WIP: Partial fix for Python 3.12 #158
Conversation
Python 3.12 doesn't call `startTest()` for skipped tests, which caused AttributeErrors for things initialized in our `TestResult.startTest()` (and `OutputFormatter.start_test()`). This is not a complete fix: there are still test failures because they expected "ran 1 tests, 1 skipped" and now get "ran 0 tests, 0 skipped". I don't know if we want to keep the old method of counting; it kind of makes sense to me not to count skipped tests as running ones. I'm not sure how to fix that without adding sys.version_info >= (3, 12) checks everywhere.
| @@ -52,7 +52,7 @@ class OutputFormatter: | |||
|
|
|||
| def __init__(self, options): | |||
| self.options = options | |||
| self.last_width = 0 | |||
| self.test_width = self.last_width = 0 | |||
| self.compute_max_width() | |||
There was a problem hiding this comment.
TBH I don't know if it's enough to initialize test_width here, or if it should also be reset to 0 in every test_success()/test_error()/test_failure().
Does it really no longer count skipped tests at all (or only skipped tests which started before they decided they want the get skipped)?
I assume that failing doctests are the main problem. In this case, a normalizer might be used as a central point to handle the version differences. |
|
There's also a risk that by skipping (Which is probably still better than the status quo, where the test runner crashes the first time it encounters a skipped test.) |
|
Marius Gedminas wrote at 2023-12-19 03:23 -0800:
There's also a risk that by skipping `startTest()` there's a combination of `-v`/`-p` options and a sequence of test runs/failures/skips that could mess up the output because I got the `test_width`/`last_width` logic wrong in this PR.
I got an alternative idea:
apparently, the (original) error happened in our `stopTest`.
This means, we already have a way to use our own `stopTest`. Likely,
we can have our own `startTest` in a similar way.
In this case, we could let our `startTest` set a marker "startTest run"
on the test instance and `stopTest` look for it, if not there, `stopTest`
could call `startTest` before its real work.
|
|
#159 is a simpler and more complete fix for this, closing. |
Python 3.12 doesn't call
startTest()for skipped tests, which caused AttributeErrors for things initialized in ourTestResult.startTest()(andOutputFormatter.start_test()).This is not a complete fix: there are still test failures because they expected "ran 1 tests, 1 skipped" and now get "ran 0 tests, 0 skipped". I don't know if we want to keep the old method of counting; it kind of makes sense to me not to count skipped tests as running ones.
I'm not sure how to fix that without adding sys.version_info >= (3, 12) checks everywhere.
See #157 and python/cpython#106584.