fiddle with the number of threads automatically used for parallel testing

[dandrake]

At #6283, we changed the parallel testing framework so that it automatically uses all the cores/threads available, but perhaps this is not the best solution.

Dave says ([#6283 comment:9]) "I would have personally not allowed the default to exceed 8", so maybe we can incorporate his limit in a way that still lets ordinary multicore computers be well-used:

• NUM_THREADS defaults to 0, which is now interpreted in the sage-ptest script as min(cpu_count(), 8) -- so the default doesn't exceed 8, as Dave suggested.
• if NUM_THREADS is -1, it just uses cpu_count().

On sage-devel, I suggested that a solution that works really well for 99+% of people is a good one -- and since most "regular" machines on which Sage is doctested have 8 or fewer cores, this still works fine for them, and with the above suggestion, people won't bring sage.math or t2.math to their knees.

Thoughts?

CC: @sagetrac-mvngu @sagetrac-drkirkby @jhpalmieri

Component: doctest coverage

Author: Dan Drake

Reviewer: John Palmieri

Issue created by migration from https://trac.sagemath.org/ticket/7011

[dandrake]

comment:1

On second thought, scratch the -1 idea. We can just make one change to sage-ptest: line 267 could be

numthreads = min(8, multiprocessing.cpu_count())

Anyone who is desperate to saturate a machine with more than 8 cores can just specify it on the command line.

[dandrake]

add in default maximum of 8 threads

[dandrake]

Attachment: trac_7011.patch.gz

patch for $SAGE_ROOT/makefile

[dandrake]

Author: Dan Drake

[dandrake]

comment:2

Attachment: trac_7011-sage-root-makefile.patch.gz

I've uploaded patches for the sage_scripts repo, and for the root makefile. The second attachment is an ordinary unified diff.

[jhpalmieri]

comment:3

Looks good to me. Very helpful comments, too.

[jhpalmieri]

Reviewer: John Palmieri

[williamstein]

comment:5

merged into sage-4.1.2