Readings of a shape are capped by those tried, not those kept, so a shape
whose every reading fails stops at the cap. The layout check grows a
window from each line start, doubling while the type draws the outlines,
and starts the next line at a mismatch, so a line split at every glyph
costs about twice its glyphs, not half their square; a glyph budget
leaves anything past it as outlines. A tracked line of 450 glyphs: 259 ms.
The lone-glyph rule looks at lines as the outlines lie, before the layout
check splits them, so tracked one-contour letters (L, I, T, H, E) become
type. A shape keeps every glyph and pose it fits exactly, upright first
(an L, and a turned L half round; an I upright or upside down), and the
line around it chooses; a one-glyph line may turn to another candidate
to take the next glyph.
The whole-outline check flattens the shape once per shape, samples every
corner and points twice the tolerance apart, and asks a flat uniform grid
of segments. With Microsoft YaHei loaded: 300 CJK outlines in 350 ms
(1.9 s before), one line in 25 ms. FontDb::outline reuses
outline_uncached, and the compositing test covers knockout_shape.
The whole-outline check tests points along every edge, so outlines with
the same corners joined another way do not match; past a work limit it
rejects. Face keys another Retype kept meanwhile are reused and charged
once. The fonts parameter test uses outlines that would match. The parity
note states the search limits.
Contours are read from a starting corner that does not depend on where
the path starts (every tied corner tried), with straight edges split in
two, degenerate curves and empty segments read as one line. A sampled
fit only shortlists a glyph: its whole outline, turned onto the shape,
must lie on it. Face keys are compact hashes, kept per process within a
budget; kept keys stay when new ones do not fit, and go when fonts are
added or rescanned.
A compound path with a locked or hidden part stays whole. A group with
an Attributes note, or any other extra, keeps its type inside. Faces are
counted once against the cap, and no font is read when every shape stays
an outline.
Hidden and locked shapes stay outlines and end the line. Isolated and
knockout groups keep their type inside. Type turned 180 degrees stays
one object (angles compared wrapped). A lone one-contour glyph beside
other shapes stays an outline unless it is the only shape selected.
The index samples only glyphs whose key (segments per contour) a selected
shape has, reads outlines past the shared outline cache, keeps each
face's keys once worked out, and loads no more faces than it takes. With
Microsoft YaHei (29,905 glyphs) loaded: 877 ms the first time, 23 ms
after.
Each retyped line is laid out and compared glyph by glyph with the
outlines; it splits where its type would draw another glyph or the same
glyph elsewhere (tracking, manual spacing, font substitutions), and a
glyph its type cannot draw stays an outline. Shapes with their own
opacity mask stay outlines. Shapes past the limits are reported as
truncated. Glyphs of plain-shape symbols (box drawing, blocks, geometric
shapes, private use) are not matched, so drawn rectangles stay art.
Type › Retype is in the Type menu.
An unmatched shape between glyphs ends the line, so the type after it is
still painted over it. Glyphs of different groups, paint or transparency
never share a line, and the new type takes the glyphs' own opacity and
blend mode. The type names the font version that matched. The glyph
budget holds inside a face too.
type.retype turns selected outlined text back into live point text when
each glyph shape is exactly a glyph of a loaded font, a font the document
uses or a family named in fonts: the same up to uniform scale, rotation
and translation. Matched glyphs map back to characters through the cmap
and become point text in that font and size at the same baseline and
turn, one object per line, with the first glyph's fill and stroke, in one
undo step. Shapes that match no font stay and are reported.
Matching (vectorcraft_text::retype) samples each contour at equal steps of
arc length and fits a similarity transform; it is bounded (faces, glyphs,
path elements, contours) and panic-free. Near matches, the ranked font
list, installed fonts not yet loaded and images are phase 2.
A Type tool click, or a Selection or Direct Selection double-click, on
type whose story is drawn in a font that isn't available as named (a
missing family, or a style its family lacks) now opens a question that
names those fonts instead of starting to edit. OK starts editing with
the caret where the type was clicked; Cancel leaves the document, its
undo history and the selection as they were. Each new edit asks again;
a click within the type being edited, or in another frame of its story,
does not. The fonts the type names are not changed.
The check reads only the clicked story's runs. As in text.missingFonts,
it leaves out CSS generic families and names over 256 bytes; it also
skips inline graphics and empty runs. The CSS generic family list moves
to vectorcraft-text so both share it. The question is the
missingFontEdit dialog, answered through the Type tool's
missingFontEdit option.
FontFace::win_cell_height adds the MVAR `hcla` and `hcld` deltas to
usWinAscent and usWinDescent at the face's location, so EMF and WMF
type with a positive font height is sized against the instance it is
drawn in. The variable test font gains the two records, and a test
checks that Bold's cell is 200 units taller than Regular's.
The comments and docs/mcp.md also say that characters the mapped face
has no glyph for are drawn from other faces at the same size and
Horizontal Scale.
Follows #7608. EMF and WMF type ignored the font's average character
width, and a positive font height was read as a fixed 0.85 of the cell.
Both changes are to how the importer reads a LOGFONT's size fields.
- When the face gives an average width (OS/2 xAvgCharWidth), a font's
average character width becomes Horizontal Scale: the face's average
width at the type size is stretched or narrowed to the font's width,
also when the size is held at its 5000 pt limit. A negative width is
read as its absolute value, and the width is ignored when the height
is 0.
- A positive height is the character cell. The em is the cell times
unitsPerEm divided by the face's usWinAscent plus usWinDescent (its
ascent plus descent when it has no OS/2 table). Type with a positive
height in existing files changes size.
- A font the system lacks is measured in the face VectorCraft draws it
in.
- FontFace gets win_cell_height and average_char_width.
LOGFONT: https://learn.microsoft.com/en-us/windows/win32/api/wingdi/ns-wingdi-logfontw
MS-EMF 2.2.13 LogFont: https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-emf/1cd7cb7a-1a7c-48d2-b4a4-c218f9d12100
The font menu's Recently Added chip, on its own row under Font Technology,
lists only the installed families with a font file created in the last 30
days, the newest at the top. It combines with the search, script, kind,
Variable, star and Document filters, lists the recently used families in
the same order without the Recent Fonts section, leaves Show Similar Fonts
to order only these families, steers the closed field's wheel, and is
shared by every font menu for the session.
The font scan keeps each file's creation time, and
FontDb::recently_added(now) dates a family by its newest file at or
before `now`, the 30th day included. Files without a creation time or
dated in the future do not count, and families added at the same instant
are in name order. The open menu lists a family no longer once its 30
days end.
First Baseline gains Em Box Height (`text.areaOptions {firstBaseline:
"emBoxHeight"}`, and the Area Type Options dropdown). The top of the
first line's ideographic em box, the largest on a line of several sizes,
touches the top of the type area inside the inset, in each row/column
cell and in vertical type. It is the em box that Top-to-Top leading
uses. Min is the least distance, as with the other kinds.
Native files save it as `emBoxHeight`. The .ai reader reads editing-data
kind 5 as Em Box Height and the writer writes it back; Legacy (6) and
unknown kinds keep today's fallback.
The dropdown's Ascent and x Height labels, and Fixed in Gujarati, get
the catalog rows nine languages were missing, beside the new label in
every catalog.
Help page: https://helpx.adobe.com/illustrator/desktop/design-with-text/add-manage-text/add-margins-to-text-and-adjust-the-first-baseline.html
A Russian or Bulgarian word written with a combining breve (`и` + U+0306
for `й`) failed the Cyrillic script check and stayed whole; the check now
reads the composed word, so it breaks at the same characters as `чайник`.
A stress mark, which composes with nothing, still keeps the word whole.
The all-capitals guard skipped a decomposed capital such as `O` + U+0308,
since a combining mark has no case, so `ÖĞRENCİ` spelled that way was
hyphenated in Turkish; the guard now counts only the letters.
The Glyphs panel asked the face for its characters on every frame, and
each call read the whole cmap, sorted it and removed repeated
characters. A face now keeps the list the first time it is asked for
(FontFace::mapped_chars) and shares it; FontFace::chars returns a copy
of it, in the same order. The panel finds a Unicode subset's characters
by searching the sorted list, and Entire Font uses the face's shared
list of unencoded glyphs instead of copying it each frame. The cells the
panel shows do not change.
Measured in one release process, old read and shared list alternating,
median of 7 samples of 100 calls, wall-clock times from a single run:
- Source Sans 3 Regular, 1,614 characters: 0.0101 ms to 0.000004 ms
- a test font of 50,000 characters: 0.2347 ms to 0.000004 ms
The saving matters for fonts with large character maps, such as CJK
fonts.
FontLikeness::distance left out every number either face does not give,
so a family whose faces give none of them (static faces with no OS/2 or
post table) was 0 from every family, and Show Similar Fonts could list
it right after the chosen family, as an exact match. Faces that share
no number to compare are now f32::INFINITY apart, so such a family
comes after every family of its script group that shares one, and
families equally far keep their order. The order of measured families
does not change. This is VectorCraft's own rule.
Builds on #7983, which added Show Similar Fonts.
Each paragraph of type on a path now sets its glyphs at its Desired word
spacing, letter spacing and glyph scaling, as point and area type do. The
spaced glyphs feed the steps along the path, alignment, the hangs and the
overflow, so text after a wider gap starts further along and a right or
centered line is aligned by its spaced length. Minimum and Maximum are not
used, vertical type is not glyph scaled, and a list marker keeps its width.
This changes existing documents: type on a path whose paragraphs carry
non-default Desired values now spaces differently and can overflow where
it did not.
SVG export writes such type as outlines with a warning, since a
<textPath> places its characters at their own advances.
Docs: docs/mcp.md, docs/type-parity.md, docs/gaps.md,
docs/target-app-parity.md.
A run whose language is German, Polish, Portuguese or Italian (any region)
now breaks at that language's hyph-utf8 patterns from craft-proof, and a
Bulgarian, Turkish or Russian run at that language's syllable rules. The
paragraph's Hyphenation options and the Preferences exceptions apply as
before, and points that would split a grapheme are dropped.
For Bulgarian, Turkish and Russian, craft-proof hyphenates a word outside
the language's script as English. For German, Polish, Portuguese and
Italian it can also return points from its own heuristics. Those points
are refused, so a word outside its language's script stays whole.
German, Polish, Portuguese and Italian need craft-proof built with
CRAFT_I18N_DIR; without it they keep today's behavior. Other languages,
English and the eleven Indian languages are unchanged.
optical_kerning_keeps_the_tsume_of_a_letter_with_coloured_harakat also
runs on an installed Arabic face. In macOS's Arial (5.01.2x) the initial
beh and its fatha fill their whole cell, so Tsume rightly takes nothing,
but the test required it to move something. The installed-face check now
measures each cluster's blank space first: with some, Tsume must move the
glyphs; with none, it must move nothing. The test font stays strict.
similar_families looked up each listed family's traits on its own, and
each look scanned every loaded face, so ordering the font menu took time
in the square of the families loaded. It now gathers the loaded faces'
traits in one pass and reads the scan's catalog once, and family_traits
uses the same helpers, so both give the same traits.