Summary: | math/lapack (and math/atlas) do not build with gcc10 | ||||||||
---|---|---|---|---|---|---|---|---|---|
Product: | Ports & Packages | Reporter: | gja822 | ||||||
Component: | Individual Port(s) | Assignee: | Thierry Thomas <thierry> | ||||||
Status: | Closed FIXED | ||||||||
Severity: | Affects Only Me | CC: | alt2600, gja822, thierry | ||||||
Priority: | --- | Flags: | koobs:
maintainer-feedback+
|
||||||
Version: | Latest | ||||||||
Hardware: | Any | ||||||||
OS: | Any | ||||||||
See Also: | https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=257122 | ||||||||
Bug Depends on: | 247542 | ||||||||
Bug Blocks: | |||||||||
Attachments: |
|
Description
gja822
2020-06-22 16:54:00 UTC
Also I've noticed this during build of atlas: /usr/local/bin/ranlib ../liblapack.a make[4]: "/usr/ports/Mk/bsd.gcc.mk" line 101: Malformed conditional (${PORTNAME} == gcc) make[4]: Fatal errors encountered -- cannot continue===> Cleaning for lapack-3.5.0_8 Is it O.K.? math/blas and math/lapack are lagging: upstream released 3.9.0 and we have still 3.5. I'm working on these ports, and it should fix the problem with recent compilers, but no ETA yet: blacs, lapacke, etc. must be upgraded together, and some exp-runs will be required. However I'm not working on math/atlas: contributions are welcome. By the way net/mpich2 exhibits similar problem (not compiling with gcc10, only with gcc8) with FORTRAN argument mismatch error. Created attachment 222281 [details]
path-Makefile
Fixes build with GCC10(+?). gfortran10 has apparently went the way of strict type checking, which will cause problems with 30+ year old fortran libraries. I added -fallow-argument-mismatch to the F77FLAGS line in the do-build: loop to make gfortran10 handle mismatches like and older fortran compiler would and atlas built. I attempted other methods like adding to EXTRAFLAGS but this seemed to break the C build. I'm guessing ATLAS checking the compiler flags causes issue when gcc10 spits back that this is a fortran flag not a C flag.
Either way it builds now fine. not sure how lapack was fixed but it built fine during a system update to 12.2 I just completed. Given this has no maintainer, I might consider taking it over. Just need to play a little to see how hard it would be to get the newest support 3.10.3 version going a little bit, and what lapacks it expects to see to optimize against. Not sure how far things have gone with upgrading lapack, lapacke, etc to later versions. I know those are tough given how many things require them.
Created attachment 222282 [details]
Makefile
Complete Makefile for those that might want to try things without dealing with a patch.
Fixed when upgrading to 3.9.1: see PR 247542. ^Triage: - Track bug 247542 as the dependent of this issue (" - Fix the build with Gcc10") and correct resolution - Assign to committer that resolved |