Unable to build sage

[Daniel Bump]

I'm unable to build Sage since yesterday on an iMac running mojave. 

I have tried (repeatedly) to build the develop branch and others including fusion_central_charge-29615. 

I mention the last one since there has been no activity on that branch since July 1 and I was able to

build it on this machine previously.

So the code on that branch has not changed, yet I can no longer build Sage even

from a clean directory.

What has changed locally is that Homebrew has been updated.

Frequently the build fails in scipy with a complaint unable to install with pip3. I am attaching

an excerpt from install.log.

Daniel Bump

[Dima Pasechnik]

Please post top config.log and also the full logs/pkgs/scipy-...log

--
You received this message because you are subscribed to the Google Groups "sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/a6e2c4fa-321c-49ff-b6fd-61534bd72051o%40googlegroups.com.

[Daniel Bump]

On Saturday, July 4, 2020 at 2:52:06 PM UTC-7, Dima Pasechnik wrote:

Please post top config.log and also the full logs/pkgs/scipy-...log

Attached.

Thanks,

Dan 

[Dima Pasechnik]

I guess this is due to gfortran 10.

We still do not support gcc 10, I think.

Can you downgrade it to gfortran 9?

--

You received this message because you are subscribed to the Google Groups "sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/117843bf-3793-47fb-a952-9e58ad31da9co%40googlegroups.com.

[Reimundo Heluani]

On Jul 04, Dima Pasechnik wrote:
>I guess this is due to gfortran 10.
>We still do not support gcc 10, I think.
>Can you downgrade it to gfortran 9?
>

For what its worth, here running arch,

1) removing line 153 from build/pkgs/gcc/spkg-configure.m4
2) adding a newer scipy: upstream/scipy-1.5.0.tar.gz
3) pulling pybind11 from this commit
https://git.sagemath.org/sage.git/commit/?id=fa20111935c171f1d99c410110967350f4facea2

allowed me to upgrade without a glitch so far

R.

>
>On Sat, 4 Jul 2020, 23:06 Daniel Bump, <[1]dwb...@gmail.com> wrote:
>
>
>
> On Saturday, July 4, 2020 at 2:52:06 PM UTC-7, Dima Pasechnik wrote:
>
> Please post top config.log and also the full logs/pkgs/scipy-...log
>
>
> Attached.
>
> Thanks,
> Dan 
>
> --
> You received this message because you are subscribed to the Google Groups
> "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an

> email to [2]sage-devel+...@googlegroups.com.
> To view this discussion on the web visit [3]https://groups.google.com/d/
> msgid/sage-devel/117843bf-3793-47fb-a952-9e58ad31da9co%40googlegroups.com.

>
>--
>You received this message because you are subscribed to the Google Groups
>"sage-devel" group.
>To unsubscribe from this group and stop receiving emails from it, send an email

>to [4]sage-devel+...@googlegroups.com.
>To view this discussion on the web visit [5]https://groups.google.com/d/msgid/
>sage-devel/
>CAAWYfq0cbnv0%3DpMchh2pPmnR%2BxaC_vVjB_Ve9NOmoJUtiLtEkQ%40mail.gmail.com.
>
>References:
>
>[1] mailto:dwb...@gmail.com
>[2] mailto:sage-devel+...@googlegroups.com
>[3] https://groups.google.com/d/msgid/sage-devel/117843bf-3793-47fb-a952-9e58ad31da9co%40googlegroups.com?utm_medium=email&utm_source=footer
>[4] mailto:sage-devel+...@googlegroups.com
>[5] https://groups.google.com/d/msgid/sage-devel/CAAWYfq0cbnv0%3DpMchh2pPmnR%2BxaC_vVjB_Ve9NOmoJUtiLtEkQ%40mail.gmail.com?utm_medium=email&utm_source=footer

[Zachary Scherr]

Seems to be related to https://github.com/scipy/scipy/issues/11611 and as was pointed out above was fixed in version 1.5.

I'd like to propose two workarounds for people who can no longer build sage and stumble into this thread (either should allow scipy 1.2.3 to build)

1). Use fortran 9:

brew install gcc@9

FC=gfortran-9 ./configure

make

2). Add a flag to fortran 10 (should work if you have gcc 10 regardless of whether you have homebrew or not):

./configure

FFLAGS=-fallow-argument-mismatch make

[Matthias Koeppe]

scipy update to 1.5.1 is now at https://trac.sagemath.org/ticket/29766

For gcc 10 support of the whole sage distribution we need at least one more package upgrade: sympow https://trac.sagemath.org/ticket/3360 

Meta-ticket on gcc 10 at https://trac.sagemath.org/ticket/29456

[Daniel Bump]

On Saturday, July 4, 2020 at 3:12:36 PM UTC-7, Dima Pasechnik wrote:

I guess this is due to gfortran 10.

We still do not support gcc 10, I think.

Can you downgrade it to gfortran 9?

I had gfortran10 which came with gcc. In installed gfortran (with Homebrew)

which gave me gfortran 8.2. I was then able to build sage.

Thanks!

Dan

 

[Dima Pasechnik]

> --
> You received this message because you are subscribed to the Google Groups "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.

> To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/1f6c7908-af7a-4438-b2da-4d8994d7ba25o%40googlegroups.com.

[dim...@gmail.com]

Your config.log says you're using gfortran 10.
Most probably it causes the failure to build scipy.

I see what happened:
https://formulae.brew.sh/cask/gfortran
says it's version 8.2, and that it conflicts with
Homebrew's gcc package, which is gcc10, which also gives you
gfortran10.

So you have a number of ways to proceed:
1) remove gcc package (you don't need it for Sage, as
We don't build Sage with "real" gcc on macOS for years. Xcode has gcc
which is actually clang, the default macOS C/C++ compiler, and this
might cause the confusion. One does need a fortran compiler, and we
support gfortran, which happens to be a part of gcc, so an extra
potential confusion here)
and re-install gfortran package (which should give you gfortran 8.2,
which works with Sage).
This is probably the quckest way to get a working Sage

2) Stay on gcc10, and merge https://trac.sagemath.org/ticket/29766
(currently marked as "needs work", due to some warning messages causing
some tests fail, nothing serious)
Cross your fingers, perhaps it will work.


HTH
Dima

[Dima Pasechnik]

An advantage of gcc10 that with it you can also use Homebrew's openblas, gsl,
perhaps R. Given that Homebrew's scipy package is built with gcc10,
this looks promising.

>
>
> HTH
> Dima
>
>