PDFW / #875: route document, booklet and book-merge exports through the
guarded shared PDF writer while retaining application export policy.
Validated the exact integration with all seven local CI gates, paired native
and WASM exports, and fresh FreeBSD and Windows ARM64 hosted checks.
A story that ends with a paragraph return has an empty last paragraph. When
it didn't fit, DesignCraft marked the story overset at the end of its text
(Preflight: "Overset text: 0 characters"; the out port turned red).
InDesign doesn't: Montilla Design's Flex Layout file has a caption ending
with a return, fitted to its three lines, and InDesign reports no errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Montilla Design's Flex Layout file names its body face "Regular " (with a
trailing space, as the designer's copy of Articulat CF did). Face lookup
already ignores spacing and case, so the text is drawn in Regular, but
font.list (Find Font) compared names exactly and reported the style
missing. It now uses the lookup's comparison (designcraft_fonts::same_style).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Flex Layout container is a frame holding its objects as nested items
(as Edit › Paste Into makes), with a `flex` record: direction, wrap,
justification, cross-axis alignment, Fixed/Auto width and height, padding
and horizontal/vertical spacing. designcraft_doc::flex lays the objects
out (a Flexbox subset: lines, justification, alignment, Auto sizing), and
the engine runs it after every edit, so the container keeps its objects
in place whatever changes.
Commands: flex.add (Add to Flex Layout; detects aligned objects: their
direction, Auto size, Space Between at their own spacing, as InDesign
does), flex.convert (a group or rectangle frame), flex.remove (objects
stay, grouped), flex.set (every panel setting, plus the alignment grid
cell), flex.reorder. Resizing a container by its handles or W/H changes
its box and makes that side Fixed instead of scaling its objects;
dragging an object inside one drops it in the nearest place.
UI: Object › Flex Layout (Add to, Convert to, Remove, Show Flex Controls),
a Flex Layout panel (Window › Panels) laid out as InDesign's (Create Flex
Layout; alignment grid, W/H with Auto/Fixed; justification menu, four
directions, wrap; padding with the link button; spacing, greyed as
InDesign does when the justification works it out or nothing wraps), and
the same section in Properties. Icons are new drawings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Guides in an IDML page (<Guide> children of <Page>) were dropped on import,
and DesignCraft's own guides weren't written. Montilla Design's Flex Layout
file keeps its 16 column guides on the B-Parent page; none came in.
Import: Location is measured from View Preferences › Ruler Origin (the
spread's top left, the page's, or the spine); Orientation, FitToPage (off:
a spread guide), Locked and ItemLayer are kept. Export writes each page's
guides under its <Page> in spread-origin coordinates, which is the ruler
origin DesignCraft writes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
InDesign writes LeadingModel="LeadingModelAkiBelow" on [No paragraph
style] of every document, Western ones included, but only its Japanese
composers use the leading model. DesignCraft applied it to every paragraph,
so a 24 pt paragraph after a 32 pt one sat 34.53 pt below it instead of the
Roman 32.8 (4 pt space after + 28.8 auto leading) in Montilla Design's
Flex Layout exercise file, and its fitted captions came out overset.
The import now drops that default when no part of the package applies a
Japanese composer (HL Composer J / HL Single J). Leading models set on
text, other styles, or in documents that use a Japanese composer are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
InDesign sets a text line when its baseline fits in the frame; the
descenders may hang below the frame's bottom. DesignCraft required the
baseline plus the descent to fit, so a frame sized to a heading's caps
oversets: Zero2Graphic's magazine tutorial draws 2.7682 cm frames for
103 pt headings (control bar, 4:12–5:44), and all three were overset.
The keep-lines IDML fixture's frames are now 236 pt tall, so they hold the
same 19 lines under either rule.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Tint row's value box was the egui Slider's own DragValue. The panel reads
the tint back from the document every frame, so egui dropped the text being
edited after the first digit (100 -> 6 -> 100 -> 1000, as #945 reports for
the CMYK fields), and a typed tint was never applied because the panel only
applied on the slider's drag_stopped().
The slider keeps applying on release, without its own value box; the value
is a widgets::NumField that buffers text while focused and applies on Enter
or focus loss, clamped to 0-100 %.
Part of #945.
The guard restored only the active document's content and selection, so a
command that edited through a nested successful command and then panicked
left the nested edit's undo entry (and no redo) behind, and a command that
opened or switched documents restored nothing at all. The guard now
snapshots every open document with its history and selection, the active
index and the journal, and restores them on a panic.
Regression tests: a nested edit followed by a panic leaves the history and
redo exactly as they were, and a command that opens a second document and
panics leaves only the original open.
find.find and find.next name change, first and attrs: the Find/Change
dialog sends one set of fields to all three commands, and main's
find-scope tests do the same. layout.pages.duplicateSpread's description
parses again (no comma inside the description of `to`).
A script whose later step failed kept the successful prefix in the document
but truncated the prefix's undo entries (and returned before the command's
aggregate undo entry was recorded), so the edits could not be undone at all.
A failed script now restores the document, selection and history from before
it started: all-or-nothing.
Regression test: frame.create followed by an unknown command is an error, the
document has no frame, nothing is left to undo, and the next edit undoes
normally. The one-undo-step success path still holds.
parse() expanded parameter entities with up to eight passes of whole-string
replacement, each pass cloning the text and re-inserting the entities'
original values. A pass count is not an output bound: a few hundred bytes of
nested entities inflate multiplicatively (a 449-byte DTD reaches ~4 MB here,
and larger chains would go to gigabytes), so a small file could exhaust
memory and CPU. Expansion now stops once it has added more than 1 MiB over
the input.
Regression test: a nested-entity DTD under 500 bytes is rejected with the
budget error, and a small inline entity still parses.
The command removed the section at a nonzero start with retain() and only then
looked it up to keep its properties, so the lookup always fell back to the
defaults: editing one field (a marker, a start number) silently reset the
style, prefix, include-prefix flag and start number. The existing section is
now read before the list is rebuilt.
Regression test: a Roman-numbered page-2 section with prefix A- and marker
First, then marker Second keeps startNumber, style, prefix and includePrefix.
Brandon Thomas