Document pictures now decode through wordcraft_geom::picture (behind a
new `image` feature of the geom crate). It reads the header, refuses the
picture before anything is allocated when the decode's peak would pass
the budget, then decodes with the header's size and the budget as
limits.
The peak is the larger of three moments: the decoder's output beside
what it holds while decoding, the output beside its RGBA8 conversion,
and that beside a premultiplied copy. A progressive JPEG's decoder
(zune-jpeg) holds an i16 for every coefficient of every component, in
whole MCUs, for the whole picture; the SOF header gives that size. A
7900 x 7900 progressive JPEG without subsampling needs 9 bytes a pixel
(561 MB), past the 512 MB budget, although two RGBA buffers (499 MB)
fit. With the fix reverted, the test's header-only file of that size
was decoded and the test process peaked at 542 MiB resident.
- Render and PDF export ask the probe in their wrappers (decode_pixmap,
load_image), so a picture past the budget gets the grey box or is
left out, as before.
- The crop dialog and every picture edit decoded with load_from_memory
and looked at the size afterwards. They decode through the probe now,
and so do wrap outlines and the RTF re-encode.
- A picture edit has the drawing budget instead of an 80 M pixel cap
checked after decoding. A picture past that budget (about 67 M RGBA
pixels) can't be drawn either, so no edit that could be seen is lost.
Windows metafiles (WMF, EMF), common for logos, crests and diagrams in
older documents, couldn't be decoded and showed as grey boxes, on screen
and in PDF.
The new crate wordcraft-metafile (layer 1) reads them from the public
[MS-WMF] and [MS-EMF] specifications into paths and bitmaps, and places
them in a picture's box; render and pdf paint the result as vectors.
It covers the drawing records Office pictures use (lines, polygons,
curves, rectangles, ellipses, arcs, path brackets), pens and brushes,
the mapping modes and world transforms, and bitmaps in DIB form with the
usual raster operations. Input is treated as hostile: every read is
bounded, and a byte budget caps what one picture may allocate.
Not yet: text records (ExtTextOut) and EMF+ records are skipped, so a
metafile whose content is mostly text shows only its lines and fills.
docxide-pdf bench (365 documents, vs main + #135): the twelve documents
with WMF/EMF pictures show them; the pictures are a small part of their
pages, so the scores move little (Jaccard passes 208 -> 209, five
documents up to 3 SSIM points better, none worse).
Co-authored-by: Oliver Zieschang <4963183+naxvog@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Clean-room foundations for WordCraft: Arc-shared document model with
paragraph/run editing, styles, lists, sections and tables; Word-style
font substitution; first-fit line breaking with tabs, lists, alignment,
pagination, tables, headers/footers and hit testing; vello_cpu page
rasteriser.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
VoiceLessQ