[lawrencecchen]

Summary

• Follow-up to ghostty-next: a scroll region scroll breaks the soft wrap across its edges (VT fuzz finding E2) #29609, from its review note 1. In ghostty-next, Screen.breakWrapAbove read the row above before it checked the flags. So IL or DL at the screen's top row restored the last history page from compression on every call. Under load, a region scroll can run once per output line.
• The fix is fork PR terminal: breakWrapAbove reads a compressed history row only when the pair was joined ghostty-next#68. The function now decides on the resident row first (its wrap_continuation, or for IL/DL the top row's flag read before the rows move). It reads a row in a compressed history page only when the pair was joined.
• The pin moves to GhosttyNextKit xcframework-84a225487372c53851b186864c867e469e7596d2-apple-v6, on release/cmux-9ce9b7d0-e2. That release is 8444cc1eb (the ghostty-next: a scroll region scroll breaks the soft wrap across its edges (VT fuzz finding E2) #29609 pin) plus only fix cmux crashing issue #68. The gitlink, package-notices.json and vt-link-graph.json move with the pin.
• Review note 2: every commit ID in the E2 section of docs/ghostty-fork.md is now a release-branch ID, and the section says so.

Testing

• Red: fork run 41240560795. The new Zig test "insertLines and deleteLines at the top row leave a compressed history page compressed" fails without the fix.
• Green: fork PR fix cmux crashing issue #68 run 41240575661, and the release run 41241826620, which built and published this kit.
• The cmux Rust replay tests on this pin: see the fleet step in a comment below.

Observed / Not run

• Observed: the red and green fork runs above. SHA256SUMS of the release lists 66e883f7130352862fd3dd4b46cf0a6d22d50750bf633f3589831bf9b2f4f99f, which is the checksum in Package.swift.
• Not run: a timing benchmark. The test proves that the common case leaves the compressed page alone. It does not measure time.

Changelog

none

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Fixes a performance issue in ghostty-next where Screen.breakWrapAbove read a compressed history row before checking flags, so IL or DL at the screen's top row restored the last history page from compression on every call. The function now decides on the resident row first and reads a compressed row only when the pair was joined.

• Pins GhosttyNextKit xcframework-84a225487372c53851b186864c867e469e7596d2-apple-v6 (checksum 66e883f7...); the gitlink, package-notices.json, and vt-link-graph.json move with it.
• Adds the Zig test "insertLines and deleteLines at the top row leave a compressed history page compressed", which fails without the fix and passes with it.
• docs/ghostty-fork.md now states that every commit ID in the E2 section is a release-branch ID.

^{Written for commit 4bf50aa. Summary will update on new commits.}

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 4bf50aa: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr29775 -f ref=cp/ghostty-wrap-edge-history-cost-2 -f platform=macos -f backend_mode=local

Not verified: macOS jobs did not run: compile and app tests were skipped on 4bf50aa683; ci-status is green without them.

CI passes on 4bf50aa683 (run 41263139029 attempt 1).
macOS jobs did not run: compile and app tests were skipped: Fast static checks skipped (the macos job needs both to succeed). Re-run CI once they pass, before merging.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); signatures are its SIGNATURES table. A machine verdict is the runner's fault, not this PR's; yours means the failing file is one this PR changes, also red on main that main's latest full suite fails the same way, seen on other PRs that it failed on another pull request's run lately.

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4130052256501 (28aa638015)

[lawrencecchen]

Review (Claude Opus 5.5 review subagent): SHIP at eb7600b

Independent review subagent (read-only). It checked that the early return is sound for every state that this pin's own paths make, the moment top_joined is read in IL and DL, the other callers, that the Zig test reaches the compressed path and fails without the fix, the URL and checksum against SHA256SUMS, the gitlink, the notices, and every docs/ghostty-fork.md hash on release/cmux-9ce9b7d0-e2.

Non-blocking finding, kept as a follow-up: the new comment in Screen.breakWrapAbove says "every clear clears both", which is not true. Two older paths, clearRows (ED 1, ED 2, DECALN) and cursorScrollAbove with the cursor at row 0, can leave wrap=true on a history row whose active continuation row has no flag. This PR does not create that state. Before this PR, IL or DL at row 0 cleared it by chance; now that stops only when the history page is compressed. The comment fix, and a root fix for those two paths with their own red tests, go in the next fork change.

[lawrencecchen]

Rust replay tests on this pin: fleet step 0e1a97219c7ae128a85709bc (Actions run 41250330654): 6 vt_replay tests pass, among them vt_replay_keeps_a_stale_wrap_continuation_row_in_its_column.

[lawrencecchen]

"cmux-next swift test" classification (run 41250318057): not caused by this PR.

• This PR changes only the GhosttyNextKit pin. The fork diff is src/terminal/Screen.zig, Terminal.zig and a CI file. No Swift source changes.
• 36 UI and app suites failed, for example Acpmux*, Home*, SidebarEditModeTests and PinnedRowHoverCardTests. None of them is a terminal or VT suite. 18 of them also failed on ghostty-next: a scroll region scroll breaks the soft wrap across its edges (VT fuzz finding E2) #29609's run 41237585801. ContextMenuStandardTests, EmptyWorkspaceRepairTests and ArchiveSurfacesTests fail at the merge-base too (fleet step 16cce5c1123dddb707523ea2, base b0e10a0).
• I rerun the failed job once on this SHA, with no source change.

[lawrencecchen]

Classification of the compile reds on head 69ae06a (after cmux-bot merged feat): base red, not this change.

• Every compile lane fails in Packages/macOS/CmuxNext/Sources/CmuxNextDesign/Overlay/NativePopupHost.swift:235 and :239, "sending 'menu' risks causing data races". The affected lanes are "cmux-next fleet Release compile", "fast tier", "app compile" and "swift test".
• This PR does not touch that file. Its diff is only the GhosttyNextKit pin, the notices and docs/ghostty-fork.md.
• The file came in with the feat merge. Its fix has owners: base red: NativePopupHost's menu-tracking observers bind the menu nonisolated(unsafe) before assumeIsolated #29871 (ac5389c, now on feat), with [superseded by #29871] fix(next): NativePopupHost compiles under Xcode 26.6 strict sending #29882 and base red: Takes back #28704: NSMenu tracking goes through one seam adapter (compile + assume_isolated/observer rows) #29856 open for the same red. I make no source change here. The next feat merge after that fix lands brings it to this PR.

[lawrencecchen]

I merged the feat tip 33995df into this branch (head 4bf50aa). It brings #29871 (ac5389c), the fix for the NativePopupHost compile base red. The PR's own diff against the merge-base is unchanged: the same 5 files. I removed and added ready-to-land again for this head. The review receipt is for the unchanged diff at eb7600b.

[lawrencecchen]

Review (Claude Opus 5.5 review subagent): SHIP at 4bf50aa

Re-review after the two base merges (69ae06a by cmux-bot, 4bf50aa by me). The PR's own diff against the merge-base is byte-for-byte the diff reviewed at eb7600b. Neither merge resolved a conflict in the PR's 5 files. Earlier non-blocking findings stay open:

• clearRows (ED 1, ED 2, DECALN): fixed in fork PR fix cmux crashing issue #68 part 2. The next pin PR, cp/ghostty-clear-rows-wrap-edges, carries it.
• cursorScrollAbove with the cursor at row 0: not fixed yet. It goes in a follow-up.