You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Performance
A bottom-up BMP (the common kind) read from a file is read from its end a block of rows at a time instead of whole: bmp -> png on a 61 MB photo holds 5.6 MB instead of 66 and runs 2.6% ahead of the crates (it had slipped 1.4% behind in 0.23.0). Converter inputs that are files can now seek (RewindableRead::seek_to).
The PNG writer's per-row calls are inlined again (0.23.0's shared FilteredZlib had lost it).
Fixed
Images to PDF: a page takes the image's physical size from the resolution it records (PNG, JPEG, TIFF, BMP), so a 300-dpi scan becomes a letter-sized page rather than one a point per pixel; images without one stay a point per pixel. Readers report resolution through RowSink::density.
Changed
The format map in DOCS/FORMATS.md (sublime paths --markdown) is readable at size: a graph shows only which formats convert directly, one plain line per pair. Under each graph a from-by-to grid gives every pair's fidelity as a shape (a filled circle lossless, a square half filled on the diagonal conditional, a hollow triangle lossy; DOCS/map/) and its number of steps.
DOCS/FORMATS.md opens with the map, lists formats in a table per category, and escapes angle brackets in notes (a note's <root> no longer disappears on GitHub).