Layout and Markdown diagnostics

Compile the current working document before export. Lipat measures text and semantic layout in the Rust compiler and returns diagnostics beside the scene. The same diagnostics cross native and WASM interfaces; no DOM measurements or PowerPoint re-import are needed.

const compiled = deck.compile({ fonts });
if (!compiled.ok) return showError(compiled.error);
showDiagnostics(compiled.value.diagnostics);

// Host policy: request a revision before exporting content known to overflow.
const overflows = compiled.value.diagnostics.filter(
  (diagnostic) =>
    diagnostic.bounds &&
    Object.values(diagnostic.bounds.overflow).some((n) => n > 0.01),
);
if (overflows.length) return requestRevision(overflows);

const exported = engine.writePptx(compiled.value.scene);
if (!exported.ok) return showError(exported.error);
showDiagnostics([...compiled.value.diagnostics, ...exported.value.diagnostics]);
savePptx(exported.value.bytes);

showDiagnostics, requestRevision and persistence belong to the host. Warnings do not automatically block export or alter content. The low-level PPTX writer accepts a resolved scene, so its diagnostics cover export compatibility; retain the compiler's diagnostics as well. Agent lipat_export already combines both sets. The playground now retains both sets, including when exporting HTML.

Structured measurements

textOverflow, layoutOverflow and outsideSlide include slideId, nodeId, bounds and recommendations. Existing code, severity and message remain available. Other compatibility/font diagnostics may have no bounds.

• textOverflow: measured wrapped-text height or a glyph/line width exceeds the text frame. Covers text, headings, code and lists, including frames entirely inside the slide.
• layoutOverflow: stack/grid flow exceeds its container height. Children retain their natural size; they are not silently shrunk.
• outsideSlide: the element's axis-aligned bounds, including rotation around its center, extend beyond the slide.

bounds.kind is text, layout or slide. measured and available use points (72 per inch). Text/layout measurements are in the compiler's layout coordinate space, before element rotation; slide measurements are axis-aligned slide coordinates. overflow gives nonnegative left, right, top and bottom excess. Text rectangles describe the required layout space, not exact glyph ink outlines.

For example, three text lines at 24 points with a 1.2 line-height multiplier require 86.4 points. In a 30-point-high frame, bounds.measured.height is 86.4, bounds.available.height is 30 and bounds.overflow.bottom is 56.4. The stable recommendations are resizeNode, reduceFontSize, splitContent, moveNode and addSlideBreaks; these are suggestions for the host/agent, not executable operations.

Measurements follow the compiler's supplied font metrics. Without them, fontMetricsFallback explicitly warns that deterministic approximations are in use. These diagnostics do not promise pixel-exact clipping detection across fonts/renderers, complex shaping, per-run font-size changes or advanced table layout. No warning is not a guarantee that arbitrary PowerPoint content fits.

Markdown import feedback

fromMarkdown keeps its existing lightweight LipatResult<DeckInput> contract. Use importMarkdown when the import flow needs measured feedback:

import { importMarkdown, createDeck } from "@byfungsi/lipat";

const imported = importMarkdown(source, {
  id: "presentation",
  engine,
  compileOptions: { fonts },
});
if (!imported.ok) return showError(imported.error);

const { document, summary, diagnostics } = imported.value;
showImportSummary(summary.slideCount, summary.boundaries);
showDiagnostics(diagnostics);
const created = createDeck(document, { engine });

The result includes the validated working document, compiled scene, diagnostics and summary. Boundary records contain the stable slideId, 1-based startLine (inclusive), endLine (exclusive), and separatorBefore: document start, or a horizontal/vertical separator with its source line. Separators are excluded from a slide's range. Empty slides have startLine === endLine. CRLF/CR line endings are normalized for line numbering. The summary describes the imported source, not subsequent edits.

Boundaries come from the actual Markdown parser. Headings remain content; standalone --- and -- lines retain their documented meaning, and separator-looking lines inside fenced code remain code. The absence of --- is not a parser error.

When import produces exactly one slide and the compiler reports overflow, the report adds markdownSingleSlideOverflow, with IDs, the relevant bounds and addSlideBreaks/splitContent recommendations. Short, fitting one-slide documents do not receive this warning. This is based on measured overflow rather than a guessed word-count threshold. Import does not automatically split, shorten or reject that content.

Reproduction and host integration

A deterministic working-document reproduction is # Report, a blank line, and 16 newline-separated Content line lines, without slide separators. At default size/theme with Arial metrics {averageWidth: 0.5, lineHeight: 1.2}, it produces one slide. The stack requires 542.4 points inside 444 available points and reports 98.4 points of bottom overflow before export.

This reproduces a diagnostics/reporting gap; the original Supernala working document was unavailable, so it does not establish the exact cause of that incident. Supernala should retain the pre-export working document/revision, show import boundaries and compiler warnings, and decide whether overflow requires revision before publishing an Artifact. Re-importing an already exported file cannot recover the original authoring layout intent. Agent instruction-following and host export policy remain separate concerns.