Sage silently resets `SAGE_NUM_THREADS` environment variable

[maxale]

Steps To Reproduce

Run Sage from shell via

SAGE_NUM_THREADS=8 /usr/local/sage/sage

and execute:

sage: import os
sage: os.environ["SAGE_NUM_THREADS"]
'1'

Expected Behavior

We should see value '8' that we set up.

Actual Behavior

Our value was overwritten with '1'.

Additional Information

The bug was introduced somewhere in between version 10.7.beta8 (which works as expected) and current 10.8.beta4.

Environment

• OS: Ubuntu 24.04.3 LTS
• Sage Version: 10.8.beta4

Checklist

• I have searched the existing issues for a bug report that matches the one I want to file, without success.
• I have read the documentation and troubleshoot guide

[antonio-rojas]

The bug is in https://github.com/sagemath/sage/blob/10.8.beta5/src/bin/sage-env#L583:

        # First, figure out the right values for SAGE_NUM_THREADS (default
        # number of threads) and SAGE_NUM_THREADS_PARALLEL (default number of
        # threads when parallel execution is asked explicitly).
        sage_num_threads_array=$(sage-num-threads.py 2>/dev/null || echo 1 2 1)

The path of sage-num-threads.py is not specified, so the script isn't found and the fallback 1 2 1 is always used.

However, this code hasn't been touched in over 5 years. Are you sure that this worked for you with the exact same setup?

[maxale]

@antonio-rojas Thank you for the analysis. I can confirm that the reported issue I observed only recently, so the reason may be that the control flow was somehow changed and the "old code" kicked in just now. Checking why this code did not break things in the release version 10.6 could shed light on what's going on behind the hood.

[antonio-rojas]

@antonio-rojas Thank you for the analysis. I can confirm that the reported issue I observed only recently, so the reason may be that the control flow was somehow changed and the "old code" kicked in just now. Checking why this code did not break things in the release version 10.6 could shed light on what's going on behind the hood.

Since the code in question did not change, the only possible explanation I can think of is that the dir where sage-num-threads.py lives was previously in PATH at the point where sage-env is imported, and it no longer is. Sounds like #39030 could be related, but I don't use sage-the-distro myself so I can't confirm.

[maxale]

I'm not sure what is sage-the-distro. I build Sage from sources from this github. The latest version, which I seen without this issue, was 10.7.beta8.

[maxale]

Since the code in question did not change, the only possible explanation I can think of is that the dir where sage-num-threads.py lives was previously in PATH at the point where sage-env is imported, and it no longer is.

I've searched sage-num-threads.py in two version I have access to. In version 10.7.beta8 (without the current issue), this file is found in two locations within main sage directory:

./src/bin/sage-num-threads.py
./local/var/lib/sage/venv-python3.12/bin/sage-num-threads.py

In version 10.8.beta4 having this issue, the file is found only in the first one:

./src/bin/sage-num-threads.py

When I've manually copied sage-num-threads.py to ./local/var/lib/sage/venv-python3.12/bin/, the issue disappeared.

So, the question is why sage-num-threads.py was in that directory before, but not anymore?

[dimpase]

I'm not sure what is sage-the-distro. I build Sage from sources from this github. The latest version, which I seen without this issue, was 10.7.beta8.

nowadays, if you have most of the needed packages installed, you can build sagelib without going through ./configure && make (this is "sage-the-distro"), but directly, more or less how a "normal" Python package is built.
In particular, you can install it into an existing Python venv, rather than being forced to use Sage's venv.

[dimpase]

sage-num-threads.py

in the process of despaghettification of the code, some loose pasta chunks were left loose :-)

OK, if this is what needs to be done, installing said file to a proper place, this is easy to fix

[dimpase]

Please review #41025