pre-commit.ci has been disabled for a while.
This PR ensures that all the files has been validated/formatted by pre-commit, to avoid spurious diffs in subsequent PRs.
During the process, ruff broke the code because it removed an "unused" import which was actually used.
A linter which breaks my code is a linter which I cannot trust, so I just removed it. I re-enabled isort instead.
* blacked
* More robust code for display, with tests
Display now includes more robust controls when checking
the input target parameters, with appropriate exception raised
(i.e. ValueError or TypeError) whether target is either
an empty string, or a not-string, not-None type, respectively.
The TypeError aligns with other similar behaviour with other Pyton
functions (e.g. str.split with integer separator).
Also, now display raises a ValueError whether the target element
is not found or not existing.
All changes are supported by tests.
* traceback lines in check_py_error & removed clones
check_py_error function now automatically includes
check of the traceback in console errors.
This way tests in basic and display have been refactored,
and lots of duplicated code removed.
* removed useless console check lines.
If check_py_errors function is running, those
console log lines are useless to check.
This re-enables CI tests on every PR 🎉.
This uses make test-integration, which runs tests sequentially.
In theory, we also have test-integration-parallel but it seems to be very flaky: many tests randomly timeout. I didn't spend too much time investigating this, it's probably worth its own investigation in a separate PR, but for now it's important to re-enable CI tests, even if they are a bit slower.
This PR re-enables tests on `worker`s. Highlights:
* by default, each test is run twice: the main thread version uses `<script type="py">`, the worker version automatically turn the tags into `<script type="py" worker>`
* you can tweak the settings per-class by using the `@with_execution_thread` decorator. In particular, `@with_execution_thread(None)` is for those tests which don't care about it (e.g., `test_py_config.py`)
* inside each class, there might be some test which should be run only in the main thread (because it doesn't make sense to test it in a worker). For those, I introduced the `@only_main` decorator
* we might introduce `@only_worker` in the future, if needed
* `@skip_worker` is for those tests which currently pass on main but not on workers. These are meant to be temporary, and eventually they should all be fixed
During the process, I tweaked/improved/fixed/deleted some of the existing tests. Some of them were at risk of being flaky and I made them more robust, others depended on some very precise implementation detail, and I made them more generic (for example, `test_image_renders_correctly` relied on pillow to render an image with a very specific string of bytes, and it broke due to the recent upgrade to pyodide 0.24.1)
I also renamed all the skip messages to start with `NEXT`, so that they are easier to grep.
Clean up a bit the testsuite and integration tests.
Highlights:
- Some of the @skipped tests just worked -- I unskipped them
- some worked after some small tweak to adapt to the new pyscript next
- some are still skipped, but I tweaked the skip message to be more precise and descriptive
- Moreover, I killed/removed the ones which no longer make sense in the context of pyscript next; in particular, I removed all the ones which tested Element (which is now gone) and the one which tested py-config features which are no longer needed (e.g., multiple interpreters).
The testsuite passes locally.
