Re: math/fftw-long / math/fftw-quad broken on Linux as it pulls in math/libquadmath, which is superfluous and broken on Linux

["Dr. Thomas Orgis" <thomas.orgis%uni-hamburg.de@localhost>]

Am Tue, 22 Jun 2021 13:13:23 +0000
schrieb nia <nia%NetBSD.org@localhost>: 

> We're already doing it for e.g. libatomic. It's pretty necessary
> because the alternative is shipping 10 different versions in pkgsrc.
> This is all over pkgsrc and not going away.

Well, an alternative would be to just ship the GCC libraries within
GCC. But I see that you want to work with the base compiler as much as
possible and not always replace the whole thing.

> It is extremely likely that you can get libquadmath to package with
> the correct library installation prefix by playing with the configure
> options. 

As I indicated in IRC, the only option I see is post-install or hacking
the generated Makefile … or add support to the wrapper to change the output of

	$CC -print-multi-os-directory

to ../lib/. It prints ../lib on Ubuntu, but ../lib64 on CentOS. This is
appended to --with-toolexeclibdir in configure, which apparently is
the directory used for libquadmath.la (for whatever reason).

But still: Installing a second libquadmath is no option for me. It is
part of the base toolchain, either with Linux itself or the toolchain I
built. It is breakage to throw in a second instance of this library.
Same for libatomic, I guess. Regarding that libomp for Darwin, I also
need be make pretty sure that dependencies to that don't get int the
way of my libgomp in the system. 

I guess all of these lirbraries that are part of a full toolchain, at
least those pulled from the GCC distribution, should get builtin.mk. 

Do you agree with that? Or can I convince you? ;-)

> The FreeBSD package for example has OpenMP disabled by default.
> I think we're unlikely to run into trouble.

You're confusing me. Either having parts of the ABI as options in a
library package is bad or not. As long as nothing explicitly depends on
fftw with MPI support or OpenMP (don't confuse these two), you won't
run into trouble, sure.

So is the rule ‚options in libraries are bad unless no other package
needs the option yet‘?


Alrighty then,

Thomas

-- 
Dr. Thomas Orgis
HPC @ Universität Hamburg