- Reword the header as one sentence with a result-level menu ("Find
[attachments] matching [all] of the following:")
- Provide a per-group menu to bind the group's descendant conditions to
the same attachment, note, or annotation (e.g., one annotation that is
both red and contains a given word, not two different ones)
- Show a hint that offers to group ungrouped sibling conditions (e.g.,
two annotation conditions at the top level, to bind them to one
annotation)
- Show a warning when conditions can't combine at the chosen result
level (e.g., an annotation condition with a note result level)
- Remove the two legacy checkboxes:
- "Show top-level items" becomes result level = top-level item and is
migrated on save
- "Include parent and child items", which has no result-level
equivalent, keeps working, stays editable, and round-trips on
searches that already have it, but it isn't offered on new searches
and is removed on save if unchecked
Give a search a result level -- top-level item, attachment, note, or
annotation -- and map every condition to that level, so one search can
mix conditions that match at different levels of the item hierarchy
(e.g., a top-level item with a given author and a red annotation on one
of its PDFs). Each condition carries the level(s) it matches at: a match
is mapped up to an ancestor or down to a descendant, level-agnostic
conditions (tags) roll up to the result level, and fields that exist on
both items and attachments (title, url, accessDate) match natively at
either. The result level is stored as a `resultLevel` marker condition
alongside the join mode.
This also removes the temporary annotation-parent hacks, which the
general cross-level mapping replaces.
A fulltextContent condition was evaluated as a global post-filter keyed
on the search's top-level join mode -- correct for a top-level
condition, but not for one inside a group, which must combine with its
siblings under the group's own join mode. The new condition grouping UI
allows fulltextContent to be placed within groups, and we need to do so
to prefill the advanced-search pane from an "Everything" quicksearch.
Materialize a grouped fulltextContent into an itemID set and emit it as an
ordinary itemID IN/NOT IN predicate, so combineConditions composes it
under the group's join mode. Top-level fulltextContent keeps the existing
post-filter unchanged.
This also removes the quicksearch full-text post-filter special case. A
quick search puts its full-text in per-word "any" groups, so the
_hasQuicksearch flag was needed to make the post-filter union those
matches rather than intersect them under the top-level "all" join. Now
that the grouped full-text is composed in SQL it never reaches the
post-filter, so the flag is gone.
With condition groups, the position of conditions and the pairing of
groupStart/groupEnd markers are meaningful, but conditions were diffed
as an unordered member set (compared by value, with additions appended).
A sync-conflict merge could reorder conditions or add/drop group
markers, corrupting the group structure.
Diff them as a single ordered unit instead, like creators. This changes
how concurrent edits to a search are reconciled: rather than merging the
two sides' conditions member by member, a conflicting edit now resolves
wholesale -- the remote condition list replaces the local one (searches
auto-merge to the remote version); a one-sided change still applies that
side's full list. Discarding one side of a rare simultaneous edit is
acceptable and avoids silently corrupting a grouped search's structure.
This was the only user of SearchConditions.equals(), so remove it.
Render the search as a tree of groups: a root group plus nested
search-condition-group elements, each with its own join-mode menu and a
remove control. Each condition row gets a "( )" button that wraps it in
a new group in place, so further conditions can be added to combine with
it under a separate join mode. Switch the builder to rebuild-from-tree --
the DOM is the source of truth and the search's flat conditions (with
groupStart/joinMode/groupEnd markers) are regenerated on each edit, so
the old conditionID-as-index tracking is gone.
Replace the flat anySQL/quicksearch-block assembly in _buildQuery with a
recursive tree of AND/OR groups, built and reduced by a new
Zotero.Search.combineConditions() helper. groupStart/groupEnd markers
delimit nested groups and a joinMode marker sets each group's mode, so a
saved search can combine conditions with arbitrary nesting and per-group
join modes. The per-condition SQL generation is unchanged.
For example, a search built as
search.addCondition('joinMode', 'all');
search.addCondition('title', 'contains', 'foo');
search.addCondition('groupStart', 'true', '');
search.addCondition('joinMode', 'any');
search.addCondition('tag', 'is', 'x');
search.addCondition('tag', 'is', 'y');
search.addCondition('groupEnd', 'true', '');
means "title contains 'foo' AND (tag is 'x' OR tag is 'y')". The 'true'
operator on the group markers is an unused placeholder -- they carry no
value, but a condition's operator can't be empty.
The quick search (matching multiple words) and the Any Field condition
previously had their own special handling in the query builder; they now
use the same grouping as everything else, so that code is gone. Behavior
for existing non-grouped searches is unchanged; new tests cover nested
groups and combineConditions() directly.
A plain select-all would mix uncombinable rows, so scope it to the
current selection's type: a library selection expands to all library
roots; a collection selection to every collection sharing a parent with
a selected one (so multi-level/multi-parent selections expand within
each branches); and a Recently Read selection to Recently Read in every
library. Other rows -- saved searches, Unfiled, Trash, etc. -- have no
useful expansion, so the tree's key handler leaves the selection
untouched instead of clearing it.
Version Read Aloud cache keys with the server-provided cacheVersion so
a version bump misses stale entries and re-fetches correct audio, prune
obsolete entries once per session, and skip caching responses sent with
Cache-Control: no-store.
Regressed by f4adb452ee, which dropped the items-tree body's top padding
to 0 for sticky section headers; the scroll container's overflow then
clipped the top of the first row's focus ring, which drew outside the
row box.
The pinned sticky section header was opaque to paint but had
pointer-events: none, so clicks, drag-starts, and drops fell through to
the item occluded underneath it.
Make the opaque content capture pointer events instead. Clicks no-op
(header rows aren't selectable), drags don't start (header rows aren't
draggable), and drops on the header are rejected so they're a no-op
rather than acting on the list underneath.
Also drop the index > 0 exception when skipping non-selectable rows in
_onSelection(), so clicking a non-selectable row at the top (i.e., the
first library's header) is a true no-op instead of scrolling to the top.
Also skip non-selectable rows in handleActivate(), since the header can
now be double-clicked; without this it would try to open the library as
an item.
Fixes#5960
Arrow-key navigation scrolled the newly selected row flush with the top
of the view, leaving it hidden behind the pinned sticky section header.
Reserve a row's worth of space at the top so the row lands below the
pinned header.
Fixes#5959
#5658 added includeDeleted to the Advanced Search outside the trash, so
trashed items kept matching: trashing a result removed its row, but
re-running the search brought it back. Exclude deleted items by default,
the same as a quick search, and only include them when viewing the trash,
where the scope returns only deleted items.
Fixes#5956
When viewing the trash, trashed collections and saved searches were appended
to the items list unconditionally, so every advanced search (and quick search)
in the trash matched all of them. Now they're filtered by name during a quick
search and excluded entirely when an advanced search or tag filter is active,
since they can't match item-level conditions or tags.
Fixes#5957
The refactor dropped the 8px (COLUMN_PADDING / 2) offset that makes up
for the inline-start padding the first cell omits, so the header label
sat 8px too far left of the item titles.
Pin the header inside the scrolling body as a zero-height position:
sticky element so its width tracks the body's content box and it lines
up with the rows without any JS geometry. Reserve a scrollbar gutter so
the macOS overlay scrollbar doesn't float over the content (the opaque
header must paint above the rows to occlude them, and so above the
scrollbar, so it can't be put under it). Drop the body's top padding for
the item tree so rows clip exactly where the header pins, and keep the
gap below the column header as a margin outside the scroll area.
Fixes#5958
Route each item by its own library -- items already in the target
library are added directly (or skipped, for a library root), while
items from other libraries are copied in -- instead of attempting an
invalid cross-library insert. Disallow a move of such a selection
rather than silently copying.
Fixes#5961
The collections passed through Notes.open into the note window were never
read -- the note editor has no collections setter and EditorInstance.collection
is never assigned on this path -- so they had no effect. New notes are still
added to the selected collection(s) directly in newNote().
When multiple collection-list rows are selected, group the combined
items by library under sticky headers (e.g., "My Library", "Group X (2
collections selected)"), or show a single summary header for a multi-row
selection within one library, with blank spacer rows separating
libraries.
When stickySectionHeaders is enabled, the header of the section at the
top of the view is pinned in an overlay that the rows scroll under,
pushed up as the next section's header arrives. Consumers identify
header rows via isSectionHeader. Row striping restarts at each section
header so every section's first row is the same shade.
stickySectionHeaders defaults to false, so existing tables are
unaffected.
When the items list contains items from more than one library, group
them by library -- in collections-list order, independent of the active
sort -- with a section heading above each library's items.
Grouping is triggered automatically by an items list spanning more than
one library, not the kind of selection behind it, so any future source
of multi-library items would be separated the same way. Today the
cross-library collection selection is the only such source.
Allow selecting multiple collections, saved searches, or library roots in the
collection tree -- within a library or across libraries -- and show the union
of their items. The selection is threaded through the pane as an array
(getCollectionTreeRows(), changeCollectionTreeRows(), etc.); the item pane, tag
selector, reports, and export operate on all selected rows.
Adding items (new items and notes, drag-and-drop, the attachment dialog,
import, Add by Identifier) targets every selected collection. Only rows that can
share an items view may be combined: collections, saved searches, and library
roots mix freely, and multiple Recently Read rows can be combined across
libraries, but other special views (Trash, Duplicates, etc.) and rows from
different visibility groups can't be shown together, so a selection mixing them
keeps only the focused row. In-window advanced search runs across all selected
collections.
Advanced search value autocomplete is now scoped to the searched library, fixing
a long-standing TODO where suggestions were drawn from all libraries regardless
of the search scope; for a cross-library selection it spans the selected
libraries.
For a cross-library selection, the tag selector shows the union of tags
(colored tags only when a single library is in scope, since colors are
per-library), and deleting a tag spans all selected libraries while
rename/color/split are disabled. If advanced search is open, the
collection and saved-search conditions are omitted, since each is scoped
to a single library.
On a fresh profile, the default sort column had no explicit sortDirection,
so the first header click set the default direction instead of reversing
the sort, and no sort indicator was shown.
In other words, don't show the children that make the parent "recently
read" as context rows.
Children that don't match the Recently Read condition (such as notes,
as well as other child attachments that weren't read since the cutoff)
are still shown as context rows.
And add tests for the new behavior, and clean up an unused local var.
- Include the Advanced Search button in the quick-search tab sequence
- Tab: quick-search textbox -> Advanced Search button -> item tree
- Shift-Tab: Advanced Search button -> quick-search textbox
- Shift-Tab: item tree -> Advanced Search button (when the deck is
closed)
- Shift-Tab: Advanced Search collapse button -> New Item button
- Fix Shift-Tab from the item tree doing nothing when the Advanced
Search deck is open; now moves focus to the deck's last focusable
element
- add tooltiptext to expand/collapse and close btns
PubMed saves can include a PMCID even when DOI-based full-text lookup
fails to produce a PDF. In that case the connector asks the desktop
attachment resolver flow for fallback sources, but getFileResolvers()
only considered DOI and URL data, so Zotero never tried the matching
PMC article URL.
Treat PMCID as a valid signal for file lookup eligibility and add a
PubMed Central OA resolver before DOI-based OA lookup. Cover the new
fallback path in attachment resolver tests and the connector
hasAttachmentResolvers endpoint.
1. 862573eab added references to fileRenaming.ftl strings but didn't
add a <link> to it, so document.l10n.formatMessages() returned
nothing. It fell back to Zotero.getString(), which threw an error
due to a missing string in en-US. We didn't catch that, so the entire
search process aborted every time.
2. message.attributes is an optional array of { name, value } objects,
not an object itself, so we were never successfully adding FTL .title
and .label attributes to the parsed strings.
This fixes the 862573eab regression, and makes search match FTL
attributes as it was always intended to.
Fixes#5947
(Confirmed on macOS, but probably Linux too)
Pass openNotExclusive to avoid acquiring an exclusive lock at the OS level
at open time, which fails with an I/O error on network shares. We still set
locking_mode=EXCLUSIVE, so the connection holds an exclusive SQLite lock for
its lifetime and keeps the WAL index in heap memory (no -shm file).
Regression in Zotero 7 (Fx115)
Fixes#4860
The temporary advanced search saved silently under an auto-generated
"Untitled" name. Prompt for a name instead (defaulting to that name),
and label its button "Save Search…" to signal the dialog. The saved-search
editor's button, which saves changes directly, is now just "Save".
https://github.com/zotero/zotero/pull/5658#issuecomment-4696398750
Deleting the saved search being edited triggered onCollectionSelected's
save-changes prompt, which doesn't make sense for a search that's gone.
Detect the deleted search and close the editor without prompting.
