New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Py3.11 sync branch with main #8814
Conversation
This removes a spurious import in the middle of a class scope presumably left over from refactoring and also removed the PYVERSION import, since that is no longer being used in the module.
`PYVERSION` is re-exported in `utils.py` for the rest of the Numba code-base. Removing this import (re-export) will break all of Numba. So this change has been reverted.
* main: Improve error message Remove glue from test names/comments Remove `overload_glue` module Remove `glue_lowering` aka `overload_glue` Add unittest for `always_run` Remove tests from `always_run` Changes how tests are split between test instances
This aims to make it so that external compilers are only probed when absolutely necessary and that the result of the probe is cached. This should speed up initialisation of the testing framework and consequently speed up tests running in subprocesses.
Bump minimum supported Python version to 3.8
As title.
…migration Updates for Python 3.8 baseline/Python 3.11 migration
…name Preserve module name of jitted class
This patch makes it so that the NRT is only compiled when compilation starts as the result of the user electing to trigger compilation. A new test is added to ensure that decorating a function with `njit` doesn't trigger NRT compilation.
This to make it easy to see if the CUDA tests are running. Patch will be reverted once its purpose is served.
… CI." This reverts commit 81cd582.
This patch: * Makes an `always_run` test decorator to use on tests that should run in all test shards, it's based on the existing test `tag` mechanism. * Updates the test runner to recognise the `always_run` tag when sharding so as to ensure that tests tagged with this tag will always be included in a shard. * Updates the unit tests to ensure fundamental properties with regards to sharding. This includes shard uniqueness, the presence of the `always_run` tests in shards and also that the sum of the test shards and `always_run` tests is the same as the full test listing. * Fixes up a few references slice->shard. Fixes numba#8657
[unicode-PERF]: use optmized BM algorithm to replace the brute-force finder
…on_windows Enable the CUDA simulator tests on Windows builds in Azure CI.
Failing tests all seem to relate to missing llvmlite build for py3.11. I'm doing the build out now |
For reviewers: actual changes start at 2c7926d, which is the merge commit from main. The subsequent commits are for fixups. |
@@ -71,7 +71,7 @@ elif [[ $(uname) == Darwin ]]; then | |||
fi | |||
|
|||
# Install latest correct build | |||
$CONDA_INSTALL conda-forge::llvmdev=11 | |||
$CONDA_INSTALL -c conda-forge numba/label/pynext_ci::llvmdev=14 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Don't have the llvmlite buildout for py3.11 yet.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the patch. I've reviewed up to the point of the merge commit from main
into py3.11
, have not checked the "fixes" as yet.
I have addressed everything so far. The CI buildscripts changes are intentionally left is to match what's on the py3.11 branch so the test results can be compared |
See #8822 that merge |
superseded by #8831 |
No description provided.