Review of #761 turned up three real defects and a feature that was visible
but not usable. All of it is fixed here, each defect with a regression test
that fails without the fix.
- A binding outlived the object it named. Deleting art left its entry in
`variables.bindings` forever: it counted as a skip in every dataset
application, `variable.list` counted it, and the saved file grew with every
delete. `Document::prune_variable_bindings` now lets them go after each
edit, as `prune_assets` does for Asset Export's.
- `Variables::prune` cleared `active_dataset` by comparing it against the
*variable* names being deleted. A dataset named like a variable lost its
active state when the variable went. The two are separate namespaces; the
block is gone (dataset.delete already did it right).
- `variable.delete` reported the number of names asked for, not the number
deleted, so deleting a name that was not there still said `deleted: 1`.
- `dataset.set` (new): a dataset's values could only be set when it was
created, so changing one meant deleting and rebuilding the row — losing its
place in the list and the active one. This replaces the row's values and
reports what it dropped.
- Window › Variables, and dialogs for the "…" commands. Every command had an
empty menu path and the command palette runs those with `{}`, so nothing
could be defined from the UI at all. The commands now carry their menu
path, the "…" ones open a Variables dialog (a name and a kind; a name and a
field per variable, its kind's editor), and `dataset.set` opens with the
row's own values.
- Locked art is neither bound nor rewritten by a dataset, and says so.
- `set_plain_text` told a node that existed but was not type that it was
missing.
- The panel had Document Info's icon, so the two were indistinguishable in
the collapsed column; it gets its own, plus New/Delete buttons, and its
rows say what their count is.
Format round trips now cover variables (the testkit fixture defines both
kinds, binds them and applies a row), and ROADMAP/README no longer list the
feature as missing.
cargo xtask ci: all 7 steps pass.
Document symbols placed inline in a run of type (beyond Illustrator; like
InDesign inline anchored objects): TextRun.inline holds the symbol, scale and
shift; the run is one U+FFFC laid out as a glyph sized to the art, centred on
the cap height. text.insertInline and Type > Insert Inline Symbol; editing
keeps the graphic whole; canvas, SVG (<use>) and PDF draw the art; Create
Outlines turns it into a symbol instance.
Each paragraph has its own alignment, indents, spacing, hyphenation, tab
stops and paragraph style (TextObject.paras; para stays paragraph 0's, so
old files and readers work). Layout, editing (split/merge), threads, the
Paragraph/Tabs/Paragraph Styles panels, ranged text.setStyle/setFormat/
tabs/paraStyle.apply, SVG export/import and the fuzzers follow.
Area Type Options gain a fit: none, autoHeight (Illustrator's Auto Size:
the frame's height follows the text after every edit, resolved in
refresh_bounds inside the edit so it shares the edit's undo step) and
shrinkText {minPercent} (beyond Illustrator: overflowing text scales its
sizes, leading and baseline shifts down at layout time by the largest
factor, in 0.1 pt steps of the first run, that fits). The Auto Size New
Area Type preference now applies. text.areaOptions and document.inspect
report overflow and the shrink factor; the Area Type Options dialog
offers fit and first baseline as dropdowns.
Area Type Options > Align: Top / Center / Bottom / Justify (AreaOptions.verticalAlign,
LayoutOptions.vertical_align). Each row/column cell is aligned on its own; rectangular
cells shift their lines, other frames and text wrap reflow until settled without
losing text. Justify shares the leftover space equally between line gaps. Area Type Options
shows Align and First Baseline as dropdowns (the parameter dialog's choices table).
Illustrator's Every-line Composer balances the breaks of the whole
paragraph for left/centre/right aligned text too, not only justified
text. VectorCraft only composed justified area text; ragged text was
always broken greedily.
- Model: ParaStyle.composer (vectorcraft_doc::Composer, re-exported by
vectorcraft-text), Every-line by default like Illustrator, saved only
when Single-line. layout() reads it from the object;
LayoutOptions.composer is now an Option override for tests.
- Composer: the justified gate is gone. Ragged lines never shrink (a
line that doesn't fit ends the node), badness is 100*r^3 with r = the
line's leftover space over a rag zone of 1/6 of the width (finite for
lines without spaces); the last line is free unless shorter than 1/3
of the width (measured in zones of 1/3), so a lone short word on the
last line pulls a word down. Hyphens cost 25 in ragged text (50 in
justified). Knuth-Plass fitness classes with adjacency demerits
(justified 10000, ragged 1000). Lines past the last width change
share one node key, active nodes are capped (4096) and paragraphs
over 200k glyphs break single-line, so work stays bounded; ragged
tries a tolerance of one rag zone, then any fitting breaks, then
falls back to greedy.
- text.setFormat {composer: "singleLine"|"everyLine"}; the Paragraph
panel menu's composer items work with a check mark (Reset Panel goes
back to Every-line).
- Tests: ragged Every-line moves "or" down in "Destroy all creatures
with toughness X or less." where greedy leaves "less." alone; the rag
of long paragraphs is no worse than greedy (sum of squared leftover
and of neighbour differences); Single-line is greedy; the 10k-char
layout budget covers ragged text; command test with undo; native
format round trip (rich_session's text is now Single-line).
No existing test needed changing: the justified tests and
hyphenated_layout_adds_hyphens (ragged, now Every-line) still pass.
Auto alignment (start of the paragraph direction) is ragged; a wrapped
line's Mojikumi flush opening bracket gives its half em back to the
composer, as in the line loop; tests cover Auto in both directions.
Layout of 10k chars, release: justified Every-line 5.6 ms, ragged
Every-line 5.8 ms, ragged Single-line 4.8 ms.
File Info's dates come from the clock, so documents made or saved a second
apart differed: journal_replay_reproduces_document and
rich_fixture_satisfies_invariants_and_undoes_to_blank failed at random.
One approach for every clock-derived date: it is a command input that
defaults to now, and the journal records the value used.
- file.new takes created? and document.save / file.saveAs take modified?
(Unix seconds, or null for none; default now), through date_param /
clock_date on the one save path (save_with), so Save, Save As and the
UI's save stamp alike; Save a Copy still leaves the dates.
- Their journal entries always record the date used, so replaying the
journal makes the same document whenever it runs. Actions recorded from
the journal (Session::journal_for_action) leave those dates out, so
playing an action later dates the document then.
- Test fixtures date their documents fixtures::CREATED, and the MCP
property test dates its local document as the headless one, so
documents built at different moments compare equal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fahri Meral