[teamleaderleo]

#23182 was one bug of a class: an incremental sidebar path drifted from what a full rebuild draws, and the drift stayed until a rebuild. This adds one invariant test for the class, and fixes the one drift it found.

The invariant

SidebarIncrementalMatchesRebuildTests runs seven operations on a real SidebarView, through the paths the app uses (the sidebar document's drag, the model's fold, a click, an external tab drag), then the daemon's echo:

• a group dragged past an empty group (cmux-next sidebar: a group dropped past an empty group keeps its members' rail #23182)
• a group dragged past a folded group, then that group opened
• a folded group dragged up
• a reorder inside a group, a loose workspace into a group, a member out of its group
• a tab dragged across groups, springing a folded group open on the way

After each step, and after every group then folds and opens and a few rows are clicked, it compares the list on screen with a list built fresh from the same model: rows, their order and frames, row views (frame, alpha, selection, untracked strays), drawn rails, the list's rail records, selection, suppressed rows and the gap. Motion is off (Reduce Motion), so the list shows the spring's ends. A mismatch prints a field-by-field diff.

The drift it found

A fold put the folded group's rail last in the list's rail records (foldLines), and showMissing (#23182's path) appended the rail it showed. The rails drew in the right place, but in a different order from a rebuild: 12 mismatches over 6 of the 7 operations, all this one cause. Both now insert a rail at its place in the list, top to bottom.

Changelog

none

No capture: nothing users see changes. The fix only reorders the list's internal rail records (groupLineViews.lines); every rail already drew at the same frame.

Observed / Not run

• Red (test only, 4d32197 plus the base compile fix): package-test-lane.sh ... SidebarIncrementalMatchesRebuildTests, fleet step 98ba6d7e5236: 12 issues, every one a rail-order diff, e.g. order: [g-infra, g-work] vs fresh: [g-work, g-infra] after Work folds and opens.
• Green (test and fix, 3de49be plus the same base fix): same command, fleet step 8b99271d3821: passes.
• This head merges feat-cmux-next at 0284c98 (with cmux-next tests: RemoteChildSessionsTests' fake client states isReadOnly (base red: tests don't compile) #23362, the base compile fix); prepush-gate clean. Not run on this merged head yet; CI runs it.
• Review: priority-view PASS on 81a8496 (comment 6315717000). Its note on tabAcrossGroups (the step passes on an unexpected drop) is a follow-up.
• Not covered: Group by Folder, filters, Pinned and custom sections, multi-row drags, and drags with motion on (mid-flight frames).

🤖 Generated with Claude Code

Agent session: cosmic-gray

[teamleaderleo]

Owned by claude-sr2/sidebar-drag-bugs on cosmic-gray · session

[github-actions]

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

Review 81a8496: 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=pr23389 -f ref=cmux/sidebar-rebuild-invariant -f platform=macos -f backend_mode=local

Not verified: macOS jobs did not run: compile and app tests were skipped on 81a8496efa; ci-status is green without them.

CI passes on 81a8496efa (run 39598285143 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.

[teamleaderleo]

PASS (review on head 81a8496, correctness first; priority-view lane, session 6807b41c).

The test goes through the app's paths.

• Folds go model.send(.toggleCollapse(.group)) → SidebarDocStore+Groups → model.onGroupFolded (SidebarListView.swift:154) → foldGroup → applyFold → foldLines. This is the incremental path, not a reload. Its guards (drag/settling/landing == nil) hold because the test finishes each landing before it folds.
• Drags go beginDrag / updateDrag / finishDrag with a real SidebarDocDrag, so the document's acceptance table and drop callback both run. The test also checks that doc.drops is not empty. The drop's settle reaches showMissing (SidebarListView+Drag.swift:318).
• The fresh list is a separate SidebarView on a copied model, and the fold state travels in the sections. The red/green pair (98ba6d7e5236 → 8b99271d3821) shows the fold path ran.

The fix is correct.

• In foldLines, the lines below the header are shifted by dy first. They stay below header.y, so inserting own at the first line with minY > header.y puts it between the lines above and below the group.
• showMissing inserts by minY, and a tie goes after, which is fine.

Non-blocking notes

1. Only the rail lines field catches this: the drawn rails are sorted before the compare. Today the order of groupLineViews.lines has no visible effect; its only reader is foldLines, which filters it by group. So the fix keeps the internal records consistent and prevents later drift; it does not fix anything users see. The PR body says this correctly. The cubic summary ("drawn rails were in a different order") overstates it.
2. In tabAcrossGroups, default: break together with #expect(drop != nil) passes even when the drop lands somewhere unexpected, and then the echo is skipped. Asserting the expected case, into Work between docs and web, would make that step bite.

I did not run it; I relied on your fleet red and green and on CI for the merged head.

[teamleaderleo]

Bead cx-zb9fl created for this PR by the cmux coordinator.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.3960833477001 (4482e41d31)
• cmux NEXT nightly 1.0.0-nightly.3960894798201 (0b1a93e9c0)
• cmux NEXT nightly 1.0.0-nightly.3961101575901 (101cc33db2)