dtourolle
4bd8d86c00
ARCH §5.3 still described only a tile cache nobody built; it now says
what 0.19.0 tiles and what it does not: a linear DNG past PROXY_EDGE
opens on a reduced copy, a finer render samples a full-resolution
window, and the export is cut into halo-grown tiles. display-and-
extension.md's FR-DSP-2 row said absent, outstanding.md said the halo
had nothing to read it and that such a file fell to the embedded
preview.
rawler now builds from third_party, which ARCH's stack table and §3.2,
the root Cargo.toml's comment ("two upstream crates ... for Android")
and third_party/README.md's bump procedure did not know; the README
also names each vendored crate's licence. panorama.md and the manual
say a composite this wide develops and exports.
Patched upstream crates
Each directory here is a crate exactly as crates.io publishes it, at the
version Cargo.lock resolves, with a local patch on top. The root
Cargo.toml routes the dependency here through [patch.crates-io]; the
directory is excluded from the workspace, so cargo test --workspace,
clippy and fmt leave it alone.
The first commit that adds a directory is the pristine copy (from
~/.cargo/registry/src/*/<crate>-<version>, minus .cargo-ok and the
crate's own Cargo.lock). Every later commit touching it is ours, so
git log -p -- third_party/<dir> is the patch and nothing else.
Each directory keeps the crate's own licence files, and the crate keeps its licence: wgpu-hal is MIT or Apache-2.0, i-slint-renderer-skia is Slint's GPL-3.0-only or royalty-free or commercial choice, and rawler is LGPL-2.1.
Carrying a patch forward
When Slint, wgpu or rawler is bumped, the version here stops matching and cargo warns that the patch is unused — the build then silently goes back to the unpatched crate. So a bump is:
- Copy the new version in beside the old one, as its own commit.
- Re-apply the diff from
git log -pon the old directory. - Point
[patch.crates-io]at the new directory and delete the old one. - Re-check what the patch was for. The wgpu-hal and Skia patches are behaviour that only a rotated Android display exercises (below); the rawler one is a linear DNG wider than 16 700 pixels, such as a panorama the merge writes.
Drop a patch entirely once upstream has the fix; each section says what upstream change would make it unnecessary.
wgpu-hal 29.0.4 — Vulkan pre-rotation on Android
Upstream creates every Vulkan swapchain with preTransform = IDENTITY
(src/vulkan/swapchain/native.rs, gfx-rs/wgpu#3345). On Android, a window
whose orientation differs from the panel's is then rotated by the
compositor on the GPU (composition=CLIENT), and on this tablet in
portrait those frames tear.
The patch adds two methods to wgpu_hal::vulkan::Surface:
current_transform(&Adapter)— the surface'scurrentTransform.set_pre_transform(transform)— thepreTransformfor the next swapchain. Opt-in: nothing calls it but the Skia patch below, and the default is stillIDENTITY, so desktop and every other caller behave exactly as upstream.
Setting it is a promise that the caller draws rotated into a swapchain sized in the panel's orientation; wgpu itself rotates nothing.
Unnecessary once wgpu exposes pre-rotation itself (#3345).
i-slint-renderer-skia 1.17.1 — rotate the canvas to match
wgpu_29_surface.rs keeps the promise the wgpu-hal patch lets it make.
On Android it reads the surface's transform whenever it configures, sizes
the swapchain in the panel's orientation (width and height swapped for a
quarter turn), calls set_pre_transform, and concatenates the matching
rotation onto the Skia canvas before Slint draws — so the whole UI,
including an imported wgpu::Texture, is drawn pre-rotated. It checks the
transform before every frame too, because a half turn (landscape to
reverse landscape) changes it without resizing the window. Touch input is
untouched: only drawing is rotated.
itemrenderer.rs widens the pixel-alignment check from "pure translation"
to "any right-angle rotation or flip without scaling". Without that, a
rotated canvas silently loses pixel snapping everywhere.
Off Android every path is upstream's: the rotation is always None.
Unnecessary once Slint's Skia wgpu surface pre-rotates on its own —
worth offering upstream, since the linuxkms backend already renders
through the same rotate-and-translate in render_to_canvas.
rawler 0.7.2 — the allocation guard counts samples, not pixels
alloc_image_plain! and alloc_image_f32_plain! (src/pixarray.rs) panic
on a buffer over 500 M elements or 50 000 along either axis. The width they
are handed is width × samples per pixel. A linear DNG therefore hits the
guard at about 16 700 pixels wide, and the 22927 × 8966 panorama
Lightroom writes is refused as ">50000 px wide". The patch raises the
guard to 1.5 G samples and 200 000 per axis. It is still a guard against
a corrupt header, just no longer one that a real photograph trips.
The copy leaves out data/testdata, 13 MB of sample files that only the
crate's own tests read.
Unnecessary once upstream sizes the guard in pixels, or drops it.