Testing padic_base_leaves.py takes a VERY long time

[jdemeyer]

On a particular server, doing the long doctests of src/sage/rings/padics/padic_base_leaves.py:

• Sage 8.0.beta9: [142 tests, 202.82 s]

• Sage 8.0.beta10: [191 tests, 626.31 s]

The test time in Sage 8.0.beta10 is simply too much (but the number of tests has increased since 8.0.beta9, so this is probably not a regression). This is a problem because it leads to timeouts when running doctests.

[chapoton]

For me, on a Dell Optiplex 9020.

lINE 242: Test ran for 65.62 s
LINE 243 : Test ran for 105.39 s

[191 tests, 304.13 s]

With many slow TestSuite(R).run()

EDIT forgot to say, on 8.0.b11

[jdemeyer]

The main thing to check is whether this is a regression or not. I plan to test some older versions of Sage.

[tscrim]

Could there be something in there which calls the optional spkg normaliz? It does not like to be run in parallel because of its own parallelization. I ran across this and have a post about it on ​sage-devel.

[roed]

I understand where some of these changes are coming from. I'll take a look next week (no internet over the weekend).

[kcrisman]

Yikes, 343 seconds on my Mac laptop too (with nothing else going on).

[jdemeyer]

I investigated the first regression and it is due to MPIR-3.0.0: #23327

I will change this ticket to be only about the second issue: the large absolute time taken by testing padic_base_leaves.py

[git]

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

​8ce532e

Make long tests in padic_base_leaves.py shorter, add documentation for timings options to unit test code

[roed]

These changes drop the long tests from 256s to 37s.

[jdemeyer]

How are the changes in src/sage/misc/sage_unittest.py related to this ticket? If they are not, please create a new ticket for that change.

[jdemeyer]

Some more tests should be marked # long time, in particular on lines 244, 246, 336, 431.

[jdemeyer]

Why do you need separate runs of TestSuite here?

sage: TestSuite(R).run()
sage: TestSuite(R).run(elements = [R.random_element() for i in range(2^10)], max_runs = 2^12, skip='_test_metric') # long time
sage: R._test_metric(elements = [R.random_element() for i in range(2^3)]) # long time

Why is TestSuite(R).run() not sufficient?

(This is just a question, there might be a good reason)

[kcrisman]

Doctesting 1 file.
sage -t --long src/sage/rings/padics/padic_base_leaves.py
    [209 tests, 49.45 s]
----------------------------------------------------------------------
All tests passed!
----------------------------------------------------------------------
Total time for all tests: 49.6 seconds
    cpu time: 44.5 seconds
    cumulative wall time: 49.5 seconds

[git]

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

​1f0a79c	Modifying long tests in padic_base_leaves.py
​748b68c	Undo changes to sage_unittest.py

[roed]

Replying to jdemeyer:

How are the changes in src/sage/misc/sage_unittest.py related to this ticket? If they are not, please create a new ticket for that change.

They got accidentally included. I've removed them.

[roed]

Replying to jdemeyer:

Why do you need separate runs of TestSuite here?

sage: TestSuite(R).run()
sage: TestSuite(R).run(elements = [R.random_element() for i in range(2^10)], max_runs = 2^12, skip='_test_metric') # long time
sage: R._test_metric(elements = [R.random_element() for i in range(2^3)]) # long time

Why is TestSuite(R).run() not sufficient?

(This is just a question, there might be a good reason)

Just want tests with random elements, not just the ones returned by some_elements, which may not reveal problems.

[roed]

I've added a few # long time tags.

[jdemeyer]

Blocker or not, it would be good to merge this quickly.

[roed]

Replying to jdemeyer:

How are the changes in src/sage/misc/sage_unittest.py related to this ticket? If they are not, please create a new ticket for that change.

This is now #23335. I'd be interested in feedback on the interface there.