Decouple-mod-test.c-from-Gnulib.patch
Description: Text Data