Skip to content

Cache fixed authored Grid atlas pages - #172

Draft
kostyafarber wants to merge 8 commits into
agent/slug-atlas-projection-reusefrom
agent/visible-first-grid-pages
Draft

Cache fixed authored Grid atlas pages#172
kostyafarber wants to merge 8 commits into
agent/slug-atlas-projection-reusefrom
agent/visible-first-grid-pages

Conversation

@kostyafarber

@kostyafarber kostyafarber commented Aug 2, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • render the authored Workspace Glyph Grid through deterministic fixed 256-root Slug pages
  • retain coherent Grid frames across source, axis and structural invalidation
  • keep Grid cell/row geometry fixed and shrink only oversized outlines
  • cache pages in a disposable Zstd CachedAtlas with integrity checks, atomic publication, carry-forward, retry-safe staging and a one-GiB LRU
  • stream cached/native page bytes with bounded backpressure and install page sets atomically
  • add real Electron GPU coverage for variable replacement, distant topology, scrolling and fixed-cell fitting

Product-boundary update

Source Han TTF is no longer this PR's representative authored-workspace acceptance fixture. Direct TTF/VF preview is now tracked by #143 and will bypass authored Shift conversion and SQLite entirely.

The pushed branch still implements visible-first/background page scheduling. The local working tree contains an uncommitted candidate that replaces visibleBuild/completeBuild with one complete atlasBuild, removes Visible readiness, and performs no atlas acquisition during scrolling.

Before that scheduler is committed:

  1. run release-native cold/warm complete residency on Hahmlet, Zen Kaku Gothic and M+ Kanji;
  2. separate native compile, packing, cache publication, transfer, GPU upload and first-frame phases;
  3. keep complete-before-present only if those representative authored fixtures pass the reviewed product gate;
  4. otherwise split scheduling from the proven fixed-page/cache/atomic foundation.

Do not use the flat 65,535-glyph Source Han TTF to force scheduling policy for 5–12k-glyph authored sources.

Verification completed on pushed/current work

  • Rust Slug/bridge tests and clippy
  • desktop tests
  • typecheck, lint and formatting
  • real macOS Electron GPU Grid suite
  • visual home suite with no intentional snapshot changes
  • manual Source Han cache measurement: 42.84 MiB artifact, about 516 ms median all-page cache load, about 1.067 s renderer-connect to complete warm frame

Current complete-scheduler working-tree validation additionally passed desktop, focused/full GPU Grid and visual tests locally. It remains uncommitted pending authored-fixture release measurement.

Stack

Draft stacked on #171. Retarget/rebase after #171 lands.

Out of scope:

@kostyafarber kostyafarber changed the title Prioritize visible Grid atlas pages Cache fixed authored Grid atlas pages Aug 2, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant