fix several tests related to constants #10013
Conversation
| def test_all_elements_without_parent(self) -> None: | ||
| node = astroid.extract_node("__all__ = []") | ||
| node.value.elts.append(astroid.Const("test")) | ||
| node.value.elts.append(astroid.Const("test", parent=None)) | ||
| root = node.root() | ||
| with self.assertNoMessages(): | ||
| self.checker.visit_module(root) |
There was a problem hiding this comment.
I don't think it's a good test. Why should adding a constant not result in messages? That's just asserting implementation quirks. We should test the opposite, we should test that synthetic constants result in the same messages.
IMO
jacobtylerwalls
force-pushed
the
fix-constant-tests
branch
from
bd66249 to
37df2ed
Compare
This comment has been minimized.
This comment has been minimized.
jacobtylerwalls
added
the
Skip news 🔇
|
Hi @temyurchenko 👋 We released an alpha of astroid 4 and updated your branch to use it. Would you mind taking a look at that last test failure? Many thanks. |
Hey, for sure! Hopefully, within the week |
|
The bug is actually caused by pylint-dev/astroid#2589, which changed the order of paths to look in, for module discovery. That PR on its own is not really unreasonable; it's just that the module path discovery is messy. |
temyurchenko
force-pushed
the
fix-constant-tests
branch
from
37df2ed to
64f6234
Compare
|
I've made a commit that is hopefully fixing the issue above. The commit message is describing the issue and the solution in detail. |
temyurchenko
force-pushed
the
fix-constant-tests
branch
3 times, most recently
from
f9bbc50 to
e1f3eea
Compare
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #10013 +/- ##
==========================================
+ Coverage 95.84% 95.85% +0.01%
==========================================
Files 175 175
Lines 19058 19056 -2
==========================================
+ Hits 18266 18267 +1
+ Misses 792 789 -3
|
I've added some tests that test regressions in the tests. I've also made the message to include the number of times a certain message is missing or extra. Without it, I had a very confusing experience, when I seemed to have the message and yet the test-suite reported that I don't. Turned out, I had to add four more of the same!
The problem was caused by pylint-dev/astroid#2589, changing the package discovery. Since tests/reporters/ didn't have `__init__.py`, the new package discovery would resolve module names as one level higher than desired (so, `unittest_reporting` instead of `reporters.unittest_reporting`). We're fixing this, by adding the appropriate `__init__.py`. We're also adding the `ModuleDescriptionDict` corresponding to the new `__init__.py` to the `expand_modules` tests. The changes in `expand_modules.py` are small optimizations mostly done in the process of understanding what is happening there.
temyurchenko
force-pushed
the
fix-constant-tests
branch
from
e1f3eea to
fb3b996
Compare
I've added some tests that test regressions in the tests.
I've also made the message to include the number of times a certain
message is missing or extra. Without it, I had a very confusing
experience, when I seemed to have the message and yet the
test-suite reported that I don't. Turned out, I had to add four
more of the same!
Type of Changes
Description
I'd be surprised if it passes the tests right now. This will need a new version of astroid to pass, specifically, pylint-dev/astroid#2602