mountScenario now assigns the mounted renderer to
globalThis.playgroundAdapter and removes it on dispose, so the
wasm module is reachable from the browser console on every
playground page (for example playgroundAdapter.module._render_stats()).
The benchmark verify-pages diagnostic already used this name by
patching the served lib.js. The source is now the canonical
writer, so the wrapper's assignment is redundant.
AI-assisted-by: deepseek-flash
The playground called _init and _set_render_options with the DPR,
which sets the ratio but leaves the Skia target, backbuffer and
UI surfaces at CSS resolution. The canvas drawing buffer is
physical (CSS x DPR), so at DPR > 1 the smaller surface anchored
to the bottom-left of the framebuffer and left the top and right
edges unpainted.
Call _resize_viewbox after setting the DPR, the same sequence the
application uses in resize-canvas!, so the renderer surfaces
match the physical drawing buffer.
Cover DPR 1, 1.25 and 2 in the adapter test.
AI-assisted-by: deepseek-flash
Apply at most one _set_view + _render_from_cache per animation
frame in the wasm playground, accumulating the wheel zoom factor
and the last cursor point instead of rendering synchronously on
every wheel event.
This mirrors schedule-zoom! in
app.main.ui.workspace.viewport.actions so playground
interactions match the application. A comment records that the
playground should call the real function once it migrates to
ClojureScript.
Extend the adapter tests with the coalesced flush, factor reset
across frames, scale clamps, and per-move pointer pan.
AI-assisted-by: deepseek-flash
Compare every on/off combination of requested Cargo features over the
Boolean-lattice edges, in Gray-code build order, with a matrix index of
configs, edge comparisons and provenance.
Single runs record explicit feature sets as schemaVersion 2 with resolved
defaults and build environments. Edge comparisons stay diagnostic; edges
clean apart from the intended toggle are marked controlled.
Validate with 68 focused tests, CLI fault checks, dry runs, and container
smoke matrices for branch-b pan cases.
AI-assisted-by: muse-spark-1.3-contributor-free
Add a regression test that keeps 1000 shadowed shapes with two
drop shadows and one inner shadow, seven FFI arguments, and
visible hidden flags. Record isolated load success against pan
and zoom context loss on SwiftShader without changing workloads.
Keep 3 warm-ups, 10 attempts, and 30s timeout provisional with
measured per-attempt costs and workflow guidance.
AI-assisted-by: muse-spark-1.3
Record manifest-default features without adding A/B selection. Keep
scored builds free of the optional function-name profiling switch.
Exercise CLI persistence, interruption, browser failure and legacy page
interactions through separate diagnostics. Share readable reports and
warn when offline comparisons include failed runs.
Validate with 50 focused tests, browser and CLI fault checks, two full
corpus workflow runs, and headed rectangle cases. Shadow graphics and
sampling calibration remain open.
AI-assisted-by: GPT-6
