Answers the CodeRabbit review on #8653.
A glyph kept as an outline ends the open line, so the next mapped glyph
went through park_text's fallback arm into flush_text, which forgot the
lost glyphs in the middle of the fill pass. flush_text is now
finish_text plus forgetting them, and park_text calls finish_text:
starting a new line draws nothing else, so it is never a boundary.
Docs: the M4 roadmap row names this follow-up.
Since #8549 the glyphs of lines finished past MAX_PARKED were kept
until the next stroke and, past 4096 of them, matched every stroke.
A `2 Tr` word after a dense page of filled-only type then came in as a
filled object and a stroked one.
- flush_text forgets them: nothing is drawn between an operator's fill
pass and its stroke pass. end_text keeps them, since it also runs
mid fill pass for a glyph kept as an outline.
- LostFills keeps the first 4096 glyphs and matches only those; there
is no overflow that matches every stroke.
- replayed counts the lines its `left` check looks at toward
MAX_REPLAY_WORK.
- A file's corners are kept only when there are exactly four that the
image can take; a longer list is skipped without decoding it. Corners
the affine map misses stay a distortion, and a scale past the
perspective limit leaves the image at its nearest affine placement.
- Distort checks the corners in a frame local to the quad, so a missed
corner far from the origin is still refused.
- Relink and Unembed refuse a named or selected distorted image before
searching a folder or writing a file. Create Gradient Mesh lists the
distorted images it leaves in `skipped`. Place with Replace refuses
only when the measured image is distorted and the file is not text.
- SVG cuts a distorted image in a scaled symbol instance at the placed
size. EMF pieces store only the pixels inside their clip. PDF mesh
pieces of a CMYK image keep a same-corner mask as its soft mask, color
mask included. Both `.ai` writers write an image at the nesting limit
at its nearest placement with a warning. DXF warns only when it writes
the image.
- The canvas cuts the mesh by the view's largest stretch. Warped images
with a singular placement keep finite corners.
- docs/mcp.md and docs/plugins.md describe the tolerances, the DXF and
WMF fallback, the partial result of Create Gradient Mesh, and when a
plug-in sees `corners`.
Answers the fourth CodeRabbit review on #8549.
When a fill pass finishes lines beyond MAX_PARKED, their unstroked
glyphs are recorded (bounded at 4096). The first stroke after that
which could have gone over one of them ends the stroke pass, so a TJ
over more than 65 lines that comes back to where it began stays a
filled object and a stroked one for each part, as documented. Any
other first stroke drops the record, so type after an overflowing TJ,
filled only or not, is matched as before.
MAX_HELD's doc now says that one operator filling the same run of more
than 4096 glyphs twice at the same places is matched as well as it can
be: the reader tells no operator's start.
Answers the third CodeRabbit review on #8549.
- stroke_last no longer restyles the glyph drawn last once a boundary
(end_fills) closed it: a stroke after an ActualText span no longer
strokes its replacement text. A glyph stroked again where it was just
drawn still stays one object.
- Past MAX_HELD, the held strokes go over the last run of consecutive
fills they match with room for all of them, not the first, so a long
stroke pass no longer takes the same glyphs an earlier operator
filled at the same place.
- Docs: the limit is 64 kept-back lines plus the open one, so a TJ over
up to 65 lines stays filled and stroked.
Answers the second CodeRabbit review on #8549.
- A stroke that goes over no filled glyph (stroked type of its own) and
the start and end of an ActualText span now end the stroke pass
(Builder::end_fills): the held strokes are matched, the parked lines
finished and the open line's fills closed, so a later stroke can't go
over an earlier fill and a later span keeps its own replacement text.
- ActualText fills are ordinary entries in the line's fill list, and
span strokes are held and matched in replay order like any other, so
an A/B/A TJ in a span strokes the replacement. The actual_fills map,
stroke_parked and has_actual_fill are gone.
- Past MAX_HELD, the held strokes are matched to the first run of
glyphs filled one right after another that they go over
(Builder::replayed), with bounded work, instead of each to its first
match, which could reach an earlier operator's fill.
- Docs state that past 64 lines in one operator each part stays a
filled object and a stroked one.
A TJ can draw the same glyph at the same place twice (its kerning goes
back), so a stroke can't be matched to a filled glyph as it arrives:
[(A) -3500 (B) 4834 (A)] TJ put the first A's stroke on the last A and
left the earlier parts unstroked. Strokes over filled glyphs are now
held until their stroke pass ends and then matched to the glyphs filled
last, in the order they were filled, so each copy is stroked where it
was filled, while a glyph filled and stroked where one filled before
stands still strokes only the newest. At most 4096 strokes are held.
Docs: ordinary text gives one filled and stroked object per part; in an
ActualText span it stays one replacement object on the first glyph's
line and style.
The reader fills every glyph of a text-showing operator, then strokes
them. When a TJ's kerning started a new line part way through, the
first line was finished before the stroke pass came back to it, so its
strokes found no filled glyph and came in as stroke-only type over
fill-only type.
A line a glyph doesn't fit is now parked while a stroke may still be
drawn over it, and a stroke-only glyph is matched against the open line
and then the parked lines, newest first. Parked lines are finished
oldest first before the open line, so drawing order is kept, and at
most 64 are parked.
The writer passed no language to craft-pdf, so exported and printed
PDFs had no /Lang and accessibility checkers failed primary-language.
Writer::new now passes the BCP 47 tag of the Document Setup language
(new vectorcraft_doc::setup::language_tag, en-US by default) through
write::Writer::new_with_language, which writes the catalog's /Lang and
the XMP dc:language. A language with no tag writes none. /MarkInfo is
left for tagged PDF (#3643).
Patch::apply wrote each cross-reference entry by walking every edit, so a
pass with many edits took time in proportion to objects times edits. It
now keeps the running total of the length changes and finds, by binary
search over the edits' ends, how far each object moved. For every edit
set that apply still accepts, the written file is byte for byte the same.
An offset that no longer fits the ten digits of a cross-reference entry
(ISO 32000-1, 7.5.4) is now refused with an error instead of writing an
eleven-digit entry that breaks the table. A change in length too large
for an isize is refused with an error instead of wrapping.
A PostScript plate has no blending modes, so an overprinting gradient
whose stops mix zero and nonzero ink of a plate was drawn whole on it,
and its paper-white part knocked out the ink below.
On a plate, the EPS writer now clips an overprinting gradient to where
it puts the plate's ink down. Linear, radial and diamond gradients leave
out the stretches between paper-white stops and the padding past a
paper-white end stop. Conic gradients leave out their paper-white
wedges, and freeform gradients leave out their paper-white cells (masked
at Level 3, clipped at Level 2). Fills, strokes (within, along and
across) and characters are covered, on process and spot plates. Where
the gradient puts ink down it is drawn as before, and a gradient with
ink everywhere is written unchanged.
On PostScript plates, a stop or point of an overprinting gradient whose
color counts as black, with up to half a percent of another ink, now
puts none of that other ink down, as a solid fill of that color already
does. PDF plates are unchanged.
A file whose catalog has a CMYK /GTS_PDFX output intent describes its
DeviceCMYK colors with that intent's profile. A DeviceCMYK image in
such a file now opens tagged with that profile, so a CMYK image that a
PDF/X export writes as DeviceCMYK opens with its profile again.
An image drawn where a /DefaultCMYK is in scope keeps no profile, since
the default replaces DeviceCMYK there. The scope is read per image from
the resources it is drawn with, and the image cache keeps the tagged and
untagged reads apart.
The scan that looks for d0 glyph procedures ran every time a page or a
form listed a Type 3 font in a font dictionary of its own, and its result
was dropped once the font had been noted. It now runs when the font is
first noted for the document. Each page still maps the font's glyphs,
and the note the import keeps for the font is unchanged.
On PostScript separations, up to half a percent of another ink in a color
that counts as 100% black is left off that ink's plate, so an overprinting
near black does not knock out the ink below it. A color linked to a swatch
the document does not have was left out of that rule, although it separates
from its own values like an unlinked color and Overprint Black marks it to
overprint. A 100% K with 0.4% cyan linked to such a name then stayed on the
cyan plate and knocked out the cyan under it.
A link to a missing swatch now counts as unlinked for this rule, on solid
fills, strokes, character fills and gradient stops. Spot swatches and
Registration are still excluded, and PDF plates still draw the faint ink.
Perspective Distort and Free Distort (`object.distort`) used to give an image
only the warp's affine approximation at its center, so its corners did not
follow the handles and it drifted away from paths warped with it.
- An image's four corners now go exactly where the warp takes them, alone or
in a mixed selection. A parallelogram stays in the image's transform; any
other shape is kept in the new `image_quad` field of the native file, which
older files do not have and still open as before.
- Corners that would fold the image over, fall on one line, cross the warp's
horizon or lie too close together for a perspective map are an error, and
nothing changes. A file whose corners break these rules opens with the image
placed by its transform. A drag is one undo step, and a distort that leaves
the corners where they are records none.
- Bounds, hit testing, transforms, Transform effects, blends, the Eyedropper,
Outline view, the Color Theme Picker and Weed Box and Lines follow the
corners. Image info measures the resolution at the image's middle, and
plug-ins get the corners.
- The canvas, printing and every exporter draw a distorted image as a mesh of
affine image pieces through the shared mesh warp, clipped by the corners.
In files and print the mesh aims to stay within 0.05 pt with cells of at
least 2 pt; the canvas cuts it for the zoom, within a quarter of a pixel
with cells of at least 4 pixels. Pieces reach into their neighbors (0.5 pt,
or 2 pixels on the canvas), so the image stays opaque at every zoom. The
mesh has at most 64 by 64 cells and no more cells across than the image
has pixels on its longer side (at least 8). No warped copy of the pixels is
stored. DXF and WMF write the nearest parallelogram and warn.
- Crop Image, Image Trace, Relink, Embed, Unembed, Create Object Mosaic,
Create Gradient Mesh and Place with Replace refuse a distorted image with an
error that names the distortion. Update Links puts a changed file's pixels
between the same corners, without applying Placement Options. Mask follows
the corners. Rasterize makes a plain image of it.
- The new error messages have rows in every complete-message catalog.
Docs: docs/mcp.md, docs/plugins.md, docs/target-app-parity.md.
A PDF/X-3 or PDF/X-4 export tagged every CMYK color with an ICC-based
space of the same profile that the output intent embeds. In PDF/X-3 and PDF/X-4, CMYK colors in the
profile of a CMYK output intent are now DeviceCMYK, and the profile is
written once, as the output intent's DestOutputProfile (ISO 32000-1,
14.11.5:
https://opensource.adobe.com/dc-acrobat-sdk-docs/pdfstandards/PDF32000_2008.pdf#page=641).
- With output.profiles none and a CMYK output intent, PDF/X-3 and PDF/X-4
write every CMYK color as DeviceCMYK, unless colors are converted to a
CMYK destination of another profile, which keeps its tag. Before, those
standards tagged every color whatever profiles said.
- RGB and gray colors stay tagged, and so do CMYK colors and images in
another profile.
- A CMYK image that carries the output intent's profile is DeviceCMYK,
and so is every CMYK image when none makes CMYK colors DeviceCMYK. A
CMYK JPEG then loses its ICC_PROFILE segments.
- Transparency groups blend in DeviceCMYK when CMYK colors are
DeviceCMYK, also when the file has no ICC-based CMYK space.
- Files without a standard keep the tags that profiles asks for.
- The Save PDF dialog's note for PDF/X-3 and PDF/X-4 says that CMYK
colors in the output intent's profile are written as DeviceCMYK.
Docs: docs/mcp.md, docs/target-app-parity.md and the document.exportPdf
params text.
Look::takes let a glyph filled only join a run whose glyphs were also
stroked, so after mode 2 text a mode 0 glyph took the run's stroke:
`2 Tr (B) Tj 0 Tr (C) Tj` came in as "BC" stroked throughout. A glyph
now joins only a run drawn alike (fill, stroke, overprints), so a glyph
filled only starts a run of its own and takes a stroke only when the
matched-stroke path (#8039) strokes it.
flush_stroke no longer strokes a whole run in place: a whole run goes
through the split and join code, so fill and stroke rendering glyph by
glyph still builds one stroked run.
A PDF/X-4 export wrote xmp:ModifyDate and xmp:CreateDate in its XMP
metadata but no xmp:MetadataDate. The PDF/X-4 metadata edit now adds
xmp:MetadataDate next to xmp:ModifyDate, with the same value, and grows
the stream's /Length to match. When the export dates the file as modified
later, post::modification_date already rewrites every xmp:MetadataDate,
so the two dates stay equal.
PDF/X-1a, PDF/X-3, PDF/A and plain PDF exports are unchanged.
A fill and a stroke at the same opacity below 1 were joined into one
type object whose opacity then applies to both together, losing the
darker overlap the file composites. stroke_filled now takes a stroke
only when the fill and the stroke are opaque.
An upright glyph filled and stroked was kept as an outline for the
reshaped-outline fallback while upright glyphs filled only were not;
it is no longer kept either, so such a line can't come in as the
outlines of its stroked glyphs alone.
Docs: the #8015 entries say what is left (translucent type, and a TJ
that starts a new line part way through).
- New tests: small opaque RGB and gray images are stored as RGB PNGs and an
image with a soft mask as RGBA; an image with one row more than the
pixel limit's square is left out with a warning.
- Gray images stay RGB on import, and a comment says why: a gray PNG
would make the image one that a fill color colorizes.
- Doc comments and the docs line give the limits in GiB, the line moves
next to the line about refused PDF parts, and it no longer gives a
timing.
A PDF, or a `.ai` read through its PDF part, was refused when one image
would decode to more than 1 GB, and the shared reader skipped images of
more than 2^28 pixels. On a 64-bit desktop build both limits are now 4 GB
and 2^30 pixels (`image_limits`), which `MAX_IMAGE_DECODED` and
`open_options()` share; 32-bit and web builds keep 1 GB and 2^28 pixels.
An image past the limit is refused with the same message.
An opaque image is encoded as an RGB PNG from its decoded samples instead
of building an alpha channel and an RGBA copy first.
Docs: docs/target-app-parity.md.
Part of #4366.
A PostScript plate drew an overprinting fill, stroke or character as
nothing when it put up to half a percent of that plate's ink down, so a
faint ink was lost from the plate. Now only a paint with no ink on the
plate is drawn as nothing, on PostScript and PDF plates alike. The one
exception is a color that counts as 100% black and is not linked to a
spot swatch or Registration: on a PostScript plate, up to half a percent
of another ink in it still counts as none, so a near black set to
overprint keeps the ink below it.