Two fixes found reviewing the paths the acceptance suite couldn't cover.
Show-while-shown-elsewhere used activate_window, which sets
active_monitor_idx and so dragged *focus* to whichever output the window
was parked on. A scratchpad should come to you: if the window is on
another monitor, remove and re-add it floating on the active workspace;
if it is already on the active monitor, focus in place as before.
add_removed_tile_floating dropped the RemovedTile on both of its early
returns (NoOutputs, monitor-not-found), destroying the window while its
client was still alive and leaving it wedged with no surface anywhere.
Both were unreachable via scratchpad_show's guards, but it is a
window-loss footgun for the next caller. It now returns
Result<(), RemovedTile<W>> so callers can stash the tile instead.
Adds two layout tests (they simulate outputs, so multi-monitor is
covered without a second physical display) and corrects the acceptance
suite's test-8 skip message, which claimed the code used active_output
— it never did; it uses active_workspace.
cargo test --lib scratchpad: 10/10. acceptance.py: 8 passed, 0 failed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stashing a window now stops an active screencast of it, rather than leaving the
consumer on a frozen last frame (a stashed window is never rendered). Covers
both stash paths - Mod+M and Mod+Shift+M's hide branch - by diffing which
windows left the layout across the action, so scratchpad_show's branch logic
isn't duplicated in the caller. Dynamic-target casts are retargeted to Nothing
rather than ended; that carve-out is documented inline.
Also widen the acceptance harness spawn wait to 30s: each test launches its own
nested niri, and a cold alacritty start occasionally exceeded the old 10s.
Deliberately not a retry, which could double-spawn and corrupt window counts.
117 unit tests and 8/8 live acceptance pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review of the destroy-path fix found two Important defects, both now closed:
a stashed window whose client died never stopped its screencast (leaking a
PipeWire stream/portal session nothing could reclaim), and the unmap path had
the identical hole (a stashed window that unmapped was never reinstated into
unmapped_windows, wedging it permanently on re-map). Also corrects a factually
wrong hook-teardown comment. 117 unit tests and 8/8 live acceptance pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds two synthetic multi-output tests (niri's layout harness supports several
outputs) proving scratchpad_show targets the FOCUSED output, plus a control
that it returns to the origin output when focus is unchanged. This is the
substitute for manual multi-monitor verification on a single-display machine.
117 layout tests pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refresh the patch with the destroy-path fix (a client dying while stashed is
now evicted instead of leaked/zombied), and add pkg/niri-scratchpad/tests/
acceptance.py: a headless IPC-driven regression test that drives a nested
patched niri and asserts on `niri msg -j windows`. It found the above bug
(unit tests miss it — the harness calls remove_window directly). Run it after
rebasing the patch onto a new niri release.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace the workspace-based minimize (declared 'minimized' workspace +
+1 offset + window-restore.py) with a native Sway-style scratchpad built
into niri. Ships as niri-patches/0001-scratchpad.patch + a niri-scratchpad
PKGBUILD. Mod+M stashes; Mod+Shift+M shows/toggles/cycles. Mod+1..9 no
longer offset.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
funman300