Caffeine button: claim the right-click so the window menu stays away

[ghackett]

Right-clicking the header's Caffeine cup under GNOME on Wayland opened the compositor's window menu (Take Screenshot, Minimize, Maximize, Close) instead of the durations menu.

Cause. The button's Gtk.GestureClick(button=SECONDARY) popped the durations popover on pressed but never claimed the sequence, so the press kept bubbling up to the header bar's GtkWindowHandle. That handle's own gesture answers any secondary click with gdk_toplevel_show_window_menu, which mutter never refuses, so the window menu took the grab and the popover (which pops a turn later) never showed. GTK's window-handle code is identical between 4.16 and 4.22, so this is not a GTK change; it is a compositor that always serves the menu. Every other right-click menu in the app already claims its gesture.

Fix. The pressed handler now sets the sequence CLAIMED before showing the menu, in MainWindow._on_caffeine_secondary.

Docs. A release note under v0.1.4 UNRELEASED in docs/releases.md. The feature guide's description of the right-click menu is unchanged and was already accurate.

Testing. ruff and the unit suite pass. There is no automated check: nothing in the e2e harness synthesizes a pointer press on the headless compositor. Please right-click the cup on the affected machine to confirm.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RjbzS2d2W6G7TQJ33ugkiy

[ghackett]

@claude review

[claude]

Claude finished @ghackett's task in 1m 6s —— View job

---

Review: Caffeine button right-click fix

• Gather context (diff, changed files, CLAUDE.md rules)
• Understand the request (code review of PR Caffeine button: claim the right-click so the window menu stays away #570)
• Review collins/window.py change
• Review .agents/collins-gtk-sharp-edges/SKILL.md and docs/releases.md updates
• Check GPL modification notice header on collins/window.py
• Post review feedback

Summary: This is a small, correct, well-scoped fix. No changes requested.

collins/window.py

• _on_caffeine_secondary (collins/window.py:3168-3175) calls gesture.set_state(Gtk.EventSequenceState.CLAIMED) before opening the popover — this matches the exact pattern used by every other secondary-click gesture in the codebase (copylabel.py:98, diffview.py:942, filetree.py:404, gitsidebar.py:1049, terminal.py:317, etc.), so it's consistent with the established idiom rather than a one-off workaround.
• The handler signature (self, gesture: Gtk.GestureClick, *_args) correctly absorbs the pressed signal's (n_press, x, y) args.
• The root cause and fix described in the PR body match the diff exactly — AdwHeaderBar's GtkWindowHandle answering unclaimed secondary presses with gdk_toplevel_show_window_menu is a real and previously-undocumented GTK trap.
• GPL modification header bump on collins/window.py (2026-09-10 → 2026-09-16) is present and correctly required since this is a pre-fork file.

.agents/collins-gtk-sharp-edges/SKILL.md

• New bullet accurately generalizes the trap (any header-bar child's right-click gesture, not just Caffeine) and references the fixing PR — good, this is exactly the kind of footgun this skill file exists to capture for future edits.

docs/releases.md

• New UNRELEASED entry is clear, correctly targeted at end users (describes symptom and cause without internals), and matches the existing entries' voice/format.

Testing note: I wasn't able to run ruff or the unit suite myself in this session (Bash tool calls required interactive approval I don't have here), so I relied on manual inspection of the diff plus a grep across the codebase confirming the CLAIMED pattern is used consistently everywhere else. The change itself is small enough (2 lines of real logic) that this doesn't concern me, but flagging it per the instructions in case CI hasn't run yet — check that the e2e required check and lint/unit jobs are green on this PR before merging, since (as the PR body notes) there's no automated regression test for the Wayland compositor behavior itself.

[ghackett]

🤖 Re the testing note in the review: CI has already run on the current head (65340c3). Lint, the unit suite, all five e2e shards and the fanned-in e2e check, packaging, both PPA source builds, the RPM build, version verification and the docs build all passed. Nothing else in the review asked for a change, so no code was pushed.