On git-master (32ebd8508d4807a7c85d2ed8e9c3b44ecd6de591) of CPython, 3 tests of test_builtin fails:

======================================================================
ERROR: test_envar_unimportable (test.test_builtin.TestBreakpoint) (envar='.')
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/yen/Projects/cpython/Lib/test/test_builtin.py", line 1618, in test_envar_unimportable
    breakpoint()
ValueError: Empty module name

======================================================================
ERROR: test_envar_unimportable (test.test_builtin.TestBreakpoint) (envar='.foo')
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/yen/Projects/cpython/Lib/test/test_builtin.py", line 1618, in test_envar_unimportable
    breakpoint()
ValueError: Empty module name

======================================================================
ERROR: test_envar_unimportable (test.test_builtin.TestBreakpoint) (envar='.int')
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/yen/Projects/cpython/Lib/test/test_builtin.py", line 1618, in test_envar_unimportable
    breakpoint()
ValueError: Empty module name

----------------------------------------------------------------------

If I revert 6fe9c446f8302553952f63fc6d96be4dfa48ceba, tests pass. This commit is from issue34756, so I add the author of that patch to the nosy list.

Environment: Arch Linux x86_64

Steps to reproduce:
$ ./configure
$ make
$ ./python -m test -v test_builtin

Wow that's super fast! I can confirm the patch fixes the issue on my machine. Thanks!

I don't understand why buildbots didn't scream and why the test didn't fail on Travis CI nor AppVeyor on https://github.com/python/cpython/pull/9457

Found a test_builtin failure in the job for the relevant commit on Travis CI - https://travis-ci.org/python/cpython/jobs/479642964#L1491. I guess it's not noticed as it's listed in allow_failures in .travis.yml. However, I'm not sure if it's the same issue or not as there are no failure details.

> Found a test_builtin failure in the job for the relevant commit on Travis CI - https://travis-ci.org/python/cpython/jobs/479642964#L1491.

This job is for code coverage. I confirm that failures are allowed. This job runs the full test suite in sequence, whereas the main Travis CI job runs tests in parallel... I tested running tests in sequence and in parallel: test_builtin fail in both cases. It's a mystery. Anyway, it's going to be fixed soon ;-)

New changeset 3607ef43c4a1a24d44f39ff54a77fc0af5bfa09a by Serhiy Storchaka in branch 'master':
bpo-35742: Fix test_envar_unimportable in test_builtin. (GH-11561)
https://github.com/python/cpython/commit/3607ef43c4a1a24d44f39ff54a77fc0af5bfa09a

New changeset 97d6a56d9d169f35cf2a24d62bf15adfc42fc672 by Miss Islington (bot) in branch '3.7':
bpo-35742: Fix test_envar_unimportable in test_builtin. (GH-11561)
https://github.com/python/cpython/commit/97d6a56d9d169f35cf2a24d62bf15adfc42fc672

I was surprised too. Maybe it is because that was very old PR.

> I was surprised too. Maybe it is because that was very old PR.

Yeah, that's the most likely explanation. Maybe the CI ran tests before the final merge, or something like that? It doesn't explain why buildbots didn't complain, but it doesn't matter if such hiccup is very rare ;-)

Maybe CI updates only files touched by the PR and the failed test was added after the initial creation of the PR? Or there is something wrong with timestamps, so outdated pyc files were used for tests?