Clearer error messages (cuda.core) #458
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
leofang
added
enhancement
…fo functions stricter
|
/ok to test |
This comment has been minimized.
This comment has been minimized.
|
/ok to test |
|
/ok to test |
|
/ok to test |
|
@leofang I pushed up what I have at the moment, mostly for backup, please don't look yet. (I need to do something else for a couple hours.) I still have 10 ACTNBL to go. |
…formative error handling.
|
Tracking progress: I've now resolved all Many changes are only tested interactively, e.g. by temporarily changing Half of the changes in the last commit (29c9fb8) are not exercised at all. I'll try to add a few minimal tests, so that there is at least a happy path. |
…ATIONS (to be moved to a separate PR).
|
@leofang I backed out the fancy driver and runtime errors (to be merged via another PR). The only thing I still want to do in this PR is to create the _utils subdirectory, as you suggested a few days ago. All other changes are ready for review. |
|
/ok to test |
…ad of AssertionError
…eeded for this PR
…erified manually), exercising the trivial NotImplementedError exceptions is not worth the trouble.
|
Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
|
/ok to test |
There was a problem hiding this comment.
We don't have to do this now, but since this is an internal module, we can just import everything in this module to under this namespace.
There was a problem hiding this comment.
mainly because the import is getting pretty long now (and the experimental module makes it worse :P )
|
…ATIONS originally created under PR NVIDIA#458 This code was removed from PR NVIDIA#458 with this commit: NVIDIA@d3df80d
* Bring back DRIVER_CU_RESULT_EXPLANATIONS, RUNTIME_CUDA_ERROR_T_EXPLANATIONS originally created under PR #458 This code was removed from PR #458 with this commit: d3df80d * Add instructions for regenerating the dictionaries. * Use DRIVER_CU_RESULT_EXPLANATIONS, RUNTIME_CUDA_ERROR_EXPLANATIONS in cuda_utils.handle_return() * Add test_driver_cu_result_explanations_health(), test_runtime_cuda_error_explanations_health() * Add test_check_driver_error(), test_check_runtime_error() * Copy https://github.com/rwgk/stuff/blob/6fc0881d2d9a37a0c6a5352ab5908b90ab0bd41e/cuda-python/reformat_cuda_enums_from_web_as_py.py
This is the main PR to address issue #198
Cleanup of error messages is considered complete.
Test coverage is expanded, but still not complete. Many changes were exercised only interactively, by temporarily changing
if condtoif not condand visually inspecting the raised exceptions.For completeness: These will be handled in separate PRs: 5dc2331, d3df80d