[pathfinder] Add DynamicLibUnknownError, DynamicLibNotAvailableError and prepare for v1.4.0 release#1688
Conversation
Replace the short-lived ValueError validation path from 1.3.5 with DynamicLibNotFoundError subclasses so downstream exception handling remains compatible while exposing more precise failure reasons. Co-authored-by: Cursor <cursoragent@cursor.com>
|
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. |
Add the 1.4.0 pathfinder release notes entry and publish the 1.4.0 docs version link in nv-versions.json for the upcoming release. Co-authored-by: Cursor <cursoragent@cursor.com>
0ce38b2 to
9ce8030
Compare
Document the planned PyPI yank for cuda-pathfinder 1.3.5 in the release notes, including upgrade guidance to 1.4.0 and links to the tracking issue and PR. Co-authored-by: Cursor <cursoragent@cursor.com>
|
/ok to test |
|
See commit 99ad650 |
This comment has been minimized.
This comment has been minimized.
Update the 1.3.5 and 1.4.0 release notes to remove the YANK NOTICE and document that the ValueError behavior in 1.3.5 was a regression that is fixed in 1.4.0. Co-authored-by: Cursor <cursoragent@cursor.com>
|
/ok to test |
| exception regression in ``load_nvidia_dynamic_lib()`` (unsupported | ||
| library names incorrectly raising ``ValueError``). | ||
|
|
||
| This bug was corrected in ``cuda-pathfinder>=1.4.0``, restoring backward-compatible |
There was a problem hiding this comment.
This reads awkward to me. Bugs are corrected in specific releases, and by implication that means >=. Combining those two notions just makes this strange to read.
Not blocking though.
There was a problem hiding this comment.
I agree, this should just be 1.4.0, but it's not misleading, I don't want to run the CI again for this, and not use [doc-only] in the PR description because that could be confusing when we look at it later. I'll go ahead and merge this as-is, then make the release.
|
Close #1684
Also prepare for cuda-pathfinder v1.4.0 release.