Drop support for GCC < 6.3 in Sage 9.7

Question

Drop support for GCC < 6.3 in Sage 9.7

Closed this issue 3 years ago · 44 comments

mkoeppe commented 4 years ago

(see Meta-ticket #32074: Remove support for outdated distros)

This ticket can be used as a dependency for upgrade tickets that impose stricter compiler requirements.

CC: @dimpase @kliem

Component: build: configure

Author: Matthias Koeppe

Branch/Commit: 7c0072e

Reviewer: Jonathan Kliem

Issue created by migration from https://trac.sagemath.org/ticket/33316

Answer 1 · 2022-02-08T21:56:45.000Z

Dependencies: #33187

Answer 2 · 2022-02-08T21:59:23.000Z

Branch: u/mkoeppe/drop_support_for_gcc___6_3_in_sage_9_7

Answer 3 · 2022-02-08T22:07:35.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`bf96a8a`	`README.md, src/doc/en/installation/source.rst: Update info on GCC versions`

Answer 4 · 2022-02-08T22:07:35.000Z

Commit: bf96a8a

Answer 5 · 2022-02-08T22:08:03.000Z

Author: Matthias Koeppe

Answer 6 · 2022-02-09T20:52:15.000Z

Work Issues: Update platforms tested on GH Actions, best on top of #33062

Answer 7 · 2022-04-20T23:33:28.000Z

Changed commit from bf96a8a to 4e8b2c2

Answer 8 · 2022-04-20T23:33:28.000Z

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

`1124ded`	`build/pkgs/gcc/spkg-configure.m4: Bump minimum version to 6.3`
`4e8b2c2`	`README.md, src/doc/en/installation/source.rst: Update info on GCC versions`

Answer 9 · 2022-04-20T23:40:54.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`9b388ea`	`.github/workflows/tox*.yml: Update platforms for drop of GCC < 6.3`

Answer 10 · 2022-04-20T23:40:54.000Z

Changed commit from 4e8b2c2 to 9b388ea

Answer 11 · 2022-04-20T23:41:12.000Z

Changed work issues from Update platforms tested on GH Actions, best on top of #33062 to none

Answer 12 · 2022-04-20T23:43:01.000Z

Changed dependencies from #33187 to none

Answer 13 · 2022-04-21T00:14:36.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`e7ff2a8`	`build/pkgs/gcc/spkg-configure.m4: Fixup`

Answer 14 · 2022-04-21T00:14:36.000Z

Changed commit from 9b388ea to e7ff2a8

Answer 15 · 2022-04-21T06:13:19.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`0713f53`	`.github/workflows/tox*.yml: Replace ubuntu-xenial by ubuntu-xenial-toolchain-gcc_9`

Answer 16 · 2022-04-21T06:13:19.000Z

Changed commit from e7ff2a8 to 0713f53

Answer 17 · 2022-04-29T00:46:43.000Z

Reviewer: https://github.com/mkoeppe/sage/actions/runs/2242737566, https://github.com/mkoeppe/sage/actions/runs/2242737575

Answer 18 · 2022-04-29T04:26:19.000Z

comment:13

On ubuntu-trusty-toolchain-gcc_9-minimal (https://github.com/mkoeppe/sage/runs/6221011844) and -standard, nauty, ecm, ntl fail with
Error: operand size mismatch for vmovdqa64'orError: operand size mismatch for vmovdqu64'; and fflas_ffpack fails with many different operand size mismatch errors.

Answer 19 · 2022-04-30T17:20:58.000Z

comment:14

I'm guessing the problem is in https://github.com/linbox-team/fflas-ffpack/blob/b51626baf2715c8b8bcb0c334856b204f9a4a901/fflas-ffpack/fflas/fflas_simd/simd512_double.inl

They are calling _mm512_and_pd with const __mm512d instead of just __m512d. The problem should mostly go away, because stuff is inlined, but it might fail with g++ 9.

It seems to work with g++ 11.3.0 for me. At least I cannot reproduce the bug, even though I have available the avx512... instructions. But then again, it works for the other workflows, so this doesn't seem to be a problem in other cases. Just in the toolchain-gcc_9-minimal workflow.

Can we configure fflas_ffpack with --disable-avx512dq in this case?

Answer 20 · 2022-05-01T01:57:29.000Z

comment:15

I'll try first if installing binutils-2.26 (and setting PATH="/usr/lib/binutils-2.26/bin:$PATH") helps.

Answer 21 · 2022-05-02T22:43:29.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`d650177`	`Merge tag '9.6.rc2' into t/33316/drop_support_for_gcc___6_3_in_sage_9_7`
`a0e037f`	`build/bin/write-dockerfile.sh [debian]: Handle EXTRA_PATH, EXTRA_SYSTEM_PACKAGES`
`dae2d08`	`tox.ini (ubuntu-trusty-toolchain): Use binutils-2.26`

Answer 22 · 2022-05-02T22:43:29.000Z

Changed commit from 0713f53 to dae2d08

Answer 23 · 2022-05-02T22:46:02.000Z

Changed reviewer from https://github.com/mkoeppe/sage/actions/runs/2242737566, https://github.com/mkoeppe/sage/actions/runs/2242737575 to https://github.com/mkoeppe/sage/actions/runs/2260846479

Answer 24 · 2022-05-03T03:47:26.000Z

comment:18

Replying to @mkoeppe:

I'll try first if installing binutils-2.26 (and setting PATH="/usr/lib/binutils-2.26/bin:$PATH") helps.

It seems to work. Ready for review

Answer 25 · 2022-05-03T06:31:42.000Z

Changed commit from dae2d08 to 33ca131

Answer 26 · 2022-05-03T06:31:42.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`33ca131`	`tox.ini, .github/workflows/tox*.yml: Add ubuntu-kinetic, linuxmint-21, fedora-37`

Answer 27 · 2022-05-09T10:22:05.000Z

Changed reviewer from https://github.com/mkoeppe/sage/actions/runs/2260846479 to Jonathan Kliem

Answer 28 · 2022-05-09T10:22:05.000Z

comment:20

LGTM.

Answer 29 · 2022-05-09T17:43:01.000Z

comment:21

Thank you!

Answer 30 · 2022-05-22T09:56:56.000Z

comment:22

If configure doesn't find a supported version we should just install it instead of giving users instructions for other configure flags:

configure: error: 
    Given --with-system-gcc=force, but no system package could be used.
    That's an error.  Please install the indicated package to continue.
    (To override this error, use ./configure --without-system-gcc)

Answer 31 · 2022-05-22T11:00:39.000Z

comment:23

Replying to @vbraun:

If configure doesn't find a supported version we should just install it instead of giving users instructions for other configure flags:
configure: error: 
    Given --with-system-gcc=force, but no system package could be used.
    That's an error.  Please install the indicated package to continue.
    (To override this error, use ./configure --without-system-gcc)

Volker, it is not a bug, it is how =force option is meant to be.

Answer 32 · 2022-05-22T11:32:13.000Z

comment:24

For reference, this behavior was implemented in #32060.

Answer 33 · 2022-05-22T16:36:56.000Z

comment:25

Breaks the Debian 9 buildbot, we can merge this after I get around to update the buildbot config

Answer 34 · 2022-05-22T16:39:17.000Z

comment:26

See #32074 - we're dropping support for Debian 9.

Answer 35 · 2022-05-22T16:39:45.000Z

comment:27

Debian 9 is still LTS

Answer 36 · 2022-05-22T16:43:00.000Z

comment:28

Replying to @mkoeppe:

See #32074 - we're dropping support for Debian 9.

Sorry, I was off by one. Debian 9 is fine. What's the problem on your buildbot?

Answer 37 · 2022-05-22T16:44:03.000Z

comment:29

so needs_work but not on the ticket

Answer 38 · 2022-05-22T21:23:44.000Z

comment:30

Replying to @vbraun:

Debian 9 is still LTS

but only till 01.07.2022, so we should be dropping it in 9.7, I guess.

Answer 39 · 2022-05-24T01:06:58.000Z

Dependencies: #33888

Answer 40 · 2022-06-13T18:15:13.000Z

Changed dependencies from #33888 to none

Answer 41 · 2022-06-13T18:15:21.000Z

comment:33

Merge failure on top of:

08609a40ad Trac #33829: sage.graphs: Do not use SAGE_TMP in doctests

eca52b2 Trac #33922: Add missing title to conf.py

e9bfcd2 Trac #33980: codespell in graphs: fix a few typos

28f70e0 Trac #31153: Fix latex of elements of multivariate polynomial quotient ring

b36c11c Trac #33983: sage_setup: Add missing dependency

6edac51 Trac #33970: Remove a period sneaked in the documentation website

6d60bf3 Trac #33977: fix pycodestyle E306 in geometry and schemes

edd7a74 Trac #33976: some pycodestyle fixes in pxd and pxi files

b0973af Trac #33975: get rid of commented old-style prints

b3d6a0b Trac #33956: some better .join with iterator inside

a33f2df Trac #33949: get rid of have_ring option in singular interface

ad1c3ac Trac #33946: fix pycodestyle E306 in rings and algebras

4942ca9 Trac #33945: fix pycodestyle E306 outside of algebras,rings,graphs,schemes,categories

8dad9ce Trac #30683: Nu Tamari

346b3b0 Trac #33968: Replace $...$ in docstrings by ...

f793558 Trac #33967: sage -t: Print git commit hash and SAGE_LOCAL, SAGE_VENV

0e48fec Trac #33963: wrong sign for symbolic Legendre polynomial at 0

34d562e Trac #33962: wrong sign for value of Legendre polynomial at 0

bd33f87 Trac #33961: compute square roots modulo powers of two in polynomial time

16af958 Trac #33940: enhance maple interface

8d921a4 Trac #33872: Upgrade memory allocator to 0.1.3

29499f8 Trac #33871: Don't run singular in spkg-configure.m4 when cross compiling

4cc978b Trac #33823: sage -t --optional='sage,!FEATURE'

6b6f0cf Trac #33776: move method distance_graph to graph.py

65e866a Trac #23443: More Schubert polynomial shenanigans

a6e696e Updated SageMath version to 9.7.beta2

merge was not clean: conflicts in .github/workflows/tox.yml

Answer 42 · 2022-06-13T18:29:17.000Z

Branch pushed to git repo; I updated commit sha1. New commits:

`7c0072e`	`Merge tag '9.7.beta2' into t/33316/drop_support_for_gcc___6_3_in_sage_9_7`

Answer 43 · 2022-06-13T18:29:17.000Z

Changed commit from 33ca131 to 7c0072e

Answer 44 · 2022-06-19T15:24:40.000Z

Changed branch from u/mkoeppe/drop_support_for_gcc___6_3_in_sage_9_7 to 7c0072e