# valw: end-of-project test checklist

Manual checks on niri that each phase leaves for one test pass once every
phase is done. Each phase appends its own section; nothing here blocks a
phase from landing. Automated checks (`cargo clippy`, `cargo nextest`,
`nix flake check`) still run per phase.

Before starting: update copland's lock (`nix flake update valw` in
`~/copland`), rebuild, and check that `valw doctor` is all `ok`.

## Preview drag-out (2026-10-02)

Already seen working while building it: a drop into Telegram delivered the
image, a cancel brought the thumbnail back, and a swipe still dismissed it.

- [ ] Drop into Dolphin → the file is copied (the default name has spaces, so this also checks `%20`).
- [ ] Drop into a browser upload field → it uploads.
- [ ] Drop into a Discord message → it attaches.
- [ ] Drop into an image editor that takes `image/png` → the pixels arrive.
- [ ] Release on nothing, or Esc mid-drag → the thumbnail comes back.
- [ ] With `timeout_secs = 3`, hold a drag past 3 s, then release on nothing → it slides out.
- [ ] While dragging, press Print → after the capture the other thumbnails come back but the dragged one doesn't; finishing the drag then behaves normally.
- [ ] Same, with 5 thumbnails on that output and the oldest one being dragged → the dragged one is never evicted.
- [ ] A fast flick to the left and an immediate release → the thumbnail comes back (or the drag runs normally); it never stays invisible.
- [ ] Drag onto the other monitor and drop there → it works.
- [ ] Swipe right still dismisses; a click still opens Satty (installed build).

### Known minors (deferred from the drag-out review)

- A drag-out press that never became a drag (no data device, or a second drag refused) ignores an expired timer; it should slide out on release.
- A receiver that opens its pipe O_NONBLOCK can get a truncated large PNG; clear O_NONBLOCK in the writer thread.
- A press during the slide-in gives a shifted grab point; use (x - offset, y).
- A swipe released back near its start opens Satty (Phase 2 behaviour); a locked swipe should never become a click.
- The (hidden, dragging) visibility/input decision has no unit test; extract it as a pure function.

## Window mode (Phase 3)

- [ ] `valw window`, click a tiled window → that window alone is saved, preview on its monitor.
- [ ] Same for a floating window, and for a window partly covered by another (the covered part is still there).
- [ ] A window on the other monitor (different scale, if any) → right size and sharpness; preview on that monitor.
- [ ] Esc during the pick → nothing saved, exit code 3, thumbnails come back.
- [ ] `--cursor` with the pointer over the window → the pointer is in the shot.
- [ ] `valw region`, press Space before dragging → the overlay closes and the window pick starts; Space while dragging does nothing.
- [ ] `--clipboard-only` and `-o -` → clipboard / stdout get the window; no stray `valw-window-*.png` left in `$XDG_RUNTIME_DIR`.
- [ ] niri's own "Screenshot captured" notification appears (expected, see the spec); the clipboard ends up with the window.

### Known minors (deferred from the window mode review)

- The 5 s wait includes niri's blocking "Screenshot captured" D-Bus notification; a hung notification daemon fails the shot and leaves the late temp file. Try decoding the temp file once on timeout.
- niri sets the clipboard itself even with `copy_to_clipboard = false` (no niri flag to avoid it); the spec's "end state unchanged" is wrong for that config. Document it.
- The Space path keeps the Wayland connection open through the pick (the spec says it is dropped). Harmless; align the spec or drop it earlier.
- Clicking something that isn't a window during the pick exits 3 silently, like Esc. Log "no window under the click".