fix(core): forget click targets on an interface reload

[LeTuR]

Intent

Resolve thurbox issue #1118 (Closes #1118): a press that arrives after an interface reload but before the next paint must never reach a plugin that did not paint the node under the pointer, for left AND right presses alike (the bug predates #1116 and hits both). Root cause per the issue: App::click_targets records plugin INDICES from the last paint; reload_interface (src/coordinator/interface.rs) rebuilds the plugin vector and reset grabbed but not click_targets, so target_at in src/coordinator/mouse.rs resolved a press to whichever plugin moved into the stale index, carrying the old node's id. The issue's own suggested fix was followed: clear click_targets next to grabbed = None in reload_interface, with the comment there updated to cover it — that is why the change is in interface.rs rather than mouse.rs, which the task brief had guessed. Constraints from the requester: write the failing end-to-end test FIRST and watch it fail for the issue's reason (done: tests/tui_e2e.rs a_press_right_after_a_reload_never_reaches_a_pane_that_did_not_paint_it runs the real binary on a pty with three stacked panes gone/aim/near, deletes gone, then sends F10 + a right press + a left press on aim as ONE write so drain_input handles all three before the next paint; before the fix near painted tb-near-R:aimL:aim, after it near hears nothing). No scope beyond the issue; do not widen any public API. Known, deliberately out of scope: pointer_grab also stores a plugin index and is not cleared on reload (reported separately, not fixed here). Known pre-existing flake unrelated to this change: tui_e2e the_wheel_scrolls_the_agents_output_back times out intermittently and fails on unmodified main 116fdb7 too (1 of 3 runs). PR description must follow the Intent/What Changed/Risk Assessment/Testing/Pipeline shape, be concise with no praise, and end with the plain last line '— LeTuR's agent'. Squash-merge title should be a conventional commit, e.g. fix(core): forget click targets on an interface reload.

What Changed

• reload_interface (src/coordinator/interface.rs) now clears click_targets alongside resetting grabbed, so a stale plugin index from before the reload can no longer be resolved against the rebuilt plugin vector.
• Updates the surrounding comment to explain that both grabbed and click_targets hold positions from the last paint, and that a reload between a paint and an input can leave them pointing at a different plugin that never painted the node under the pointer (Stale click targets after an interface reload route a press to the wrong plugin #1118).
• Adds an end-to-end test (tests/tui_e2e.rs) that runs the real binary on a pty with three stacked panes, deletes one, then sends a reload followed by a right press and a left press on another pane as a single input batch, asserting the press never reaches a pane that did not paint that node.

Risk Assessment

✅ Low: The fix is a minimal, correctly-targeted one-line addition (clearing click_targets alongside grabbed in reload_interface) that closes the exact stale-index race described in issue #1118 for both left and right presses, backed by a real pty-driven e2e test that reproduces the bug via a single batched write (F10 + right + left press) and verifies both the fixed path (target_at returns None until the next paint) and the previously-buggy path; the deliberately out-of-scope pointer_grab index is untouched as intended, and the updated comment accurately reflects the new behavior.

Testing

Baseline cargo nextest run --all already passed. Beyond that, I ran the specific new e2e test end-to-end on a real pty: on the fixed commit it passes, and after temporarily reverting only the interface.rs fix (keeping the test), it fails for exactly the reason issue #1118 describes — a stale plugin index causes the pane that moved into a deleted pane's old slot to receive a press (both right-click and left-click) meant for a different pane it never painted. Restored the fix afterward and confirmed the worktree is clean. The related #1116 right-button routing test still passes, showing no regression. This directly demonstrates the intended end-user behavior (mouse press routing after a live interface reload) rather than relying on unit-test pass/fail alone.

Evidence: tui_e2e regression test failing before fix (base commit 116fdb7)

a_press_right_after_a_reload_never_reaches_a_pane_that_did_not_paint_it panicked:
"a press after a reload reached a pane that did not paint the node" — frame shows tb-near-R:aimL:aim, i.e. the stale pane 'near' received both the right-press and left-press meant for 'aim' after the reload.

Evidence: tui_e2e regression test passing after fix (target commit 17af69b)

test a_press_right_after_a_reload_never_reaches_a_pane_that_did_not_paint_it ... ok
test the_right_button_reaches_on_context_and_the_left_one_still_reaches_on_click ... ok

- Outcome: 🔧 1 issue found → auto-fixed ✅ across 2 runs (11m39s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

🔧 **Test** - 1 issue found → auto-fixed ✅

• 🚨 tests failed with exit code 100
• cargo nextest run --all

🔧 Fix: Confirm click_targets fix passes; other failures were disk-quota flakes
✅ Re-checked - no issues remain.

• cargo nextest run --all
• cargo nextest run --test tui_e2e a_press_right_after_a_reload_never_reaches_a_pane_that_did_not_paint_it (target commit 17af69b) — passes
• Manually reverted src/coordinator/interface.rs to the pre-fix (base commit 116fdb7f0) version while keeping the new test, reran the same test — fails with the exact reported symptom (stale pane 'near' receives 'aim's right+left press: tb-near-R:aimL:aim), then restored the fixed file and confirmed git status is clean
• cargo nextest run --test tui_e2e -E 'test(a_press_right_after_a_reload) + test(the_right_button_reaches_on_context)' (target commit) — both pass, confirming no regression in the adjacent #1116 right-button routing test

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

— LeTuR's agent

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[greptile-apps]

Greptile Summary

This PR prevents stale click-target indices from being reused after an interface reload and adds an end-to-end regression test for both right and left presses.

• Clears click_targets when rebuilding the plugin vector.
• Updates the lifecycle comment to document why pre-reload targets are invalid.
• Verifies through the PTY-driven test that batched presses cannot reach a plugin that did not paint the targeted node.

Confidence Score: 5/5

The PR appears safe to merge; the state reset is narrowly scoped to invalid paint-derived targets and the regression test covers the stale-index routing failure.

Reloading now invalidates click targets at the same point the indexed plugin vector is rebuilt, while the next paint restores valid targets; no actionable correctness, security, or repository-rule issues remain.

Important Files Changed

Filename	Overview
src/coordinator/interface.rs	Clears paint-derived click targets during interface reload so stale plugin indices cannot route presses into the rebuilt plugin vector.
tests/tui_e2e.rs	Adds a PTY-driven regression test covering batched reload, right-click, and left-click input before repaint.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Paint interface] --> B[Record click targets with plugin indices]
    B --> C[Reload interface and rebuild plugin vector]
    C --> D[Clear grabbed and click targets]
    D --> E{Next paint completed?}
    E -- No --> F[Press has no valid target and is ignored]
    E -- Yes --> G[Record targets for rebuilt plugins]
    G --> H[Route subsequent presses using fresh targets]

Loading

_{Reviews (1): Last reviewed commit: "fix(core): forget click targets on an in..." | Re-trigger Greptile}