Uh oh!
There was an error while loading. Please reload this page.
pathfinder: add toolkit-info parser and display-driver release helpers - #2012
Closed
rwgk wants to merge 5 commits into
Closed
pathfinder: add toolkit-info parser and display-driver release helpers#2012rwgk wants to merge 5 commits into
rwgk wants to merge 5 commits into
Conversation
Introduce a small toolkit-info utility that reads the CUDA_VERSION macro from cuda.h so follow-up guard-rails changes can infer CTK major.minor from toolkit headers without depending on version.json. Made-with: Cursor
Centralize encoded CUDA version parsing and validation so toolkit and driver version helpers stay aligned and cuda.h parsing gets consistent string conversion and error reporting. Made-with: Cursor
Introduce ``DriverReleaseVersion`` and ``query_driver_release_version()`` in ``cuda.pathfinder._utils.driver_info`` so callers can read the display-driver release shown as ``Driver Version`` in ``nvidia-smi`` (for example ``595.58.03``, branch ``595``). The helper queries ``nvmlSystemGetDriverVersion()`` via ``ctypes`` against the NVML library loaded through pathfinder's existing dynamic-lib loader, and parses the result into a frozen dataclass exposing both the raw text and the branch number used by NVIDIA's published minor-version compatibility tables. Tests cover the dataclass parser, the cache-clear lifecycle, the ``QueryDriverReleaseVersionError`` wrapping, and a fake-NVML end-to-end path that asserts the shutdown call always runs. Co-authored-by: Cursor <cursoragent@cursor.com>
Remove ``test_query_driver_release_version_returns_parsed_dataclass`` and ``test_query_driver_cuda_version_returns_parsed_dataclass``: both shadow the dataclass parsers (``DriverReleaseVersion.from_text`` and ``DriverCudaVersion.from_encoded``), which already have direct coverage, and the wrapping/error paths are still exercised by the remaining mocked tests. Co-authored-by: Cursor <cursoragent@cursor.com>
…nt contract Collapse the duplicated ``nvmlShutdown()`` calls in ``_query_driver_release_version_text`` into a single ``try/finally`` so the cleanup always runs in one place. The asymmetric error-precedence rule is preserved via ``sys.exc_info()[1]``: when both the NVML body and shutdown fail, the body's error wins (Python keeps the shutdown error on ``__context__`` for debugging); when only shutdown fails, the shutdown error surfaces. Add comments above the matched ``nvmlInit_v2()`` / ``nvmlShutdown()`` pair noting that NVML's init/shutdown is reference-counted, so this balanced pair is safe even when the caller has already initialized NVML elsewhere in the process. Pre-empts a question raised in review on PR #2000. Add two focused tests filling out the cleanup matrix: - ``test_query_driver_release_version_text_raises_when_only_shutdown_fails`` asserts a non-zero shutdown status surfaces when the body succeeded. - ``test_query_driver_release_version_text_body_error_wins_when_both_fail`` locks in the body-error-wins precedence when both calls fail. Co-authored-by: Cursor <cursoragent@cursor.com>
Contributor
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. |
rwgk
commented
May 4, 2026
ContributorAuthor
/ok to test |
|
rwgk
commented
Aug 4, 2026
ContributorAuthor
Closing this PR (and PR #1977) so I can rebuild the work as a native stacked PR series using GitHub's newly announced stacked pull requests feature, which is currently rolling out in public preview: This indicates that the new feature is already available in this repo (the list is empty because there are no stacks yet): I’ll link the replacement PRs here once they're available. |
rwgk
commented
Aug 5, 2026
ContributorAuthor
The corresponding new PR is #2505 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Extracted from #1977 (xref #1038) so the underlying
cuda.pathfinder._utilsplumbing can be reviewed independently of the larger compatibility guard-rails work that consumes it. Five commits, scoped strictly tocuda_pathfinder/cuda/pathfinder/_utils/{driver_info,toolkit_info}.pyand their unit tests; nothing else is touched.Changes
toolkit-info parser. Add
cuda.pathfinder._utils.toolkit_info, a small utility that reads theCUDA_VERSIONmacro fromcuda.hso callers can infer CTK major.minor from toolkit headers without depending onversion.json.Shared encoded-CUDA-version decoding. Centralize encoded CUDA version parsing and validation so toolkit and driver version helpers stay aligned and
cuda.hparsing gets consistent string conversion and error reporting.Display-driver release metadata. Add
DriverReleaseVersionandquery_driver_release_version()incuda.pathfinder._utils.driver_info. The helper queriesnvmlSystemGetDriverVersion()viactypesagainst the NVML library loaded through pathfinder's existing dynamic-lib loader, and parses the result into a frozen dataclass exposing both the raw text (e.g."595.58.03") and the branch number (e.g.595) used by NVIDIA's published minor-version compatibility tables.try/finallyaround NVML init/shutdown. The matchednvmlInit_v2()/nvmlShutdown()pair runs under a singletry/finally, withsys.exc_info()[1]preserving the asymmetric error precedence: when both the body and shutdown fail, the body's error wins (Python keeps the shutdown error on__context__for debugging); when only shutdown fails, the shutdown error surfaces. Comments document that NVML's init/shutdown is reference-counted, so this balanced pair is safe even when the caller has already initialized NVML elsewhere in the process.Tests. Direct coverage of the dataclass parsers, the cache-clear lifecycle, the
QueryDriverReleaseVersionErrorwrapping, the fake-NVML end-to-end path, and the full success/failure cleanup matrix (body-ok+shutdown-fail, body-fail+shutdown-ok, body-fail+shutdown-fail).