Re: signbit(), C++, and Solaris 10

[G. Branden Robinson]

[please keep both lists in follow-ups]

At 2024-11-22T13:54:27-0800, Paul Eggert wrote:
> On 2024-11-22 10:23, G. Branden Robinson wrote:
> > Not the freshest release...
> 
> I checked, and the problem still occurs with bleeding-edge groff (your
> commit 2562cf23726fc154b657d35a1d856f944f8461ad dated Wed Nov 20
> 23:46:16 2024 -0600) and with Oracle Solaris 12.6. So it doesn't sound
> like it's a problem specific to compiler or groff version.
> 
> Also, you can't simply remove the include directives, as some .c /
> .cpp files will require that you add them.
> 
> I briefly tried to build with Oracle Developer Studio 12.6 but ran
> into other issues (e.g., with static_assert). Not being a C++ expert I
> can't help much here.

I have the following workaround.  Start the build.

$ gmake

Watch it fail on the first C++ source file that tries to #include
math.h.

$ gsed -i '3212,3291d' lib/math.h
$ gmake

The build will then run to successful completion.  Even the automated
tests pass.[1]

I'm attaching the deleted material.  This appears to be the entire
stanza of `GNULIB_defined_signbit`-related stuff.

Maybe Solaris 10 has been field-upgraded in some way that invalidates
tests that gnulib is doing?

Is there something I can specify to "configure" or make(1) to override
gnulib's decision here?

Regards,
Branden

[1] With one exception, depending on configuration, that I'm pretty
    confident has nothing to do with this.  See
    <https://savannah.gnu.org/bugs/?66504>.

signature.asc
Description: PGP signature