Skip to content

fix(cuda.core): add 3 missing CUresult entries to frozen fallback table in cuda.core v1.2.1 - #3066

Closed
lijinf2 wants to merge 1 commit into
NVIDIA:release/cuda-core-v1.2.1from
lijinf2:fix/6843633-frozen-curesult-table
Closed

fix(cuda.core): add 3 missing CUresult entries to frozen fallback table in cuda.core v1.2.1#3066
lijinf2 wants to merge 1 commit into
NVIDIA:release/cuda-core-v1.2.1from
lijinf2:fix/6843633-frozen-curesult-table

Conversation

@lijinf2

@lijinf2 lijinf2 commented Oct 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

cuda.core 1.2.1's frozen CUresult fallback table is missing 3 members introduced in CTK 13.4:

  • CUDA_ERROR_MULTICAST_RESOURCE_FULL (47)
  • CUDA_ERROR_INSUFFICIENT_LOADER_VERSION (227)
  • CUDA_ERROR_FABRIC_NOT_READY (919)

(name/value mapping per cybind/assets/headers/driver/13.4.1/cuda.h: line 2818, line 3000, line 3429)

Any env with cuda-bindings>=13.4.1 fails test_frozen_driver_table_covers_all_curesult_members. This adds the 3 entries and text from the CUresult docstrings.

Testing

Repro'd with cuda-core[cu13]==1.2.1 + cuda-bindings==13.4.2 in a clean venv: the test fails before this patch and passes after.

@lijinf2
fix(cuda.core): add 3 missing CUresult entries to frozen fallback tab…
74376e7 
…le (#6843633)

cuda-bindings 13.4.2 exposes 106 CUresult members while cuda.core 1.2.1's
frozen fallback table (capped at CTK 13.1.1/13.3) only covers 103, so
test_frozen_driver_table_covers_all_curesult_members fails on every
Python version tested.

Add the 3 members introduced by cuda-bindings 13.4.1+:
- CUDA_ERROR_MULTICAST_RESOURCE_FULL (47)
- CUDA_ERROR_INSUFFICIENT_LOADER_VERSION (227)
- CUDA_ERROR_FABRIC_NOT_READY (919)

Text is copied from the cuda.bindings.driver.CUresult member docstrings
(cuda-bindings==13.4.2), cleaned the same way
DocstringBackedExplanations.get() would clean them.

Bug 6843633
@copy-pr-bot

copy-pr-bot Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@github-actions github-actions Bot added the cuda.core Everything related to the cuda.core module label Oct 9, 2026
@lijinf2 lijinf2 self-assigned this Oct 9, 2026
@lijinf2 lijinf2 changed the title fix(cuda.core): add 3 missing CUresult entries to frozen fallback table fix(cuda.core): add 3 missing CUresult entries to frozen fallback table in cuda.core v1.2.1 Oct 9, 2026
@lijinf2 lijinf2 added this to the cuda.core 1.2.1 milestone Oct 9, 2026
@lijinf2 lijinf2 added the bug Something isn't working label Oct 9, 2026
@lijinf2

lijinf2 commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ok to test 74376e7

@lijinf2
lijinf2 enabled auto-merge (squash) October 9, 2026 17:51
@github-actions

github-actions Bot commented Oct 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor
Doc Preview CI
Preview removed because the pull request was closed or merged.

rwgk
rwgk requested changes Oct 9, 2026
View reviewed changes

@rwgk rwgk left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

cuda_core/cuda/core/_utils/driver_cu_result_explanations_frozen.py should never be touched. Give me a moment to dig up a recent relevant PR.

@rwgk

rwgk commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

I found the PR I had in mind:

I'm still working on getting my head around the nvbug situation.

@rwgk

rwgk commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

GPT-6.1-Sol ultra analysis:

Here’s how we got there:

  1. September 2 — PR 2383 refreshed the frozen table, added this coverage test, and instructed release maintainers to keep updating the table. The 1.2.0 release line branched immediately afterward.
  2. September 9 — your PR 2792 reversed all three changes: restored the frozen snapshot, deleted this exact test, and removed the refresh instructions. Your comment explicitly explains the intended policy. Comment.
  3. September 23 — cuda-core 1.2.1 was prepared from that older release line without PR 2792. Consequently, PR 3066’s target branch still contains the test and instructions telling agents to refresh frozen tables. Stale instructions.
  4. September 30 — PR 2968 already recognized this failure and deselected the test for released cuda-core ≤1.2.1 in nightly CI, explicitly citing PR 2792. Existing workaround.
  5. SWQA runs the 1.2.1 tag’s tests with bindings 13.4.2. The bug reports precisely that assertion: 106 enum members versus 103 frozen entries. Rob’s bug comment recommends adding the missing entries, which PR 3066 implements.

So I can’t establish whether Jinfeng’s agent read your comment. The branch it worked on never received that comment or its associated correction, and both the branch’s checklist and the bug guidance pointed toward updating the table.

There is also a CI coverage gap: the release branch builds against CTK 13.4.2 but tests older toolkit minors, which select corresponding published bindings. Building 13.4.2 therefore does not ensure cuda-core was tested with bindings 13.4.2. Dependency selection.

I recommend replacing PR 3066’s table expansion with a backport of PR 2792’s correction. For SWQA testing the existing immutable 1.2.1 tag, use the same targeted deselection already present in nightly CI.

@lijinf2

lijinf2 commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

Fixed in backport PR #3068

@lijinf2 lijinf2 closed this Oct 9, 2026
auto-merge was automatically disabled October 9, 2026 20:38

Pull request was closed

@lijinf2
lijinf2 deleted the fix/6843633-frozen-curesult-table branch October 9, 2026 23:10
github-actions Bot pushed a commit that referenced this pull request Oct 10, 2026
cuda-python-bot
Clean up PR preview folders for 7 closed/merged PRs
a485805 
Removed preview folders for the following PRs:
- PR #3024
- PR #3059
- PR #3062
- PR #3063
- PR #3065
- PR #3066
- PR #3067
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@rwgk rwgk rwgk requested changes

Assignees

@lijinf2 lijinf2

Labels

bug Something isn't working cuda.core Everything related to the cuda.core module

Projects

None yet

Milestone

cuda.core 1.2.1

Development

Successfully merging this pull request may close these issues.

2 participants

@lijinf2 @rwgk