alonso.torres 8a2c2b7246 🐛 Preserve resize overrides across rotated variant switches
A clean target can acquire tiny size drift while taking the old copy's
rotation. Exact main-size comparisons then discard a valid resize.
Compare geometry with its float tolerance, including fixed layout sizes
and composite bounds; keep nongeometry values exact.

The portable rotation/resize regression fails both height and selrect
checks before the fix. Cover ordinary and fixed layout sizing, and
check child placement from the original library shrink separately.
The full variant suite passes on JVM and JS: 56 tests / 337 assertions.

Benchmark lower-quantile ratios against the Phase 1e baseline:
main 10/100/1000 copies 0.52/0.60/0.60; three levels 0.51;
chain 0.52; library 0.54; many shapes 0.46; variant switch 0.65.
No slowdown flags.

AI-assisted-by: GPT-6
2026-10-28 15:53:09 +01:00
..
2026-10-28 15:53:06 +01:00
2026-09-29 09:07:34 +02:00
2026-03-11 15:24:40 +01:00