Why deferred / why not per-cell collapse
Grid-level collapse + min(span, columns) auto-capping cover the common cases. A per-cell collapse="lg" was considered and rejected: cells collapsing at different breakpoints produce ragged intermediate layouts (one tile full-width while neighbours are still multi-up → holes the auto-placer fills awkwardly). The coherent primitive is responsive spans — each cell knowing its exact width at each step — but that is a larger syntax feature with no concrete demand yet.
Sketch (finalize on demand)
- A per-cell responsive span syntax, e.g.
cols="4 lg:2 sm:full" (Tailwind-style col-span-4 lg:col-span-2), resolving to breakpoint-scoped grid-column spans; optionally rows too. - Stays coherent (no raggedness) because every cell declares its own width at each breakpoint, rather than an opaque collapse trigger.
- Composes with the grid-level
collapse (which still sets the final stack point).
Acceptance Criteria (when scoped)
- A per-cell responsive span syntax parses to breakpoint-scoped column (and optional row) spans.
- It composes with grid-level
collapse and the auto-cap, without raggedness. - Documented; tests cover a multi-breakpoint cell.
References
- WORK-348 (grid-level collapse + auto reduction), SPEC-085