pkgsrc-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: [2022Q3] mpich-3.4.3 compilation failure



>> > 	OK, I have found information on several web pages and I
>> > have added to option.mk:
>> >
>> > CONFIGURE_ARGS+=   FFLAGS=-m64 FCFLAGS=-m64
>> >
>> > 	In /etc/mk.conf, I have added:
>> >
>> > PKG_OPTIONS.mpich=f90  
>>
>> Hmm.  The package's options.mk already has
>>
>> PKG_SUGGESTED_OPTIONS=  f90
>>
>> so that option should be "default on".
>
> -m64 being necessary? Do you have totally empty flags otherwise? I
> build on Linux/x86-64, but also usually with something set in -march.

I see that I also built this package on NetBSD/amd64 9.99.97,
where I also got PLIST mismatches similar to what I did on the
initial test system (which was NetBSD/macppc 9.99.99 which
pretends to be 10.0).  The mpi-ch package built for me on
NetBSD/amd64 without the above additional CONFIGURE_ARGS.

On both systems the build ended up depending on gcc10 for
gfortran (NetBSD doesn't supply an OS-bundled fortran compiler),
and according to the bundled man page in the gcc10 package that
variant should support both f2008 and f2018 in addition to the
older variants.  The gfortran compiler accepts both standard
settings.

>> Conditionalize use of AI_V4MAPPED (not present on NetBSD).
>
> Are we talking about NetBSD on the initial report?

Don't know, but seems somewhat likely.  Bertrand?  I'll look a
bit more into the "official optionalness" of this symbol, I seem
to recall that getaddrinfo() is one of the rare APIs documented
in an RFC.

> I welcome patches for potability (of course also presented
> upstream). I worked on the MPI packages, but need other people
> to do this testing on NetBSD. I remember feedback on that
> before the branch, though.

I sent an e-mail to the package MAINTAINER but have not received
a reply so far...

> It seems, though, that we don't deal with NetBSD specifics
> here, but a compiler that does f90 without f08 support. I
> presumed that people would use a compiler that supports Fortran
> 2008 at least.

What makes you say that?  Ref. the above wrt. which compiler is
used.  Maybe I'll need to look further into why those files are
either not built or installed.

> What's your compiler version? Maybe we need a minimal
> requirement ... or add more complicated logic that finds
> Fortran 2008 support. There's no f08 language define for
> pkgsrc, or is there?

mk/gfortran.mk has some logic to pick which gcc version to depend
on for gfortran, but seems to only recognize the fortran and
fortran77 variants, and does not have any logic to pick or prefer
a specific gcc version depending on the standard level support
for any newer variants via the USE_LANGUAGES variable.

Best regards,

- Håvard


Home | Main Index | Thread Index | Old Index