v0.24.0

Parallel flex-row fragmentation, and the results of a systematic sweep for values the engine computed and never read.

Changed: these move existing documents

Six fixes make declarations take effect that were previously discarded, or apply a CSS rule the engine had wrong. If a document relies on any of them, its layout will change, and that is the point.

• max-width / min-width on a text block with no border, padding or background. A leaf text node honoured width and ignored the min/max family, so <p style="max-width: 315pt"> ran the full column. Eight of the thirty shipped template images move because of this alone: the Northmoor set's prose-measure classes (.intro, .measure78) were inert and now are not. Affects the JSX path equally, through <Text style={{ maxWidth }}>
• word-spacing is part of text measurement. It was applied only at PDF-write time through the Tw operator, so lines were broken as though it were zero and then drawn wider, so text could run past its container. Lines now break where a browser breaks them
• Print media queries evaluate against the page CONTENT box. They briefly evaluated against the page box, on an unverified claim that a browser agrees. Measured, Chrome matches none of Bootstrap's 768/992/1200 breakpoints when printing A4 or Letter, and neither does Forme now. A Bootstrap .container that took a desktop width and overran the page no longer does
• A stretched column paints its band on every page the row crosses. Under align-items: stretch, a column whose content ended before its neighbour's painted nothing on later pages; a browser continues the band
• A fixed-height box that does not fit fragments instead of vanishing. A breakable View with a fixed height that did not fit in the remaining space lost that height entirely: with a background it was drawn at content height, with none it was not drawn at all, and everything below it moved up by the difference. No page break, no warning. It now fragments across the boundary the way a browser does, measured: a 500pt box starting 600pt down a 733.9pt page paints 134pt on that page and 366pt on the next. This was corrupting one of the shipped demo templates, whose baseline had recorded a footer with no position at all
• An auto-width block fills, and its auto margins are zero. CSS 2.1 10.3.3: a block-level box in normal flow with width: auto fills its containing block, and any auto horizontal margin computes to zero; auto margins only center a box whose width is constrained, which max-width still does. Forme shrank these to fit and centered them. Bootstrap's .container is again the shape that suffers, for the same reason as the media-query fix above and with the opposite symptom: with its width absent when printing it is exactly an auto-width block, so it collapsed to its own padding, and its percentage-width children then measured against nothing. If you relied on an auto-width block centering, give it a max-width

Added

• Parallel flex-row fragmentation. A flex-direction: row that crosses a page now continues as parallel columns on each page, instead of laying its children out sequentially. This includes wrapped rows (flex-wrap: wrap), so the render defect that used to report the sequential outcome is gone along with the outcome. A row that crosses no longer relocates whole either: it starts in the space that is there, and the slivers the old relocation guard prevented are now prevented per column, by each column's own widow/orphan control
• word-spacing in the HTML CSS subset, block-level, with em/rem
• <html lang> reaches /Lang. The attribute was parsed and never read, so a PDF/UA file could carry a language contradicting its own content while the warning told the author to set the attribute they already had. lang precedence is now uniform on every render: an explicit option, then the document's own declaration, then "en" with a warning
• em and rem on gap, border-radius and border-width, which parsed correctly and did nothing. border: 0.5em solid painted at the medium default rather than the width asked for

Fixed

• Render warnings reach every surface. renderTemplateWithLayout() reported "warnings: none" on every render since it shipped, and forme dev served a complete warnings badge that nothing populated. Font-embedding warnings under pdfUa, missing glyphs, clamped table columns and sequential row splits were all silent on those paths
• A border no longer cancels the rest of an element's style. A paragraph with a border, padding or background lost roughly fifteen other properties, among them border-style, text-transform, letter-spacing, orphan and widow control, position and its offsets, and vertical-align
• Per-side border colours on the HTML path, which collapsed to one colour
• Inline elements keep their background, padding and border
• dir="rtl" as an attribute now reaches the engine's BiDi
• border-style with no explicit width paints, per CSS's medium initial
• border-collapse is honoured on display: table elements
• A table whose fractional columns sum to exactly 1.0 no longer reports itself as clamped. 0.6 + 0.4 of the available width left a float remainder of about 3e-14, negative at some widths, so half of those tables reported themselves over-full. The threshold is now a hundredth of a point

Internal

• The docs gallery freshness gate now verifies which engine build produced the images. It hashed the source and rendered with the binary without checking they corresponded, so a stale build satisfied it completely. 27 of 30 images were stale when that was finally measured
• Byte-wall fixtures for the box/text split, relative units, word-spacing and fragmented columns. Each was verified to fail before being trusted
• The SDK WASM rebuild is unconditional in RELEASE.md. Conditional on "if engine/ changed", it was a judgement call whose failure is silent: an SDK keeps a stale WASM and its byte-parity test still passes, against a JS package that has moved