The overlap sweep moves a horizontal line down the page, and its time
per event grows with the number of segments on that line. A long, low
path, such as rows of a zigzag, puts a whole row on the line at once.
stroke_region, unite_all_with and try_unite_all_with now estimate how
many segments the sweep holds at once in each direction, from up to
4,096 segment boxes (one from each run of segments, at a position that
changes from run to run). When the estimate for the input turned a
quarter turn is under two thirds of the upright one (2,048 segments or
more), the operation runs on the turned input and the result is turned
back. A quarter turn swaps and negates coordinates, so no point moves
by rounding. Offset Path is not turned. Callers of unite_all,
try_unite_all and outline_stroke take the turned sweep through these
functions.
Tidy's lookups of the input curves around a piece and of the kept
anchors can search along y: once lookups with more than 64 candidates
along x have scanned 16 per item, a lookup sorts the list along y once
and uses that window when it holds fewer candidates (under a quarter as
many for curves). Where the y window rarely helps, only one lookup in
32 measures it. is_kept counts the few points within its tolerance
directly instead of searching for the window's end.
In the sweep_axis tests, a turned result has as many subpaths as the
result without the turn, and the area where the two differ is no
larger than the refit precision times the outline's length.
On one open path of 200,000 anchors in rows of 1,000, Outline Stroke
takes 17.3 s of processor time instead of 31.3 s and Unite 2.9 s
instead of 6.5 s in a release build. Unite of 2,000 circles takes the
same time as before.
Reset Bounding Box set the angle to 0 on the selected objects only. A
rotation changes the box angle of the selected objects and of every
object inside them, so the members of a rotated group kept their angle:
after rotating two rectangles 20 degrees, grouping them, rotating the
group 10 degrees and resetting, Ungroup brought back a box at 30
degrees, and a second reset reported 2 objects to change.
The command now also squares every object inside the selected ones: the
children of groups, compound paths, clip groups, blends, envelopes,
repeats and intertwines, and the art of linked opacity masks. The art of
an unlinked mask does not turn with the object, so it keeps its angle.
Ungroup, Release Clipping Mask, Release Compound Path, Blend Expand,
Release Envelope, Release Repeat, Release Intertwine and Release Opacity
Mask then bring back boxes square to the page. The whole reset stays one
undo step.
Two results change. `changed` counts every object reset, the objects
inside included, so the steps above return 3 where they returned 1. A
group whose own box is square but whose members are turned used to
return 0 with no undo step; it now squares the members in one step.
Docs: params text, docs/mcp.md, docs/target-app-parity.md.
When only anchors of a path or compound path move, the area an edit
changed is now the box of the segments that start or end at a moved
anchor (and the line that closes an open subpath's fill at a moved
end), before and after, grown by the farthest the strokes reach. The
canvas then draws that part of the frame again instead of the whole
object. Objects with a non-solid paint, dashes, a width profile,
arrowheads, a brush, effects or an opacity mask still count whole.
At 1600 x 1200, single-threaded, the frame after one anchor of a filled
closed path moves takes 16.6 million CPU cycles instead of 56.4 million
with 50,000 anchors, and 65.6 million instead of 180.0 million with
200,000 anchors.
A PDF saved with some of its artboards has a page for each of them, while
the editing data keeps them all. The importer took the pages after the
first to show the artboards right after the first page's artboard. In a
PDF of artboards 2 and 4 of 4, the second page's type went to artboard 3.
In a PDF of artboards 1 and 3 of 4, the second page's type went to the
left edge of artboard 2.
Each page after the first now goes to an artboard of its size that comes
after the previous page's artboard and leaves an artboard for each later
page. The artboard chosen for the first page also leaves one for each
later page. When several artboards fit a later page, the importer
compares their art with the page and takes the closest one, or the first
of them on a tie. It compares at most 16 artboards for all the later
pages together, the same limit it uses for the first page. A later page
with no comparisons left takes the first artboard that fits it.
When the PDF has fewer pages than artboards and its first page shows the
first artboard, the first page's type and page art still go to the art's
box or to the editing data's artboard.
The Preferences dialog has a Search Preferences field above the category
list. It has the keyboard focus when the dialog opens, and Cmd/Ctrl+F
focuses it again. While it has text, the list shows every setting whose
label, section or category contains it (ignoring case, in the interface
language or in English), or "No settings found". Choosing a setting shows
its category, scrolls to its row and highlights it. Escape in the field
with text clears it; Escape in the empty field closes the dialog.
The dialog's `__search` field holds the text, so `ui.dialog.set` drives
the search too.
shortcuts.export and shortcuts.import read `path` only when it was a
string, so a `path` such as 5 or true was dropped and the file dialog
opened. They now return "`path` must be a string" and open no dialog;
an omitted or null `path` still asks for the file.
shortcuts.set (`id`, `shortcut`), shortcuts.list (`query`) and
shortcuts.preset (`name`) give the same error for a value that is not a
string, and shortcuts.set gives "`force` must be true or false" for a
`force` that is not true, false or null. Nothing changes when a param
is rejected.
A short move with View > Snap to Pixel on returned to where it began and
showed a "last location" label, skipping the pixel step. Snap to Last
Location is now off while Snap to Pixel is on, for the Selection tool and
for paths Direct Selection moves by an anchor, so the move lands on whole
pixels as it does with the option off.
A path selected whole that Direct Selection moves held by one of its anchors snapped only through the anchor-drag targets, so Snap to Last Location never put it back. The move now goes back to where it began within Snapping Tolerance, labeled "last location", as a Selection tool move does; an Alt copy is not pulled back.
Preferences > Smart Guides gets Snap to Last Location (`snapToLastLocation`,
off by default). With it and Smart Guides on, a selection moved by dragging
with the Selection, Group Selection or Direct Selection tool, Shift too, goes
back to where the move began while the move is within Snapping Tolerance,
labeled "last location". An Alt copy is not pulled back onto its original.
Source: https://helpx.adobe.com/illustrator/desktop/measure-and-align/grids-and-guides/smart-guides-options.html
The distance, the label and the copy rule are VectorCraft's own.
Layout kept no record of a run's face, so each paragraph looked up the face
of its first character three times for its line metrics and once more to
shape it. The layout now keeps each run's face the first time it is looked
up and uses it for shaping, for paragraph metrics, for on-path type's empty
line and for the column offset of vertical point type. Glyphs, positions,
bounds and baselines do not change.
A test-only helper lays text out looking every face up at each use, and a
new test requires point, vertical point, area and path type laid out both
ways to match exactly.
The layouts of 5,000 short point type objects take a median of 215 million
CPU cycles of the laying-out thread instead of 236 million in a release
build (11 alternating runs of each build on Windows 11, measured while
other builds were running).
Turkish error rows translated parameter keys and values that the English
writes without backticks, so a message named a key or value the command
does not accept: `group must be true or false` read as "grup doğru veya
yanlış", `index must be ≥ 0` as "dizin", and `span` with "object" and
"selection" as "yayılma", "nesne" and "seçim". The rows for the docking,
option and setting errors (onto, below, above, anchor, before, panel,
zone, column, group, action, move, tool, kerning, balance, brightness,
decimals, lang, percent, plane, scale, control, span, index) now keep the
key and its values as the English has them. `included` and `others` lose
the double quotes the English does not have.
Bulgarian and Catalan rows that translated bare true and false now keep
them. Hindi messages whose double-quoted values were translated or
misspelled ("auto", "contents", "none", "start", "corner", "selection" and
others) now keep the values as the English spells them.
A new test checks that every translated status or error message keeps
each value its English puts in double quotes.
While the Type tool edits text, Cmd+Enter on macOS and Ctrl+Enter on
Windows and Linux typed a line break. It now ends editing as Escape
does: the typed text stays, the type stays selected, and the tool used
before editing is selected again. A held chord ends editing once; its
repeats type nothing and do not open the Move dialog.
Enter, Shift+Enter and Ctrl+Enter on macOS still type line breaks, and
so does the chord with Alt or Shift. Cmd+Ctrl+Enter on macOS ends
editing too. A command bound to the chord still runs instead, and
editing continues. While an IME composes, the chord stays the IME's.
The bundled craft-fonts Japanese, Georgian and Devanagari faces now get
the same baseline shift as installed fallback faces, so every glyph in
an interface label sits on the baseline of its family's first face.
Each fallback face's shift is measured against the first face of the
family it is added to (Inter for the proportional families, JetBrains
Mono for `mono`, egui's face for `Monospace`). Families whose first
faces need the same shift share one copy of the face; the others get a
copy named after their first face.
Tests: craft_fonts_sit_on_the_baseline_of_each_family checks each
bundled face in every family against its own baseline, and
fallback_faces_sit_on_the_ui_baseline now checks every family,
monospace ones included.
An installed face added for characters the interface fonts lack (Microsoft
YaHei for Simplified Chinese on Windows, for example) now gets a FontTweak
y_offset_factor computed from its ascent, descent and line gap against the
first face of the Proportional family, so its glyphs share that face's
baseline instead of centering the difference in row height. The bundled
faces and Latin text do not move.
This follows designcraft#394 (baseline_tweak from line metrics) and
effectcraft#1395. vectorcraft-text gains line_metrics, and a test face with
a CJK system face's vertical proportions checks the baseline at 13, 24 and
40 pt in the interface families.
An instance in perspective selected with a turned instance stays as it
was while the other is reset and returned in one undo step. Stored
natural sizes that are negative on either side give the art its own
size, unmirrored, as a zero size does.
The Symbols panel menu's Reset Transformation and the Control bar's Reset
(shown while symbol instances are selected) give the selected instances
their symbol's own scale again, without rotation, shear or reflection, as
one undo step (symbol.instance.resetTransform). Each instance keeps the
center of its art's bounding box in place. Instances in perspective and
instances already at their symbol's scale are left as they are, no undo
step is added when nothing changes, and both controls are enabled only
while a selected instance would change.
Docs: docs/mcp.md, docs/target-app-parity.md.
Three tests failed on a loaded machine because they asserted elapsed
time. Each now counts the work that the slow behavior it guards
against would repeat, with a counter compiled only into tests:
- The CSS cascade test counts selector tests, at most two per element
(testing every rule against every element tests N x N). It styles
5,000 elements, so the slow behavior fails on the count rather than
on the hang guard, which now waits 300 s.
- The stacked-glow test counts the silhouettes drawn, 30 for 30 glows
with and without the shadow cache.
- The Zstandard frames test counts the bytes looked at to find the
zero padding.
The hang guards for invalid dash patterns, text layout and script
checks of a pipe wait 300 s instead of 20 s.
The History panel gets a New Document from State button at its bottom,
which runs the new edit.newDocumentFromState command. The command makes
a new document holding the art of a History panel row ({index?}, the
current state when omitted) and makes it the active document. An index
that names no state is an error and changes nothing.
The new document takes the next Untitled-N name, is marked unsaved, has
nothing selected, no steps to undo or redo and no modified date, and
leaves symbol, opacity-mask and pattern editing; the art drawn in a
pattern's tile is stored in the pattern, as Done stores it. Its links point at the files the
source finds, including files found in the source's folder away from the
links' paths. Those files are looked for only within the folders the
asking caller may read; for a caller without limits, within the folders
the caller that opened the source may read. The source
document, its steps and its saved state stay as they are; typing or a
drag in progress is first kept as a step, as Undo keeps it. While an IME
composes, the command waits, as Undo does.
Docs: docs/mcp.md, docs/target-app-parity.md.
The mesh tests now count each feedback path's subpaths, so a mesh that
drew its boundary instead of its mesh lines fails them. The progress-doc
lines name the Pathfinder, mesh and symbol instance outlines and the
limit's count of evaluated objects.
With Real-time Drawing and Editing off, each pointer move of a drag ran one
boolean union of a group's members to draw its outline. A group of 256
overlapping ellipses took seconds per move, and at 1,024 overlapping ellipses
the union went over its budget and no outline showed.
A move or handle drag now draws the outline of each visible member of a group
or intertwine, and of each object a blend, envelope, repeat, Pathfinder effect
or object effect evaluates to, without uniting them. A gradient mesh draws its
mesh lines and a symbol instance its bounds, as their selection highlights do.
The limit of 100,000 counts outline points and evaluated objects, so a
selection past it shows its moved bounds.
A compound shape or Pathfinder effect still evaluates its boolean on each move,
and so does a clip group whose clipping path is a group.
appearance.addFill, appearance.addStroke and appearance.duplicateItem can
put a new topmost row on a series mark with the paint the mark already
shows. When another mark of the series had set a different stored paint,
the mark's top paint did not change, so the graph kept the other mark's
paint and drew it again. A new topmost visible row now stores its paint,
as an Eyedropper append already does.
The Appearance commands do not store a stroke paint or weight for a series
that draws a Filled Lines band, but they do store a stroke opacity set on the
band, and the band's 0.25 pt black outline takes it when the graph is drawn
again. docs/mcp.md now says so, and a test covers it.
The params text of the Appearance commands that update a series' stored
paint said the graph stores the paint and weight of the topmost visible
fill and stroke. Each of them also stores that row's opacity, as
docs/mcp.md says. The text now names the opacity too.
paint.setFill, paint.setStroke, paint.none, Apply Fill to Stroke and Apply
Stroke to Fill stored a series' paint on the graph. appearance.setItem, the
Eyedropper, appearance.transfer, Clear Appearance, Reduce to Basic Appearance
and the add, remove, move and duplicate row commands stored no paint or weight.
An Eyedropper pick of a blue fill at 40% onto a red column drew red at 40%
after the next redraw.
These commands and Show All Hidden Attributes now store the paint and weight
of the topmost visible fill and stroke of a path directly in a series group,
in the same undo step. A color, weight or opacity that a command writes to that
row replaces the stored value even when the mark's own value does not change. A
row that has no paint before and after the command does not change the stored
paint. These commands store no stroke for a series that draws Filled Lines
bands, and a band no longer takes its color from an empty stored stroke.
The stored opacity also comes from the topmost visible row: hiding a top row
stores the row under it, and an opacity set on a hidden top row is not stored.
Follows #5501.
Edit > Keyboard Shortcuts gains Export Text..., which saves the set shown
in the dialog, as edited so far, as a UTF-8 text file the user names in a
save dialog with a .txt filter. shortcuts.exportText {path?} does the same
for the current set over MCP and the control channel.
The file holds a "Keyboard Shortcuts" line and the set's name, then a
Tools and a Menu Commands section with the rows the dialog lists, in its
order and named as it shows them: the name, a tab and the shortcuts as the
dialog shows them on the current platform. A row with no shortcut ends
after the tab. A path that is not a string, or a file that cannot be
written, returns an error and nothing is written.
Docs: docs/mcp.md, docs/target-app-parity.md.
The queued and replayed wheel events of the font menu are read only by
native-only code (the viewport-layer plugin and the replay loop). Their
fields that only that code reads (frame, menu, position, event and
amount of QueuedWheel, and target of ReplayWheel) and the push that
fills the queue now carry the same cfg(not(target_arch = "wasm32")) as
their readers, so
`cargo check --target wasm32-unknown-unknown -p vectorcraft-ui-egui`
no longer warns that they are never read. Native builds do not change.
Choosing a PDF/X standard without a PDF/X preset writes a file with no bleed
(BleedBox = TrimBox) even when Document Setup sets one, and the export gave no
warning about it. The export now warns when a PDF/X file has no bleed on a side
where the document's bleed has one. The warning comes ahead of the flattening
note, because the status bar shows only the first warning. The code adds a
warning and does not change what is written.
These cases do not warn: the PDF/X presets, Use document bleed settings, a
nonzero bleed on every side the document sets, a document with no bleed, files
that aren't PDF/X files, and Export for Screens rows whose artboards Include
Bleed grows.
The warning has a row in every catalog. The Summary's warning and error rows in
the Save PDF and Print dialogs now go through the catalog, so they show its
translations.
Turkish rows translated backticked parameter keys and command ids, so a
message named a key the command does not accept: `stops` read as
"duraklar", `ids` as "kimlikler", `aspect` as "boyut", and `plugin.list`
as "eklenti.liste". Each of these rows now keeps the key in backticks as
the English has it. The rows that list accepted values ("corner",
"merge", first, previous, next, last, grayscale, websafe, true, false)
keep the values too, and the gradient stop rows read "stop" as a
gradient stop rather than the verb.
Seven Bulgarian rows quoted a placeholder with „“ where the English and
the rest of the catalog use backticks; they now use backticks.
A new test checks that every translated row in every catalog keeps each
backticked span of its English text.
The Shape Builder, Live Paint and run_around tests in planar/tests.rs
measured elapsed time against a 5 s limit, and the large edit tests
(large_edits.rs, with the Lock and Hide All Artwork Above and perspective
Alt-drag tests that share `soon`) against 20 s. On a loaded machine they
failed without any change in the code.
- pathops: a test-only counter adds up the items `RectIndex::at` reads
and the segments `nearest_u` searches. The four planar tests assert a
bound on that count.
- doc: `tree_walks` counts, per thread, the walks of a document's tree
that lookups by id make. `soon` fails when a command walks the tree
more than 64 times, and keeps a 300 s wait for a hang. The tests that
named a 20 s limit are renamed.
Looking objects up one at a time in Delete or Offset Path, or scanning
the whole index or ring in the planar code, makes these tests fail.
The only callers of `menus::menu_open` are the native palette-window
code in `panel_docking.rs` and its native tests, so on wasm32 the
re-export and the function were unused and `cargo xtask wasm` printed
two warnings for them. Both now carry `#[cfg(not(target_arch = "wasm32"))]`.
The menu bar still records whether a menu is open on every target.
The progress doc listed the moved-path stroke outline reuse twice, once
with the timings first measured and once with the timings measured
against current main. The line with the earlier timings is removed.
dotCooCoo