Node::strip_ruler_guides and Node::map_ruler_guide_positions asked a
layer whether its whole subtree held a ruler guide, then asked each
sublayer the same question again when filtering the children and again
when the recursive call started. A layer nested d deep was tested
2d + 1 times, so a chain of N layers cost about N^2 layer tests.
strip_ruler_guides runs when a document's objects are placed or pasted
(Clipboard::from_document), when a copy of the document is saved
(Document::guides_to_saved) and when a temporary editing layer is
removed (Document::remove_edit_layer). map_ruler_guide_positions runs
when an imported EPS is scaled.
A new held_guides reads the layer once and records which children are
ruler guides and which sublayers hold one; strip_held and map_held then
visit and copy only those sublayers. The guides removed, the order f is
called in, the art and its order, and the sharing of layers without a
guide are the same as before. Neither method records an undo step.
Deleting ruler guides and moving or scaling an artboard's guides asked
each layer whether its whole subtree held a matching guide, then asked
every sublayer the same question again on the way down. A layer nested
N deep was scanned N times, so a chain of N layers cost N(N+1)/2 layer
scans.
One read walk now records which children match and which sublayers hold
a match, and the edit visits and copies only those layers. The guides
removed, the count returned, the selection, the order the guides are
changed in and the sharing of untouched layers are the same as before.
The same walk replaces the repeated subtree scans in the other guide
edits in this file:
- tidying the ruler guides of an opened file (guides below art, in
objects, in opacity-mask art, in symbols and patterns)
- saving the guides as a list
- removing an editing layer while keeping its guides
When a read document is brought within the canvas, each saved guide was
checked against the removed artboard ids and the removed layer ids with
linear scans, and each guide node against the removed artboard ids, so
the work grew with the number of guides times the number of removed ids.
The ids are now collected into ordered sets, and each guide is checked
in logarithmic time.
Removing what lies beyond the canvas scanned each container's whole
subtree to see whether anything reached beyond, then scanned each child
again while removing, one level down at a time, so a deeply nested
object was measured once for every group around it. Removing guide nodes
probed each layer's subtree the same way before descending into its
sublayers. Both now make one read-only walk that records, by index,
which objects reach beyond (or which guides are removed), and the
removal follows that record. The objects clipped by a clipping path
beyond the canvas are still not measured.
Containers with nothing to remove are still not copied. The objects and
guides that remain, their order, the counts, the retired artboard ids
and the saved selections are unchanged.
Spreading two keys along a spine whose length overflows splits nothing, so
it returned the end anchors, and a closed spine was then opened at its start.
spread_anchors now rejects a length that is not finite before it makes any
target, so two keys get no anchors and the path stays as it was, as for
three or more keys.
A fill or stroke row that the Appearance commands gave a graph label, its
object opacity, and the characters the Eyedropper gave it lasted only
until the graph was drawn again: the graph stored nothing for a label's
rows, and the Eyedropper's character attributes were not stored.
- The graph keeps, for value axis numbers, category labels and legend
labels, the paint, weight, opacity and blend mode of a label's topmost
visible fill and stroke row, their order and place relative to the
Characters row, and the label's object opacity and blend mode
(labelStyles valuesRows, categoriesRows, legendRows). Every label of
the kind gets those rows when the graph is drawn again, with a placed
gradient fitted to each label.
- Every appearance.* command stores them, and so do graphicStyle.apply
and redefine and the paint, stroke, gradient and transparency edits
aimed at a row, even when the label already showed them.
transparency.set without a row and object.setProps store only the
object opacity and blend mode. With several labels of a kind, the last
one edited wins.
- appearance.copyFrom onto all of a label's text stores the character
attributes it applied and, from type, the paragraph attributes; a
transparency pick stores the object opacity and blend mode, and
append stores the rows.
- paint.setFill and paint.setStroke store a character fill or stroke
only when the characters take the paint (the whole label, or all of
its text highlighted with the Type tool); aimed at a row, they store
the row.
- Document Color Mode and swatch edits reach the kept rows.
Docs: docs/mcp.md, the commands' params text, docs/target-app-parity.md.
Giving the keys of a blend whose spine came from Replace Spine anchors of
their own flattened and measured the whole spine again for every key,
then walked its segments one by one. 1,000 keys on a spine of 4,096
segments flattened it 999 times.
The spine is now flattened once into one buffer. A split flattens only
its two halves and measures the segments after it again, in order, only
as far as the next key, so the arc lengths are summed as before and the
keys get the same anchors. A binary search finds each key's segment.
A spine so large that its length overflows no longer gets an anchor
split at a parameter that is not a number: spreading keys along it
returns nothing and leaves it as it was.
stops and stop flattened the path twice, once for the points and once
for its length, and stop did both before rejecting an index past the
last object. place validated the path with attachable_path first, and
transform measured the old and new paths up to three times each.
A private Measured value now holds the flattened first subpath with its
length and closed flag, and path_length, stops, stop and transform use
it. stop returns None for an out-of-range index without measuring, and
a move of the group measures no path.
attachable_path now uses path_length, so a path whose length overflows
to infinity is no longer offered as one objects can attach to when
placement cannot measure it.
In SVG and PDF output, each piece of a warped image whose pieces can show
through was drawn through a mask of every later piece that reaches it. On a
later piece's antialiased edge the viewer multiplies the earlier piece's
coverage by the mask and draws the later piece over it, so an image with
both opaque and partly transparent pixels came out lighter at its opaque
pixels along the triangles' edges (alpha 184 of 255 in the new SVG test).
A later piece now goes into an earlier piece's mask only when a pixel of
either piece's image within 2 pixels of their overlap is less than opaque
(vectorcraft_effects::WarpSeams). Where two pieces overlap over opaque
pixels, the later one is drawn over the earlier one without a mask.
Type with an inline graphic was resolved only in the layers, so type in
the art of another symbol, a graph design, a pattern tile or an opacity
mask kept the size its graphic had when it was made. The symbol holding
that type, its instances and the inline graphics that show it were
measured from the stale layout.
- Inline graphics are resolved in every tree that holds type. Symbols
are resolved in dependency order, so a symbol's type is laid out
before the type that shows that symbol inline. Edits, opening,
reverting and Refresh Font List use the same pass.
- The pass visits each object once: it resolves the children first and
copies a node only when something below it changed. It measures each
symbol's art once per pass.
- An inline graphic of a symbol whose art shows that symbol inline,
directly or through other symbols, is an empty slot. The symbols on
such a cycle are found once and cached until the symbols change.
- A changed inline graphic also clears the text's cached baselines.
Edit > Edit Colors > Convert to Indexed Color... (edit.colors.toIndexed)
stores the embedded images in the selection as indexed-color PNGs: a
palette of at most 256 RGB colors, each pixel the index of one. The
dialog previews live and OK is one undo step.
- The palette is chosen as for a PNG-8 export (selective, perceptual,
adaptive, web, gray, black and white), as evenly spaced levels of red,
green and blue (uniform), or from the document's swatches. Dithering
is diffusion, pattern, noise or none, with an amount.
- Pixels under half opacity become one transparent entry; the others
become opaque in their own color. Linked images, colorized images and
color glyph images keep their pixels.
- An image that is already an indexed PNG and would show the same pixels
is left as it is, and no undo step is recorded when no image changes.
- An image of more than 134,217,728 pixels (33,554,432 in the web app)
is refused, read from its header before decoding.
- With Select All, locked and hidden objects that hold an image that is
not an indexed PNG yet are counted in notChanged, without converting
them.
- The quantizer gains quantize_to, which maps pixels to a given palette.
Docs: docs/mcp.md, docs/target-app-parity.md.
An envelope around type of only spaces and a rectangle mapped from the
union of the rectangle and the type's laid-out box, so the empty type
moved and scaled how the rectangle was distorted. Type that draws
nothing now adds no box while other content in the envelope has one,
in groups and layers and as a clipping path too. When nothing else in
the envelope has a box, the type keeps its laid-out box, so an envelope
of only empty type keeps its frame, Make with Mesh and Reset with Mesh
still work, and Release still gives back the warp shape.
This applies to Make with Warp, Make with Mesh, Make with Top Object,
Reset with Mesh, Edit Contents and the stroke margins. Type that has
not been measured yet keeps its laid-out box as before.
The PDF/.ai import's boxes for envelopes with Distort Appearance
measure each object as the whole contents measure it, so the margins
the import tries match the contents' geometric box. The three import
tests that measure type without registering the glyph measure now
register it first, so another test registering it while they run no
longer changes their result.
An envelope or the Warp effect cuts an image into triangle pieces that
reach a little past their edges, so neighbors overlap and no seams show.
On an image with alpha, the overlap drew that alpha twice: a 50% image
came out about 75% opaque along every cell edge and diagonal.
- The pieces now form a knockout group, and each piece is marked as a
warp piece (`warp_piece`) whose knockout shape is its whole triangle,
so where pieces overlap only the top one draws.
- The canvas knocks each piece out by its triangle at full coverage.
- SVG and PDF draw the pieces of a partly transparent image (its pixels,
opacity, color mask or colorize paint) through one mask each, of the
later triangles that reach it; pixels on a triangle's antialiased edge
come out slightly lighter there. Pieces of an opaque image are drawn
one over another as before.
- SVG writes one copy of the image for pieces cut by opacity masks
(Preserve Shape Using Transparency) too.
- A cell count past the generated-object budget makes an empty group
instead of overflowing.
Docs: docs/target-app-parity.md.
Open and Place store some images again instead of keeping their bytes:
a photo turned by its orientation, a TIFF the image decoder reads, and
a CMYK TIFF with one plane per ink. Those images lost the ICC profile
their file embeds, and 16-bit samples were cut to 8 bits.
- raster_image keeps gray and RGB samples at 8 or 16 bits in the
stored PNG, with the gray or RGB profile in its iCCP chunk. A
profile of another color space, such as the CMYK profile of a CMYK
image read as RGB, is left out. A TIFF's profile is read with the
tiff decoder, which reads it whatever the image's size.
- stored_cmyk_tiff and turned_cmyk write CMYK8 or CMYK16 as the file
has it, with the CMYK profile as TIFF tag 34675 (UNDEFINED). A CMYK
JPEG's profile is put together from all its APP2 segments.
- The docs/target-app-parity.md line on EXIF orientation says 8 or 16
bits, and a line after it describes the kept profile.
Object > Path > Remove Anchor Points removed each selected anchor with
its own Vec::remove, shifting the rest of the subpath every time, so
removing k anchors from an n-anchor subpath took O(k*n) moves.
- pathops: remove_anchors removes a subpath's anchors the last first,
reading each anchor's current neighbors, handles, length and closed
state as removing them one at a time does, and compacts the anchor
list once. remove_anchor and remove_anchors share one refit, so the
result is bit for bit the same.
- engine: the command groups the selected anchors by subpath, works out
the next selected anchor with each subpath reading only its own
selected anchors, and takes emptied paths out of the selection
together (Selection::remove_all).
- Tests compare remove_anchors bit for bit with remove_anchor over 600
generated subpaths and focused cases, count the anchors moved (2001
anchors, every other one removed: at most one move per anchor left,
500,500 one at a time), and compare the next-anchor selection and
the command on many paths and subpaths with removing one at a time.
An <image> whose EXIF orientation or TIFF Orientation tag turns or flips
its pixels, linked or as a data URL, is turned before usvg reads it: the
pixels are stored as they are shown, the width and height swap, and the
box fits the turned image under preserveAspectRatio, as CSS
image-orientation: from-image describes. A turned CMYK JPEG or TIFF keeps
its inks as a CMYK TIFF, as does a CMYK TIFF that is not turned.
- Data URLs of type image/tiff, and untyped TIFF bytes, are read.
- TIFFs with extra channels, CCITT Group 3 TIFFs and planar CMYK TIFFs
are read as Open and Place read them.
- Converted images decode within the decode allowance, and a TIFF that
claims more strips or tiles than the decoder can count is refused
before any decoder opens it.
- exif_orientation moves to vectorcraft-doc so Open/Place and SVG import
share it.
Tests: crates/svg/tests/it/image_orientation.rs.
Spine::at_length walked the first subpath's segments from the start
until one ended at or past the arc length, so placing every blend step,
spread key, blend widget and Objects on Path attachment cost one
comparison per earlier segment. It now searches the anchors' arc
lengths, which never decrease, and picks the same segment: the first
whose end is at or past the length, so an exact anchor length and
repeated lengths from zero-length segments keep their segment. The
point and tangent evaluation is unchanged.
Tests compare the search with the old walk on open and closed spines,
one segment, curves, zero-length segments, every anchor, both ends,
out-of-range and non-finite lengths, and check that a 4,096-segment
spine needs at most 13 comparisons.
The previous commit took the spans from how far each piece bends away
from the split whenever both handles there were within ONE_CURVE of the
anchor. A curve that barely moves at the split but doesn't bend there
(a straight one, say) has short handles and no bend, so that gave 0/0
and the side fell back to the polygon shares. Now the spans come from
whichever pair is the longer: the handles, or the bends (which go with
the squares of the spans). The check of every piece against the
rebuilt curve is unchanged.
Test: a_split_curve_keeps_the_top_object_patch_where_it_barely_moves_and_runs_straight.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A blend made with Create Spine set to Never now shows the Appearance
widget with the Selection and Blend tools, so its color Shift can be
dragged on the canvas as on a blend with a spine. The Easing widget
stays hidden, since easing needs a spine.
The widget sits 16 px above the halfway point, by length, of the
straight lines between the key centers that the steps are laid out
along, and a drag follows those lines' direction there. The lines are
not drawn. `blend_layout_line` exposes that line; `blend_spine` still
returns `None` for such a blend.
Split where the curve stops (a cusp), both handles at the split lie on
the anchor, so their lengths say nothing of how the curve's parameter
divides between the two pieces, and the side fell back to the polygon
shares. When both handles lie within ONE_CURVE of the anchor, the
spans now come from how far each piece bends away from it (the second
derivatives, in proportion to the squares of the spans). The existing
check of every piece against the rebuilt curve is unchanged.
Test: a_split_curve_keeps_the_top_object_patch_when_split_at_a_cusp.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A live-effect plug-in that paints its pieces draws each piece's brushed
stroke along the piece's path at the piece's stroke width. Culling left
room for brush art from the object's own stroke width around the
pieces, so brush art of a moved or wider piece was cut off at the edge
of the view. Culling now leaves 16 x the piece's stroke width + 8 pt
past each brushed piece, and does so for a symbol instance whose art
holds such a path (measured once per symbol) and for a blend, envelope
or repeat that holds one. Object > Rasterize measures the brush art
drawn along each piece and leaves out the object's own brush art, which
is not drawn.
Groups and layers now leave room for their members' brush art. The
cull bounds of a group or layer united its members' bounds without the
room each brushed member leaves, so the painted bounds of a group, and
the area redrawn when it changes, left out brush art past its members'
strokes. That applied to plain brushed paths as well as painted pieces.
Each member is now measured as the renderer measures it when it draws
the group, in the members of a group with a fill, stroke or adjustment
of its own and in a group with crop marks or Vector Halftone too.
A top-object envelope gives each piece of a side a share of the content
in proportion to its control polygon's length. When anchors split a
curve without changing its shape, the pieces' polygons don't divide the
side where the curve's own parameter does, so the envelope bent its
contents differently although its outline was the same: Add Anchor
Points on a top object moved the art, and a top object whose .ai mesh
saves each side's curve split into thirds drew its type up to about
half a point off its drawn art, which follows the whole curve's
parameter.
Coons::from_top_object now checks whether a side's pieces, from corner
to corner, are all pieces of one cubic. The pieces' spans of that
curve's parameter come from the handle lengths either side of each
split, the curve from the outer ends and handles, and every piece must
lie within 0.02 pt of the curve's piece over its span (each control
point). Such a side runs along that curve, by its parameter. Any other
side keeps the polygon shares, including a side only partly made of one
curve and a smooth side drawn in many pieces that isn't one curve as a
whole. Corners found from the outline can still move when anchors are
added, as before. Gradient meshes made from a path are unchanged.
Existing top-object envelopes with such sides redraw.
a_split_curve_moves_the_top_object_share_to_its_new_anchor, which
pinned the earlier behaviour, becomes a_split_curve_keeps_the_top_object_patch.
Tests: a_split_curve_keeps_the_top_object_patch,
a_top_object_joins_a_split_curve_saved_with_rounded_coordinates,
a_top_object_keeps_smoothly_joined_curves_apart,
a_top_object_side_partly_of_one_curve_keeps_the_polygon_shares and
a_top_object_side_of_many_arcs_stays_on_them (vectorcraft-doc),
envelope_top_object_keeps_its_look_when_anchors_are_added
(vectorcraft-engine) and
a_top_object_whose_top_is_one_curve_split_in_thirds_opens_live
(vectorcraft-eps).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The kept grid now records whether a type measure was registered, so a
grid built before registration is not reused when the registered inputs
number happens to be 0.
BFlat