perf: vmk port stack — VirtualList, quick wins y height-skip (22 commits) - #1
Open
LuisAlbertoMK wants to merge 23 commits into
Open
perf: vmk port stack — VirtualList, quick wins y height-skip (22 commits)#1LuisAlbertoMK wants to merge 23 commits into
LuisAlbertoMK wants to merge 23 commits into
Conversation
…concurrency unbounded - Internal plugins (10 auth plugins): for...of secuencial → Effect.forEach concurrency:unbounded. Seguro porque son auth plugins independientes sin shared state durante init. - Config hooks: for...of secuencial → Effect.forEach concurrency:unbounded. Cada hook recibe el mismo config, sin dependencias entre sí. - External plugins mantienen carga secuencial (determinismo de hooks). - Triple verify: typecheck 0 errors in file ✅ / static review PASS ✅ / build binary 130.5MB smoke test PASS ✅ (cherry picked from commit d2f49d8)
(cherry picked from commit 3b824da)
- tool_output global: 500 lines/10KB - truncateLimit added to V1+V2 config schema (Local + Remote) - MCP.toolTruncateLimit() exposed in mcp/index.ts - Used in session/tools.ts to override global maxBytes per server - Config: codebase-memory=8KB, engram=4KB - AGENTS.md: new MCP Token Budget Rules section - BACKLOG: anomalyco#2 Parallel Plugin Loading -> Skip (already parallel) - CYCLE: removed empty Cycle 6 - METRICAS/BITACORA/VMK-MANIFEST: session tracking (cherry picked from commit 8a144ec)
…essages
Create VirtualList component that reduces SolidJS/Yoga overhead by
rendering only items in the visible viewport + overscan buffer (3-5 items)
instead of all messages.
Key design decisions:
- Poll scroll position at 10fps (no scroll events in opentui)
- Use estimated heights (3 lines per item) — no reactive layout events
- Integrates into session component replacing <For each={messages()}>
- Maintains all existing rendering logic (revert, user/assistant message types)
Build and lint clean (pre-existing error in error-component.tsx only).
- 1 new file: packages/tui/src/component/virtual-list.tsx
(cherry picked from commit aef949d)
Add real height measurement of rendered items via polling refs. - Yoga-computed heights read in the same 100ms polling loop as scroll tracking - Cache Map<index, lines> accumulated over time, fallback to estimate (5 lines) - Range calculation uses cached heights for accurate padding/scroll positioning - Extra box wrapper per item captures total height including margins/borders Build and lint clean (pre-existing error-component.tsx only). (cherry picked from commit a0e463e)
…ast --help/--version paths Ported from vMK-dev commits 84c6784 + 533fa4a, adapted to current upstream entrypoint. Uses a cmdLoaders map of static import() specifiers so Bun bundles command modules at build time without evaluating their top-level code until invocation (compiled-binary safe). First-argument --help/--version take fast paths; subcommand flags keep the original scoped-help behavior.
…ng buffer - VirtualList: two-tier polling (100ms active → 500ms idle) reduces CPU 60-80% when not scrolling - registry.ts: generation-counter cache on materialize() — returns cached Materialization when tool registrations haven't changed, avoiding per-turn Map rebuild + permission filtering - compaction.ts: LRU cache 10→100 entries + WeakMap for fast identity lookup on repeated estimate() calls with same object reference - message-updater.ts: incremental index shift on eviction instead of O(n) rebuildIndex() — shifts Maps by evictCount, re-scans only tail - build.ts: fix post-build rename to handle .exe extension on Windows - script/index.ts: fix package.json path (packages/opencode → packages/cli) - web/package.json: fix workspace dependency (opencode → @opencode-ai/cli) Build verified: opencode-vMK.exe (122MB) created successfully. All 13 bash-parser tests pass. Typecheck clean. Lint 0 errors. (cherry picked from commit 7eca550)
No behavior change — the 4-pass O(n) scan moves verbatim into a pure, Solid-free function so it can be unit-tested and benchmarked in isolation. Prepares Ciclo 1 of experimento/mejora-autonoma-2026-09-03.
Ciclo 1 of experimento/mejora-autonoma-2026-09-03. 3 approaches evaluated with equivalence-verified benchmark (2500 random queries each): - A prefix-sum+binary search (WINNER): 82-91% faster at n=100/1000/5000 - B two-pointer: discarded — amortized pointer helps but O(n) padding remains - C chunked K=32: discarded — within-chunk walks + O(n) padding The prefix structure rebuilds only when the heights cache changes (memo), never per scroll tick. Baseline algorithm kept exported for regression comparison. Tests: exact equivalence + boundary pinning (7/7 green).
…M growth Ciclo 2 of experimento/mejora-autonoma-2026-09-03. itemRefs retained one stale entry (and its Zig-backed BoxRenderable) per ever-rendered item with no cleanup — growth proportional to scroll distance. 3 approaches evaluated: - A onCleanup guard (winner): releases on row disposal, el-identity check guards against re-mount ordering - B isDestroyed skip in tick (kept as defense-in-depth) - C WeakRef: discarded — the Map is the sole retainer, plain delete is simpler and deterministic Simulation metrics (2000 window shifts): retained entries 2005 -> 18 (-99.1%); tick measurement loop CPU -48.9% for long-scroll sessions.
|
This PR doesn't fully meet our contributing guidelines and PR template. What needs to be fixed:
Please edit this PR description to address the above within 2 hours, or it will be automatically closed. If you believe this was flagged incorrectly, please let a maintainer know. |
|
Hey! Your PR title Please update it to start with one of:
Where See CONTRIBUTING.md for details. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Base:
ef2792511d(fork/dev)Highlights
Evidencia
Nota