[Python-Dev] New work-in-progress bisection tool for the Python test suite (in regrtest)

[Nick Coghlan]

On 4 July 2017 at 22:10, Victor Stinner <victor.stinner at gmail.com> wrote:
> 2017-07-04 13:22 GMT+02:00 Nick Coghlan <ncoghlan at gmail.com>:
>> That means if test.bisect is shadowing the top level bisect module
>> when backported, it suggests that the test.regrtest directory is
>> ending up on sys.path for the affected test run (e.g. because the
>> tests were run as "python Lib/test/regrtest.py" rather than via the -m
>> switch).
>
> I don't think that Lib/test/ is in sys.path. It's more subtle than
> test. When you run "./python -m test test_bisect",
> Lib/test/regrtest.py imports "test.test_bisect", and so test_bisect is
> imported with __package__=['test'].
>
> With test_bisect.__package__=['test'], "import bisect" in
> Lib/test/test_bisect.py imports Lib/test/bisect.py.

Right, for test_bisect specifically, the implicit relative import
problem applies, and "from __future__ import absolute_import" is the
relevant fix.

That concern just doesn't apply to the *stdlib* modules doing a normal
top-level "import bisect".

> The question is more when Lib/multiprocessing/heap.py got
> Lib/test/bisect.py instead of Lib/bisect.py. I didn't dig into this
> issue. The Python 2 import machinery blows my mind :-)

*This* is the case that I think is due to "Lib/test" being on sys.path
when the tests are run:

```
$ ./python -i Lib/test/regrtest.py --help
[snip output]
>>> import sys
>>> sys.path[0]
'/home/ncoghlan/devel/py27/Lib/test'
```

Using test_urllib2 as the example:

```
$ touch Lib/test/bisect.py
$ ./python -m test.regrtest test_urllib2
Run tests sequentially
0:00:00 [1/1] test_urllib2
1 test OK.

Total duration: 800 ms
Tests result: SUCCESS
$ ./python Lib/test/regrtest.py test_urllib2
Run tests sequentially
0:00:00 [1/1] test_urllib2
Warning -- os.environ was modified by test_urllib2
[snip output]
test test_urllib2 failed -- multiple errors occurred; run in verbose
mode for details
1 test failed:
    test_urllib2

Total duration: 107 ms
Tests result: FAILURE
```

Cheers,
Nick.

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia