Pax requested the keyboard shortcut be removed — too easy to fat-finger
when navigating with the keyboard, and "Open in Default App" still
ships an external process that may steal focus from the app. The
right-click menu's Open in Default App action stays, both on browse
thumbnails and in the preview pane right-click — only the bare-key
shortcut goes away.
The deleted block was the only Key_O handler in BooruApp.keyPressEvent,
so no other behavior changes.
Verified manually:
- press O on a selected thumbnail → nothing happens
- right-click thumbnail → "Open in Default App" still present and opens
- right-click preview pane → same
Two related fixes for the File → Batch Download Page (Ctrl+D) flow.
1. Saved-dot refresh
Pre-fix: when the user picked a destination inside the library, the
batch wrote files to disk but the browse grid's saved-dots stayed dark
until the next refresh. The grid was lying about local state.
Fix: stash the chosen destination as self._batch_dest at the dispatch
site, then in _on_batch_progress (which already fires per-file via
the existing batch_progress signal) check whether dest is inside
saved_dir(); if so, find the just-finished post in self._posts by id
and light its grid thumb's saved-locally dot. Dots appear incrementally
as each file lands, not all at once at the end.
The batch_progress signal grew a third int param (post_id of the
just-finished item). It's a single-consumer signal — only
_on_batch_progress connects to it — so the shape change is local.
Both batch download paths (the file menu's _batch_download and the
multi-select menu's _batch_download_posts) pass post.id through.
When the destination is OUTSIDE the library, dots stay dark — the
saved-dot means "in library", not "downloaded somewhere". The check
uses Path.is_relative_to (Python 3.11+).
self._batch_dest is cleared in _on_batch_done after the batch finishes
so a subsequent non-batch save doesn't accidentally see a stale dest.
2. Tab gating
Pre-fix: File → Batch Download Page... was enabled on Bookmarks and
Library tabs, where it makes no sense (those tabs already show local
files). Ctrl+D fired regardless of active tab.
Fix: store the QAction as self._batch_action instead of a local var
in _setup_menu, then toggle setEnabled(index == 0) from _switch_view.
Disabling the QAction also disables its keyboard shortcut, so Ctrl+D
becomes a no-op on non-browse tabs without a separate guard.
Verified manually:
- Browse tab → menu enabled, Ctrl+D works
- Bookmarks/Library tabs → menu grayed out, Ctrl+D no-op
- Batch dl into ~/.local/share/booru-viewer/saved → dots light up
one-by-one as files land
- Batch dl into /tmp → files written, dots stay dark
The infinite-scroll backfill loop in _on_reached_bottom accumulates
results from up to 9 follow-up API pages until len(collected) >= limit,
but the break condition is >= not ==, so the very last full batch
would push collected past the configured page_size. The non-infinite
search path in _do_search already slices collected[:limit] before
emitting search_done at line 805 — the infinite path was emitting the
unclamped list. Result: a single backfill round occasionally appended
more than page_size posts, producing irregular batch sizes the user
could see.
Fix: one-character change at the search_append.emit call site to mirror
the non-infinite path's slice.
Why collected[:limit] over the alternative break-early-with-clamp:
1. Consistency — the non-infinite path in _do_search already does
the same slice before emit. One pattern, both branches.
2. Trivially fewer lines than restructuring the loop break.
3. The slight wasted download work (the over-fetched final batch is
already on disk by the time we slice) is acceptable. It's at most
one extra page's worth, only happens at the boundary, only on
infinite scroll, and the next backfill round picks up from where
the visible slice ends — nothing is *lost*, just briefly redundant.
Verified manually on a high-volume tag with infinite scroll enabled
and page_size=40: pre-fix appended >40 posts in one round, post-fix
appended exactly 40.
The browse grid's multi-select right-click menu collapsed library and
bookmark actions into a single "Remove All Bookmarks" entry that did
*both* — it called delete_from_library and remove_bookmark per post,
and was unconditionally visible regardless of selection state. Two
problems:
1. There was no way to bulk-unsave files from the library without
also stripping the bookmarks. Saved-but-not-bookmarked posts had
no bulk-unsave path at all.
2. The single misleadingly-named action didn't match the single-post
right-click menu's clean separation of "Save to Library / Unsave
from Library" vs. "Bookmark as / Remove Bookmark".
Reshape: split into four distinct actions, each with symmetric
conditional visibility:
- Save All to Library → shown only if any post is unsaved
- Unsave All from Library → shown only if any post is saved (NEW)
- Bookmark All → shown only if any post is unbookmarked
- Remove All Bookmarks → shown only if any post is bookmarked
Mixed selections show whichever subset of the four is relevant. The
new Unsave All from Library calls a new _bulk_unsave method that
mirrors the _bulk_save shape but synchronously (delete_from_library
is a filesystem op, no httpx round-trip). Remove All Bookmarks now
*only* removes bookmarks — it no longer touches the library, matching
the single-post Remove Bookmark action's scope.
Always-shown actions (Download All, Copy All URLs) stay below a
separator at the bottom.
Verified:
- Multi-select unbookmarked+unsaved posts → only Save All / Bookmark All
- Multi-select saved-not-bookmarked → only Unsave All / Bookmark All
- Multi-select bookmarked+saved → only Unsave All / Remove All Bookmarks
- Mixed selection → all four appear
- Unsave All from Library removes files, leaves bookmarks
- Remove All Bookmarks removes bookmarks, leaves files
Final commit of the gui/app.py + gui/preview.py structural refactor.
Updates the four call sites that were importing through the
preview.py / app.py shims to import from each entity's canonical
sibling module instead, then deletes the now-empty shim files.
Edits:
- main_gui.py:38 from booru_viewer.gui.app import run
→ from booru_viewer.gui.app_runtime import run
- main_window.py:44 from .preview import ImagePreview
→ from .preview_pane import ImagePreview
- main_window.py:1133 from .preview import VIDEO_EXTENSIONS
→ from .media.constants import VIDEO_EXTENSIONS
- main_window.py:2061 from .preview import FullscreenPreview
→ from .popout.window import FullscreenPreview
- main_window.py:2135 from .preview import FullscreenPreview
→ from .popout.window import FullscreenPreview
Deleted:
- booru_viewer/gui/app.py
- booru_viewer/gui/preview.py
Final gui/ tree:
gui/
__init__.py (unchanged, empty)
app_runtime.py entry point + style loader
main_window.py BooruApp QMainWindow
preview_pane.py ImagePreview embedded preview
info_panel.py InfoPanel widget
log_handler.py LogHandler (Qt-aware logger adapter)
async_signals.py AsyncSignals signal hub
search_state.py SearchState dataclass
media/
__init__.py
constants.py VIDEO_EXTENSIONS, _is_video
image_viewer.py ImageViewer (zoom/pan)
mpv_gl.py _MpvGLWidget, _MpvOpenGLSurface
video_player.py VideoPlayer + _ClickSeekSlider
popout/
__init__.py
viewport.py Viewport NamedTuple, _DRIFT_TOLERANCE
window.py FullscreenPreview popout window
grid.py, bookmarks.py, library.py, search.py, sites.py,
settings.py, dialogs.py (all untouched)
Net result for the refactor: 2 god-files (app.py 3608 lines +
preview.py 2273 lines = 5881 lines mixing every concern) replaced
by 12 small clean modules + 2 oversize-by-design god-class files
(main_window.py and popout/window.py — see docs/REFACTOR_PLAN.md
for the indivisible-class rationale).
Followups discovered during execution are recorded in
docs/REFACTOR_NOTES.md (gitignored, local-only).
Step 12 of the gui/app.py + gui/preview.py structural refactor — the
biggest single move out of app.py. The entire ~3020-line BooruApp
QMainWindow class moves to its own module under gui/. The class body
is byte-identical: every method, every signal connection, every
private attribute access stays exactly as it was.
main_window.py imports the helper classes that already moved out of
app.py (SearchState, LogHandler, AsyncSignals, InfoPanel) directly
from their canonical sibling modules at the top of the file, so the
bare-name lookups inside BooruApp method bodies (`SearchState(...)`,
`LogHandler(self._log_text)`, `AsyncSignals()`, `InfoPanel()`) keep
resolving to the same class objects. Same package depth as app.py
was, so no relative-import depth adjustment is needed for any of
the lazy `..core.X` or `.preview` imports inside method bodies —
they keep working through the preview.py shim until commit 14
swaps them to canonical paths.
app.py grows the BooruApp re-export shim line. After this commit
app.py is just imports + log + the four helpers (run,
_apply_windows_dark_mode, _load_user_qss, _BASE_POPOUT_OVERLAY_QSS)
+ the shim block. Commit 13 carves the helpers out, commit 14
deletes the shims and the file.
VERIFICATION: full method-cluster sweep (see docs/REFACTOR_PLAN.md
"Commit 12 expanded verification" section), not the 7-item smoke test.
