Pin craft-ui to c4b8405d, which adds a top edge and the canvas-side top
corner to icon-strip flyouts and saves the moved top per group. Accept
its new MoveFlyoutTop docking action, and restore the saved top when a
group returns to its place. A docking regression checks the cursors, an
upward top-edge and free-corner drag with the bottom and icon-column edge
fixed, and the top after reopening and in a saved layout.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex <codex@openai.com>
Each 3D facet is stroked with a narrow seam of its own paint (0.05 to
0.75 pt; Revolve's is always 0.75 pt) so neighboring facets meet without
an antialiasing hairline. In Extrude & Bevel, Inflate and Rotate under a
white light, and in every face of the 3D (Classic) effects (Extrude &
Bevel, Rotate and Revolve (Classic); shaded, No Shading, gradient and
pattern alike), that seam kept the default Miter Join with its limit
of 10. Revolve's seams and the colored-light seams
were already round. A facet seen almost edge-on is a sharp sliver, and
its mitered seam reached up to ten times its half-width past the facet's
tip: thorns of up to 3.75 pt past the solid, along the tops of type seen
tilted and at every edge-on bevel facet.
#2706 (multiple lights) deliberately kept white-light seam joins as they
were, so that legacy white-light output stayed byte-identical to main
(its comment "Preserve existing white-light paints, alpha composition and
seam joins exactly", and four PNGs compared byte for byte). That kept
the thorns too. This change reverses that one part on purpose: the
paints and alpha composition stay exactly as #2706 left them; only the
join changes.
- Outputs change, the document model doesn't. Exports draw the new
joins (SVG writes stroke-linejoin="round", PDF and .ai draw round
joins), and Expand Appearance writes round seams into the document.
But nothing a document stores selects the join, and the format is
unchanged, so there is no version gate: an old file drawn now simply
has no thorns.
- Round joins. Projected facets rarely have right-angle corners (a
square seen at an angle is a parallelogram with acute corners, and an
exact 90° corner sits at a √2 limit), and round is what the seams of
colored lights and Revolve already draw. The cost: where a facet has
no neighbor (Rotate's plane, every silhouette corner) its convex
corners are drawn rounded by up to 0.375 pt instead of sharp.
Alternatives weighed and not taken: a Miter Join with a limit near √2
at every seam, the composed clip included, which would bevel slivers
but chamfer most projected corners as well; and seaming only the edges
facets share, or clipping facets to their union as solid_surface
does, which would keep every outer corner sharp but means a new
geometry pass for every 3D effect. Round is the smallest change that
matches the seams already drawn.
- A seam no longer reaches past its facet by more than half its width,
so the art's visual bounds shrink by up to 3.375 pt where a thorn
stood (10 × 0.375 less the 0.375 a round seam still reaches).
White-light output is no longer byte-identical to #2706's comparison;
docs/3d-lighting-performance.md says so.
Every seam stroke is made by lighting::seam_stroke. The colored clip in
composed_art already outlined its seam with round joins and is
unchanged. Tests: every seam of Extrude & Bevel, Extrude & Bevel (Classic, each
branch), Rotate (Classic), Inflate and Rotate, under white and colored
light, has round joins and reaches no more than half its width past its
face, at views that make a sliver; Revolve (Classic)'s seams are round
on each surface; a mitered seam fails them, and so does the legacy
white-light Inflate test, which now expects Round.
The Brushes rows name each type as the New Brush dialog does (Art Brush,
Scatter Brush) and say "missing from the Brushes panel" for a brush the
panel does not hold. The panel now shows these values through the
catalogs, so a localized interface no longer shows them in English.
document.info and the saved report keep them in English. Every catalog
gets a row for "missing from the Brushes panel"; the type names already
had rows.
Remove Background refused a grayscale image colorized with a process
color by looking at the pixels the image shows. With a color mask whose
masked pixels are not gray, as an opened file can hold, the check let
the image through, and the gray pixels it cleared took an alpha channel
under the process color. The check now reads the image's own pixels,
which are the ones cleared.
docs/mcp.md and the Color Mask line in docs/target-app-parity.md now say
that a mask that hides no pixels leaves a grayscale image free to take a
process color.
View > Proof Setup > Working CMYK simulates the working CMYK space of
Color Settings in Proof Colors and Separations Preview, also in a
document assigned another CMYK profile. The active context keeps the
working context beside the assigned profiles, and the proof table cache
key includes the working settings, so a change of the working CMYK space
shows in the next proof.
A weeding line at v0 + k*step can round just past the edge of an outline
segment that lies on it: across 0.1 to 0.5 the one line sits at
0.30000000000000004, and (0.30000000000000004 - 0.1) / 0.2 is
1.0000000000000002. The candidate range for that segment then came out
empty, so the line was not stopped there and ran over the art's edge.
The candidate range is now widened by a small tolerance before rounding;
`near` still leaves out lines the segment does not meet.
docs/mcp.md and the params text of `addon.cutting.weed` now state the
`spacing` range (more than 0, at most 100000 pt), the 1000-line limit
each way, and that a request giving both `count` and `spacing` is
refused.
A spot-colored fill or stroke whose opacity is 0 prints no ink, but
Spot Color Objects still listed its object and objects.spots counted
it. Such fills and strokes are now left out; an object stays listed
while another of its fills or strokes prints the spot color. Pattern
Objects and Gradient Objects are unchanged.
With automation read roots in force, document.json returned each entry of
svg_scripts unchanged, so a JavaScript file linked from outside the read
roots kept its full path in the reply. The reply copy now names such a
file by its file name alone. An entry given as a URL is kept, and the
stored document is not changed.
The Pencil no longer snaps to Smart Guides or Snap to Point and shows no
Smart Guides while hovering or drawing: a stroke's points, its ends
included, and both ends of its straight lines are where the pen went.
Snap to Grid and Snap to Pixel are separate and still move its ends
(#2958 is about the grid). Closing, continuing a selected path and Edit
Selected Paths are unchanged; the Knife's straight cuts still snap.
- object.expand.info reports no gradient or fill to expand for a
gradient that repeats in more runs across its box than are written,
matching object.expand, which leaves it a gradient. For a path that a
painting live effect draws, it measures each piece the effect returns.
- CSS writes a repeating gradient's stop positions, and a repeating
radial gradient's radius, with enough decimals that their rounding
adds up to at most a hundredth of a period across the box (at least
four for positions and three for the radius).
- docs/mcp.md says that sharp steps and swatch-linked stops are the last
to be left out rather than always kept, and that the CSS conic limit
counts stops before midpoints and Perceptual curves add more.
Color Mask checked the whole document for a grayscale image colorized
with a process color only when the document passed that check before
the run. With another such image already in the document, the check
was skipped, and a process-colorized gray image could take masked
pixels with transparency.
Each image whose mask changes is now checked on the pixels the new mask
shows, made from the image's own pixels, so an image elsewhere in the
document no longer affects the result. A mask that hides nothing keeps
the image's own pixels and is still set. When new pixels replace a
masked gray image's own pixels under a process color, the mask made for
them hides nothing, and Apply Color Mask refuses them. The check shares
its colorize predicate with paint's check_colorized.
A stroke at an opacity of 0 draws nothing, so its brush is no longer
listed in the Brushes category, brushNames or brushes. Pattern Objects,
Gradient Objects and Spot Color Objects likewise leave out fills and
strokes at an opacity of 0, and Objects no longer counts objects whose
only such paint is at opacity 0.
A reply hid a hidden library's absolute path wherever it appeared, also
inside another loaded library's id when the match began after the id's
start. With `/libs/Kit.gpl` hidden, the readable id
`loaded//project/libs/Kit.gpl` came back as `loaded//projectKit.gpl`,
and a request that used it failed.
Each whole readable id in a reply (with or without `loaded/`, or in a
`file:` URL with no host or `localhost`) is now found first, and a
hidden id or path that lies inside it stays. The same path after a
drive or in a URL with another host is still hidden.
jmverges