NetterSilverandClaude Opus 5.5
2f4a6d906a
PDF export rasterises every slide as a backdrop and a content layer and hands both to the writer. Profiling (gdb stack sampling) showed most of the main thread outside rendering going to per-pixel work on those rasters: compositing the backdrop onto white (not vectorised), the writer hashing every backdrop to deduplicate it, and the straight-alpha conversion scanning every content pixel. - Composite the backdrop with saturating adds and skip 16-pixel blocks that are already opaque; skip blocks with no partly transparent pixel in the straight-alpha conversion. Same formulas, same bytes. - Keep the last opaque backdrops (up to 32 MiB of pixels) and hand the writer the same image again when a slide renders an identical backdrop. The writer stored equal backdrops once already; now it also skips hashing and compressing the copies. Output is byte-identical. deckcraft-cli bench, export .pdf step, 15 interleaved runs each on a 16-thread Ryzen (A/A noise +0.9 %): - sample x4 (32 slides): median 531 -> 322 ms (-39 %), min 529 -> 317 ms - same deck, every slide a different background: 557 -> 441 ms (-21 %) - sample (8 slides): 149 -> 103 ms (-31 %) Exported PDFs are byte-identical. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DeckCraft PDF export
DeckCraft uses craft-pdf for writer construction, metadata, fallible font preparation and
final serialization. Slide rasterization, the selectable text overlay, font discovery and
shaping, hyperlinks, bookmarks, notes and handout layout remain app-owned. The existing
renderer also resolves and rasterizes WMF/EMF pictures before export.
Only the shared write feature is enabled, with default features disabled. Krilla retains
its raster-image and simple-text features, including notes/footer shaping. PDF embedding is
disabled: DeckCraft does not use it. The production exporter has no Hayro parser, interpreter,
image-decoder, fallback-font or CMap dependencies. Tests use hayro-syntax 0.7.2 directly,
with only std enabled, to inspect the generated PDFs.
The exporter runs inside a native unwind guard. Writer failures return PdfError::Write.
A font that cannot be prepared for embedding now produces an error instead of silently
omitting its selectable text or notes. The shared Krilla 0.8.2 patch avoids a second panic
in page/surface cleanup while unwinding and permanently poisons a document after cleanup
failure, so a retained writer cannot serialize it. Successful serialization paths are
unchanged. Native unwind guards cannot catch WASM aborts or process-wide allocation failures.
Building and validation
Rust 1.95 or newer is required. The workspace pins craft-pdf 0.1.0 and its Krilla 0.8.2
patch to https://github.com/storytold/craft-libs revision
39adbdf756569c8e03adb747d5e063f468ca98c1. Both packages resolve from that immutable Git source;
a sibling checkout is unnecessary. Publish and review this shared-library prerequisite
before publishing the app migration, so fresh CI and release builds can fetch it.
Run cargo test -p deckcraft-pdf for export contracts and cargo xtask ci for all workspace
gates. PDF contract tests also run with wasm-bindgen-test-runner on
wasm32-unknown-unknown. The pdf-deps gate verifies the two writer source pins,
retained Krilla image/text features and the development-only parser boundary.
This migration does not add vector slide artwork or change text-shaping/ToUnicode policy.