- presentation.json gets its own 256 MiB inflation limit instead of the 2 GiB
per-entry allowance, and is parsed in place: only a file that needs its
null-number repair is copied.
- A media member that fails decoding or its CRC is charged for the bytes it
inflated, so later members cannot reuse the shared media budget.
- repair() and Presentation::validate() check slide layouts against a set
built once instead of scanning every layout per slide.
The shaper lets one call grow to 256 glyphs per input character, as an
embedded font's substitutions decide, and every glyph was then copied out.
Items longer than 1 KiB are now shaped in pieces (cut at a space where one
is near, never before a combining mark or after a joiner), each with the
rest of the paragraph as context and the item's script, and a piece that
expands past 32 glyphs per character falls back to one glyph per character
from the cmap.
* Animations: index morph matching and timeline lookups by shape
Morph pairing searched every shape of the next slide for each shape of the
current one, four times over, rebuilding shape text inside each comparison.
Timeline state looked up a shape's effects, paragraph owner and base colour
by scanning every effect, group and colour on the slide, once per shape and
paragraph drawn. Both are now indexed once, keeping the same match order.
* Animations: drop wall-clock bounds from the large morph and paragraph tests
On a loaded runner a ten-second bound can fail after every functional
assertion passed. The inputs stay large enough that a quadratic regression
hangs the test instead.
Found while testing crates/engine/src/cmd/file.rs:489 (save, with save_bytes at :341-349 and format_for_path at :440)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The round trip now also checks the written presProps.xml: inside p:showPr, the extension uri and
p14:showMediaCtrls with its namespace declared on the element; the default writes nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review of #1693 found that cast_effects built the picture content that a shadow,
a glow and a reflection are drawn from without converting a tiled fill to
points, so inside a scaled group the reflection of a tiled picture showed
child-space tiles while its face showed the right ones.
The content now goes through paint::tiled_picture_per_point with per_point_xy,
as the face does.
Test: a tiled picture's reflection in groups scaled 0.25, 1600 and 2x0.5 shows
the same colours as in an unscaled group; it fails without the change.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Review of #1693 found two more places that drew a tiled picture fill in the
shape's own (child-space) units inside a scaled group.
A group nested in a scaled group stored its own tiled fill unconverted, so the
members that take it (grpFill) got oversized tiles. The owning group now
converts the fill with per_point_xy when it stores it, so inheriting members,
directly or through a chain of grpFill groups, get the converted tile.
A picture shape (p:pic) whose blip fill is a tile passed its own fill straight
to paint::picture. It now goes through paint::tiled_picture_per_point, which
tiled_per_point also uses.
Tests: a group's tile passed to its members and a picture shape's own tile, each
in outer groups scaled 1, 0.25, 1600 and 2x0.5; both fail without the change.
Docs: the Picture and texture fill checklist row is re-assessed (remains P) with
its status line and revision history, and the ROADMAP progress log row mentions
both cases.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Brandon Thomas