SAGE_FAT_BINARY is broken in sage-9.6

[williamstein]

On Sage-9.5 I build Sage like this and it worked on everybody else's computers (under ubuntu 20.04 on x86_64):

export SAGE_FAT_BINARY="yes"
export SAGE_INSTALL_GCC="no"
cd sage
make configure
./configure
make

See https://github.com/sagemathinc/cocalc-docker/blob/master/scripts/install_sage.sh

I had thousands of downloads and zero complaints.

With Sage-9.6, the binaries are broken (yet again), so I must continue to ship sage-9.5 to my users of cocalc-docker until this is fixed.

Support for SAGE_FAT_BINARY is a critical piece of functionality for Sage, which appears 22 times in the codebase.

Evidently, I created SAGE_FAT_BINARY in the first place 13 years ago, and I think it's a key piece of Sage's functionality.

I'm also now also a "binary-providing downstream developer" and need this to work in order to provide binaries.

CC: @slel

Component: build

Keywords: SAGE_FAT_BINARY

Issue created by migration from https://trac.sagemath.org/ticket/33863

[mkoeppe]

comment:1

I think this just means that a quality assurance workflow is missing in your Docker-building scripts. I could imagine that a good way of ensuring portability is to (1) define the minimum architecture on which the Docker image should run; (2) provide a self-hosted runner; (3) set up some GH Actions that tests the built image with the runner.

[slel]

Description changed:

--- 
+++ 
@@ -15,7 +15,7 @@
 
 With Sage-9.6, the binaries are broken (yet again), so I must continue to ship sage-9.5 to my users of cocalc-docker until this is fixed.
 
-Support for SAGE_FAT_BINARY is a critical piece of functionality for Sage, which [appears 22 times in the codebase]()https://github.com/sagemath/sage/search?q=sage_fat_binary).
+Support for SAGE_FAT_BINARY is a critical piece of functionality for Sage, which [appears 22 times in the codebase](https://github.com/sagemath/sage/search?q=sage_fat_binary).
 
 Evidently, [I created SAGE_FAT_BINARY in the first place 13 years ago](https://github.com/sagemath/sage-prod/commit/80bda10cfc3612a5fc570dd8b6d50de553f44183), and I think it's a key piece of Sage's functionality.
 

[slel]

Changed keywords from none to SAGE_FAT_BINARY

[williamstein]

comment:3

Just to be clear - I have not been able to reproduce this myself yet. It could be an error by one of my users, or they are using some obscure hardware. I don't know yet.

The user that reported this just tried the sage-9.5 binary I also built, and reported the same problem with that. Thus there might be no regression at all from sage-9.5 to sage-9.6. If I'm not able to reproduce this problem and identify a specific thing that is built improperly within a few weeks, then I agree that we should close this ticket as invalid.

[williamstein]

comment:4

More details: they are using a "Intel i7-4930K (12) @ 3.900GHz", which is a 2013 processor.

[williamstein]

comment:5

I've decided to keep cocalc-docker at sage-9.6 for the time being, since the only user who has clearly reported problems so far also can't use the sage-9.5 version either.

[williamstein]

comment:6

OK, upon further checking, I think that processor does not support AVX2, and SAGE_FAT_BINARY really does not work properly regarding AVX2, according to #32434

So this ticket is a dup of #32434, and I'm closing this. I hope that somebody fixes #32434 since it looks easy (?).