Terry J. Reedy added the comment:
Currently, all requires() tests pass when the file they occur in is run as
'__main__'. This is especially needed when the file ends with the now standard
boilerplate.
if __name__ == '__main__':
...
unittest.main(...)
as there is currently no way to set resources within the unittest.main call.
The problem is that this permissiveness does not apply to subsidiary files
discovered from and run by a main file, even though it should. The current
workaround is to explicitly set use_resources for the benefit of subsidiary
files, as in test_idle.py.
As I see it, the main point of this patch, somewhat obscured by the title, is
to extend the current resource permissiveness from tests *in* main files run as
main to tests in other files discovered and run from a main file. It also
extends the permisiveness to any test not run by regrtest (ie, by unittest).
The key change is
- if sys._getframe(1).f_globals.get("__name__") == "__main__":
+ if not regrtest_run:
'regrtest_run == True' is currently spelled 'use_resources is not None', so the
latter could be used instead to replace the frame-check without otherwise
adding new code.
Extending the permissiveness from main files to subsidiary files strikes me as
a no-brainer. Splitting a (potentially) large file into a master file and a
package of subsidiary files should not affect which tests are run.
More interesting is extending the permisiveness to tests run under unittest
with "python -m unittest target". Target can be a master file, a test file, or
a test case or test methods. Subfile targets can only be run with unittest, not
regrtest, and there is no way to enable resources for such targets.
python -m unittest idlelib.idle_test.test_xy.TextText # runs
python -m unittest idlelib.idle_test.test_xy.GuiText # skips
So the patch enables something that is currently not possible.
Serhiy is concerned about the possible booby-trap for users that run slow
resource intensive tests. Some thoughts:
* Running tests as main is mainly done interactively, and usually by developers
at that. Humans can stop a test and ignore errors better than buildbots.
* There are multiple ways to run a file from the command line. The test chapter
of the manual can document that
python -m test.test_xyz
is more or less equivalent to
python -m test -uall test_zyz
with -v possibly tossed in. Anyone who does not want that can still run under
regrtest by using the currently documented
python -m test test_xyz
* Anyone running a test file loaded in an Idle window very likely wants to run
all tests possible.
* Documenting that running under unittest enables all resources is trickier as
long as resources are cpython and regrtest specific. I think I would mention
this in the test chapter, where resources are discussed, rather than the
unittest chapter.
*If -u is added to unittest (and 'use=' to .main), a default of all would be
the right thing for subfile targets, even if not for file and multi-file
targets.
---
----------
nosy: +terry.reedy
stage: -> patch review
_______________________________________
Python tracker <rep...@bugs.python.org>
<http://bugs.python.org/issue18492>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
