Re-review only the new commits
Re-reading an entire PR after every push wastes both reviewers and buries the actual change. Diff since your last review, then verify your open findings against the runtime diff.
Procedure
- Find the commit you last reviewed. Note it when you first review, or read it from your review's SHA:
gh api repos/:owner/:repo/pulls/482/reviews --jq '.[].commit_id'. - Diff from that commit to the new head:
git diff <last_reviewed_sha>..HEADaftergit fetch origin pull/482/head. - Confirm the author addressed each prior finding by inspecting the intervening change, not the comment thread alone — a resolved thread may hide code that was reverted.
- Watch for collateral edits:
git range-diff origin/main..<old_head> origin/main..HEADshows whether a force-push or rebase silently altered commits you already approved. - If history was rewritten, re-read the rewritten hunks — a rebase can drop or reorder a fix. Do not trust a clean thread state.
- Re-approve only for the new range and note the new head SHA in your approval.
- If the diff since your review exceeds the size at which you can still verify it, request a fresh split rather than skimming.
Pitfalls
- Trusting the "resolved conversation" checkmark without looking at the code change that resolved it.
- A force-push that squashed the fix into an unrelated commit, so a line-by-line diff of the last commit misses it.
- Re-approving the whole PR while only the last commit was actually looked at.
- Reviewing a stale local branch because you did not fetch the updated pull ref.
Verification
git fetch origin pull/482/head:pr-482
git range-diff origin/main..<last_reviewed_sha> origin/main..pr-482
git diff --stat <last_reviewed_sha>..pr-482
Report the new head SHA, which earlier findings are confirmed fixed, and any hunk you had not seen. Approval covers only the range you diffed; state that range.
Worked example
You reviewed a1b2c3d and left two blocking: items. The author force-pushes; git range-diff origin/main..a1b2c3d origin/main..pr-482 shows the first fix wrapped into an unrelated commit and the second reverted during the rebase. Diffing a1b2c3d..pr-482 reveals the second finding is unresolved despite its thread being marked resolved — re-request the change rather than re-approving.
A clean re-review reads the intervening diff in under a minute; if it takes longer, the PR has grown past a fix and needs a fresh split rather than another pass.