Learning center

Does an outdated review comment still apply to your branch?

A comment made on an earlier commit is a claim about that commit. Before you fix or dismiss it, find the commit it reviewed, test the claim against your current head, and say in the thread which commit changed the answer.

What "Outdated" tells you#

A review comment can become outdated when a later commit changes the lines it was attached to. GitHub's REST documentation says a comment made on an older commit may become outdated "if a subsequent commit modifies the line" it targets. Create a review comment The Conversations menu on the Files changed tab lists which conversations "are unresolved, resolved, and outdated." Commenting on a pull request

The label is about lines, not about the claim. It misleads in both directions:

  • A comment can be outdated and still true, because you reformatted the line, or moved the code, without changing what it does.
  • A comment can stay current and be false, because the push that settled it changed a different file, or arrived through a merge of the base branch.

So the label tells you where to look. It does not answer the question.

A claim that was true at the commit it reviewed#

In November 2026, a Home Assistant pull request changed the FiveM integration to raise UpdateFailedUnreachable instead of UpdateFailed when the server is offline. Copilot reviewed the first commit and reported that the coordinator imports UpdateFailedUnreachable, "which the helper module neither defines nor exports, causing an ImportError that prevents configuration and setup." Copilot's review One reviewer requested changes: "What copilot says, plus prek is failing". Review

The author pushed two formatting commits and merged the dev branch into the pull request. The next day a second maintainer requested changes again: "Check Copilot feedback, which is still not resolved. :)" Review The author answered:

The Copilot review is wrong. UpdateFailedUnreachable is defined in homeassistant/helpers/update_coordinator.py on this branch (added upstream in #188683, and this branch includes it through the merge from dev). The import resolves, so FiveM loads normally. Resolving.

Author's reply

Both statements can be checked against the commits. At the reviewed commit, 377ae4d, the helper file has no UpdateFailedUnreachable at all. At the merge commit 4498aae, it defines class UpdateFailedUnreachable(UpdateFailed):. Helper at 377ae4d, helper at 4498aae The class arrived with pull request #188683, which merged into dev on 13 November, a day before this pull request opened from an older base.

So Copilot was right about the commit it reviewed, where the import would have failed, and the author was right about the head. "The Copilot review is wrong" is the one part of the reply that does not hold. The precise statement is that the claim stopped applying when the base was merged in. The rest of the reply is what made the thread checkable: the file, the upstream pull request and the merge that brought the class in. The second maintainer approved, and the pull request merged on 16 November 2026.

Notice which push settled it. The fix did not touch coordinator.py, the file the comment was on. It changed a helper module through a merge from the base, which is the case where the outdated label is least likely to help.

A claim that a later push fixed#

In May 2026, Gemini Code Assist reviewed a vLLM pull request and said that computing num_redundant_experts by iterating through layers breaks under pipeline parallelism: on ranks that hold no MoE layers, self.num_redundant_experts would wrongly stay 0. Its review summary called the calculation "unsafe for Pipeline Parallelism." Review comment The comment was attached to a loop that set the value from the first MoE layer found on the rank.

The author force-pushed twice after the review, and the merged file answers the comment. The model now reads the value from configuration, which is the same on every rank:

self.num_redundant_experts = (
    vllm_config.parallel_config.eplb_config.num_redundant_experts
)

Merged AXK1.py The claim no longer applies to the head, and it deserved the change it got. What the thread lacks is one line saying so. A later reader, or a reviewer trial counting which comments changed code, has to compare commits to learn that the comment was accepted.

Recheck the claim at your head#

  1. Find the commit the comment reviewed. GitHub shows it on the review, and the API returns it as the comment's original_commit_id.
  2. Restate the claim as a condition you can test: an import that fails, a value that is wrong on some input, a caller that breaks.
  3. Test that condition at your current head, including files the comment was not attached to. Read the code, run the test, or import the module.
  4. Reply with the result and the commit that changed it, then resolve.
What you find at the headDecisionReply
The condition still holds, outdated or notChange the codeThe fixing commit
Your commit fixed itNo change"Fixed in" and the commit
A merge of the base fixed itNo changeThe base commit or pull request that fixed it
It never held, even at the reviewed commitDismissThe evidence, as in a dismissal reply

Keep the second and third rows apart from the fourth. A comment that was right about an earlier commit was a correct finding, even if the head no longer needs a change.

One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. Eligibility and fair-use terms.

Start free with GitHub

Related reading