putinterval, copy of a string, roll, counttomark, clear to mark and cvn
worked on up to 4 MiB or 100,000 objects for the one operation each is charged,
so a small file looped over them could keep the importer busy for hours within
the operation budget. Each now spends one operation per 256 bytes or objects.
Fixes#3584
- A composite (Type 0) font is noted by its descendant CIDFont's BaseFont
(ISO 32000-1 9.7.4.1), not its own BaseFont, which ends in "-" and the
CMap name (9.7.6.1, e.g. Name-Identity-H); without one, the CMap name is
taken off the end of the BaseFont.
- A font name's family is the longest part before a hyphen (or the comma of
Arial,Bold) that names an available family, so Noto-Sans-Bold is Noto
Sans, Bold and Noto-Sans-CJK-JP is Noto Sans CJK JP; a shorter family
still matches when no longer one is available.
- A missing font is split only at a comma or before the first part that is a
style word (Helvetica-BoldOblique); otherwise the whole name is its family
(MS-PGothic is reported as MS PGothic, not MS).
Synthetic tests (no host fonts): Families::of builds the family list.
# Conflicts:
# docs/file-format-parity.md
# docs/gaps.md
# docs/type-parity.md
Files::svg parsed nested SVG images with usvg without the check the top-level
import makes, so an overflowing order inside an <image> still reached usvg's multiply.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
usvg 0.48 casts the filter's order attribute through i32, so a value like
1066570220609 saturates and the u32 order_x * order_y multiply overflows.
Only debug and overflow-checks builds panic on that; release builds already
import the file — the multiply wraps, and VectorCraft drops the filter as
an unsupported primitive with a warning, so nothing reaches the document,
render or export (observed on main in a --release build: both repro files
import without error; with a one-entry kernelMatrix the wrapped product
2147483647² mod 2³² = 1 even passes usvg's length check and leaves a filter
claiming a 2147483647×2147483647 kernel, still dropped downstream).
The import now parses order the way usvg does (svgtypes numbers, the same
as-i32 cast with its saturation, the first side defaulting the second, both
positive or both back to the 3×3 default) and refuses exactly the one case:
when the product of the two sides overflows u32 (checked_mul). Any order
usvg can multiply keeps importing as before, however large on its own, and
a missing, unparseable or non-positive order still falls back to 3.
Regression tests cover the overflow repros, the release wrap case, large
but fitting orders (100, 80 80, 65535 65535, 2147483647 1), the
pass-through shapes, and sane kernels, in debug and in release.
Co-Authored-By: GLM-5.3-Flash <noreply@z.ai>
Highlight Alternate Glyphs no longer drops a lam-alef cluster's glyphs
outright: the ligature is clear when required shaping, or shaping with
liga left on (older fonts), gives it, so an optional feature such as
salt that changes it is highlighted again. is_lam_alef takes any
combining mark by Unicode general category (Arabic Extended-A/B/C
included) instead of a partial list. The test font maps the fatha to a
mark and skips marks in its ligature, so the mark-spanning test checks
the exact ligature glyph and cluster; it gains a salt lookup and test.
Also a one-line equivalent nonminimal_bool repair in pathops'
pathtools that strict Clippy needs on main.
Found while testing crates/engine/src/cmd/variablecmds/csv.rs:146 (csv_escape) and :158-163 (cell_to_value)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The PLOTSETTINGS part of each LAYOUT wrote the paper image origin as a point
(148 and 158), but the origin is two values at 148 (x) and 149 (y), and 158 is
not a PLOTSETTINGS code. Strict readers such as Autodesk Fusion can reject the
file. The ezdxf re-save the reporter found Fusion accepts differs from our export in
these objects, not in the geometry.
- 148/149 for the paper image origin
- shade plot mode, resolution level and DPI (76/77/78) from DXF 2004
- plot flag 1024 (model type) on the Model layout
- the plot style ACDBPLACEHOLDER without a subclass marker
Test r2000_and_later_files_have_the_structure_cad_apps_require checks the
required sections and table entries, handles below $HANDSEED, references
that resolve, and LAYOUT group codes.
# Conflicts:
# docs/file-format-parity.md
run_guarded put back only the active document's art, selection and interaction.
An undo step pushed before the panic stayed (Undo then stepped to a state equal
to the current one) and Redo was lost; other documents a command touched kept
their half-applied edits; the active document was restored by index.
It now snapshots every open document with its history and restores them, and
the active document by uid.
Fixes#3599
An envelope whose contents are being edited is written the other way
round from one that isn't: its contents, a group on `%_` lines, then
its dictionary, which holds the mesh that shapes them. Only a group of
meshes was left out before an envelope's dictionary, so those contents
were read as art and their undistorted shapes drew over the distorted
art that follows on plain lines (and that the page draws).
The group on `%_` lines before an envelope's dictionary is now left
out whatever it holds; text in it stays as hidden text, as the
dictionary's own source text does.
# Conflicts:
# docs/file-format-parity.md
Art inside a dictionary in the editing data, such as an area text's
frame (`/ConfiningPath`) or a mask's art, can set a transparency
(`Xy`). The writer goes on from it: the objects after an area text
whose frame sets one have no `Xy` of their own. The reader dropped
that state when the dictionary closed, so those objects took the
transparency before the dictionary instead; in a user's file a Lighten
glow before an area text made everything after it fade into the light
background.
The last transparency the art inside a dictionary sets is now the
current one after it (and the enclosing dictionary's). Dictionaries
outside the layers (the setup's styles) are left as they were.
# Conflicts:
# docs/file-format-parity.md
Area type made from the editing data's text document was laid out with
its first baseline the font's ascent below the frame's top. The text
document keeps where each of its lines starts, and the file's page draws
them there: for some fonts the first line sits well above the font's
ascent (Oswald's at about its cap height), so headings came in lower
than on the page and the layers were taken to differ from it.
When the file's first line is more than half a point from where it is
laid out here, the first baseline is now fixed where the file has it
(First Baseline: Fixed), measured in the text document's own frame so a
copied object's type moves with its frame.
# Conflicts:
# docs/file-format-parity.md
Type made from an Illustrator file's text document took each character
attribute its run doesn't set from the document's base attributes, not
from the character style the run names ([Normal Character Style] or
another of the file's). Where the two differ the type came in wrong: a
file whose base fill is white and whose Normal Character Style is black
came in with its text white, invisible, so the layers differed from the
page wherever the page's own type didn't stand in for it.
A run now takes its own attributes, then its character style's (and the
styles that one is based on, up to 16), then the document's.
Found while testing crates/ui-egui/src/panels/color.rs:199-205 (parse_hex)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/doc/src/units.rs:163-165 (readout)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/doc/src/setup.rs:111-113 (Quotes::opens / is_open), used by Quotes::apply at :122-131
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The tangent and perpendicular case is done for every tool the public pages name; its evidence stays in the revision row and in gaps.md and ui-parity.md.
With write root ...\out, a save to ...\out\x.svg:evil was accepted: it stays inside the root, but writes a hidden stream no folder listing shows. A ':' in any name after the drive is now refused with the device names. The check is a plain name test, so it is covered on every platform; the check-then-open race (item 1 of the report) is a documented limitation and is not touched.
library_file_name kept CON, NUL, COM1 and the like, which Windows treats as
devices, not files. It now adds a - after such a name (CON-.json).
Part of #3602
The canvas keeps a hairline visible by raising a stroke's weight to half a device pixel. A width profile multiplies that weight, so a 1 pt stroke with a 6x width point swelled to twice its size at 25% zoom and shrank back on zooming in. The minimum now goes to the profile's widest point instead; plain strokes are unchanged.
The grid options' public page lists a Preview option: the repeat on the canvas follows the fields while the dialog is open. Object > Repeat > Options had none, so a change showed only after OK. Now a Grid Repeat's dialog has the checkbox: Cancel puts the original back without a history step, OK keeps the previewed values as one undo step. Radial and Mirror Repeat Options are left as they are, since their public pages list no Preview.
Each patch of a gradient mesh was cut into 8 x 8 flat quads for SVG export,
Object > Expand and the other outputs that cannot shade a mesh, whatever its
colours, so a patch of one colour became 64 identical pieces. Neighbouring
quads are now joined into square blocks wherever the colour changes by no
more than about one 8-bit step across the block; the block's outline runs
through its quads' corners, and a patch that folds over itself is not joined.
A mesh read from a file can have any number of patches, and the view any zoom: a 150 x 150 mesh from a 3 MB file exported at 20x asked for about 23 million quads (2 GB), and the command's own cap of 50 rows and columns does not apply to a loaded one. One tessellation now makes at most 1,048,576 quads, the subdivision per patch giving way before that, and a mesh with more patches than that is malformed, like a point count that does not fit.
The Bosnian catalog still carried the Color (,) and Gradient (.) rows, which no longer match the toolbar's tips. It now translates Color (, or Shift+,) and Gradient (. or Shift+.) like the other catalogs.
The Color and Gradient tips now read "Color (, or Shift+,)" and
"Gradient (. or Shift+.)". The shortcuts follow the key positions, so on
layouts where Shift at those keys types another character the tips name
the keys to press. The 25 catalogs with these rows and the toolbar
accessibility test follow.
The key-position match for Shift+Comma and Shift+Period now applies only
when Shift is the only modifier. Cmd/Ctrl+Shift and Alt+Shift presses at
those keys keep their logical key, so the type-size shortcuts on German-like
layouts behave as before. The typing test also checks that no paint command
runs while text is edited.
When the conjugate a ∘ self ∘ a⁻¹ overflows, conjugated now forms every
entry of the product as a sum of terms whose factors are split into a
mantissa and a power of two, and scales the whole product by one power of
two from its w row at the end. Scaling each partial product to a largest
entry of 1 could round an entry to 0 that the next factor would restore,
so a projection with entries far apart in size, such as a scale by
(1e160, 1e-160) seen through the same scale, was removed and the art drew
flat.
When the conjugate of a projection overflows, it is now formed one factor
at a time, and each partial product is divided by its largest entry
magnitude. Scaling the affine factors up front let the w row underflow to
0, so a scale by 2 seen through a move by 1e308 returned None and the
projection was removed, though its conjugate is finite.
When conjugating a projection overflows, the scaled-down result is now divided by the largest entry in its w row. A valid conjugate is no longer dropped by the determinant cutoff at the arbitrary scale, and every point of a kept projection passes the horizon test in apply instead of none.
When moving type or a symbol instance in perspective makes the
conjugated projection overflow, Homography::conjugated now forms the
product from factors scaled to a largest entry of 1 and returns it at
that scale, which maps every point to the same place. When even the
scaled result is not invertible, the transform removes the projection
and the art is drawn flat, instead of keeping the old projection while
the flat art moves.
Homography::apply checked that w is in front of the horizon but not the quotients, so a point with w just above 1e-9 and a large x or y, or a NaN w, mapped to an infinite or NaN point. apply now returns None unless the mapped point is finite. inverse, conjugated and affine_at return None when their result has an entry that is not finite.
The move drag remembers the plane it places loose art on. Pressing 4
during the drag, or dragging on after it, previews and commits the move
on that plane instead of dropping it. A 1-3 key pressed before the drag
threshold also sets the plane the move uses.
perspective.move takes constrain: true instead of a plane-space
direction. For each object, the command snaps the move to the nearest
axis or 45-degree diagonal of that object's own plane, so a selection on
several planes keeps every object on a line of its plane. The tool sends
the raw pointer with the flag.
On a diagonal, Snap to Grid now moves along the diagonal by the nearer
per-axis correction, so that edge lands on its gridline.
perspective.move takes an optional constrain: [dx, dy], a plane-space
direction. For an in-plane move, Snap to Grid's correction is projected
onto that direction, so a move the Shift key constrains to a plane axis
or a 45-degree diagonal stays on that line when an object edge lies near
a gridline. The Perspective Selection tool sends the direction for
Shift moves and copies. Moves without the param behave as before.
Holding Shift while dragging objects with the Perspective Selection tool
snaps the move, in the plane's own coordinates, to the nearest plane axis
or 45 degree diagonal. This applies to plain moves and to Alt and
Ctrl+Alt copies. The preview, the final move and the measurement label
use the snapped point. Pressing or releasing Shift, pressing 5 and
pressing 1 to 4 during the drag recompute the preview from the pointer;
a move perpendicular to the plane is not constrained.
The targets are in paint order, so the first selected object is not always the first target. The active row is now matched against the first selected object wherever it is among them.
Reduce to Basic Appearance kept the topmost visible fill and stroke even
when another fill or stroke row was selected in the Appearance panel.
It now keeps the selected row in place of the topmost one of its kind,
and in each other selected object the row equal to it. With ids, or
with no selected row, it keeps the topmost ones as before.
With a fill row selected in the Appearance panel, Add New Fill now
inserts a copy of that row directly above it and selects the new fill.
Rows already above the selected one stay where they are. Before, the
new fill was always a copy of the topmost fill, added at the top.
The selected row belongs to the first selected object. In a mixed
selection, an object that shares that row gets the copy above its own
equal row; an object that does not share it gets a copy of its top
fill on top, as before. With a stroke row selected, or with `ids` or
`target: "contents"`, Add New Fill works as before.
The panel drops rows that were Cmd- or Shift-clicked together when the
active row changes outside the panel, so after Add New Fill (button,
menu or shortcut) only the new fill is selected.
transparency.makeOpacityMask with two or more layers below the mask
object (or a layer and objects of other layers) wrapped the layers
themselves in a new top-level group, so a layer ended up inside a
group. A layer among several masked items now stands for its art: the
masked group takes the layer's visible, unlocked objects, a hidden,
locked or template layer gives none, and the layers stay where they
are. One layer under a mask object, or one layer alone, is masked as
before.
Pattern tiles were drawn with the default render options, so an image
whose pixels cannot be decoded drew nothing inside a pattern fill even
when the render marks undecoded images, as the canvas does. Tiles now
follow the caller's mark_undecoded_images setting, and the tile cache
keys on it.
An opacity mask with such a pattern now treats the mark as part of the
mask, as it does for an undecodable image outside a pattern.
bflatastic