Hi @mihaimaruseac, thank you for your previous feedback in #96866.

I've opened this new PR with the same code changes, but with a cleaned-up commit history (now squashed into a single commit to conform with TensorFlow's standards).

Appreciate it if you could take a look when you have time. Thanks again!

@mihaimaruseac Thanks for the feedback!
I've cleaned up the commit history using git rebase -i, and the PR now contains a single, clean commit as expected.
The previous version inadvertently included some commits from master, but that has been corrected.
Appreciate your patience and guidance!

Thanks again for your earlier feedback regarding the commit history.
In the previous PR, I had mistakenly merged master into my feature branch, which introduced unrelated commits. I now understand how that affects the internal import process.
For this PR, I’ve switched to a clean branch and used git rebase -i upstream/master to ensure a linear, single-commit history.
I appreciate your guidance — please let me know if anything else needs adjustment.

Hi @mihaimaruseac ,

Thank you again for reviewing and merging PR #96429 via the Copybara‑Service—it’s an honor to have made my first contribution to TensorFlow!

I’ve just squashed my two commits and force‑pushed on this PR. Could you please take a look and approve again when convenient? Thank you!

Also, I noticed that in PR #96429 my three commits were merged into master rather than being squashed. I had assumed Copybara‑Service would handle that. To align better with TensorFlow’s workflow, would you recommend I generally keep one commit per PR going forward?

I appreciate your guidance and support!

Best regards,
Qingwei

Hi,

#96429 was closed by a merge commit. On master branch there is only the merge commit, your commits are on the merged branch.

It is recommended to not edit the PR after the approval, unless really needed (test failures, requested fixes). Remember that the PR is being imported into the internal system and each modification adds additional delays as the new change also needs to be imported and reviewed again. It might happen that the new change is being reviewed by other people, adding more delays as new people also need to get up to date with the history of the change, etc.

Please only edit PRs if requested by a reviewer. Don't merge master back into the PR, only rebase if GitHub shows a merge conflict.

Right now, because this PR was changed after approval, it needed a new approval here to be imported into the system. I missed that notification, so the PR stalled until today.

Hi @mihaimaruseac,

Thank you so much for the follow-up and re-approval! I appreciate your time and the clear guidance on the review and merge process. I've taken note of the recommendation to avoid modifying PRs post-approval unless necessary, and I’ll be sure to follow that going forward.

Looking forward to seeing this merged—thank you again!

Best regards,
Qingwei

Hi @mihaimaruseac ,

I noticed that the CI failure was caused by a SHA‑256 mismatch error while downloading the pygments==2.19.1 wheel:

Expected sha256 61c16… or 9ea15…, but got ab1688…

This indicates that the requirements_lock.txt file needs to be updated to reflect the new checksum. My plan is to re-run:

bazel run //:requirements.update

or the equivalent lockfile generation, and then commit the updated lock file to resolve the mismatch.

Does that approach look good to you? Thanks for your guidance!

Best regards,
Qingwei

I think the actual error to fix is

  /tensorflow/dtensor/cc/small_constant_optimization_test.cc:22 ClangTidy: no header providing "TF_DeleteStatus" is directly included
  /tensorflow/dtensor/cc/small_constant_optimization.cc:108 ClangTidy: no header providing "TF_TensorData" is directly included
  /tensorflow/dtensor/cc/small_constant_optimization_test.cc:13 ClangTidy: no header providing "TF_NewStatus" is directly included
  /tensorflow/dtensor/cc/small_constant_optimization_test.cc:15 ClangTidy: no header providing "std::optional" is directly included
  /tensorflow/dtensor/cc/small_constant_optimization_test.cc:20 ClangTidy: no header providing "TF_INVALID_ARGUMENT" is directly included

The dependency hash mismatch should be fixed by a concurrent change (already happening)

Hi @mihaimaruseac

Thanks for pointing out the missing headers flagged by Clang-Tidy. I’ll make the necessary changes and push an update within the next two days.

Also, I noticed that these Clang-Tidy issues weren’t visible in the build logs I checked (e.g., build.log or CI output). Could you let me know where I can view these warnings directly in the future?

Appreciate your time and help!

Best regards,
Qingwei

Actually, no need to change anything, it got fixed internally

Hi @mihaimaruseac ,

Thank you for the update! Glad to hear it’s been resolved internally. I’ll leave the PR as is then.

Let me know if anything else is needed on my side.

Best regards,
Qingwei

Thanks, Michael!