perf(web): reuse completed Markdown prefixes while streaming

[juliusmarminge]

Streaming a response reparses completed code fences on every chunk. Preserve the parsed prefix through a closed top-level fence and reparse the changing suffix, following the completed-block reuse in Legend Markdown.

The existing whole-document transforms, sanitization, source positions and rendering remain in use. Reference definitions, footnotes, CR/CRLF and BOM input fall back to a full parse. Plain prose, completed messages and custom syntax plugins use the existing path. The cache belongs to one streaming renderer.

Five alternating production-browser runs at 1440 × 1000, each with 20 warm-up updates and 120 measured updates:

Client update time	Base, including minimap change	This layer
Median of run medians	21.5 ms	17.4 ms
Median of run p95s	34.0 ms	30.3 ms

The fixture contains 12 completed TypeScript fences of 30 lines plus growing prose. Every paired run improved; the median reduction is 19.1%. Both builds use the same browser process, origin, fixture and probe. These are synchronous client-update measurements, not provider throughput or user telemetry. Raw samples · Probe, comparison script and parity evidence.

82 focused tests and web typecheck pass. The external CommonMark corpus adds 17,426 full-tree comparisons, including source positions, across its 652 examples. Browser verification retained selected text, DOM identity and exactly the same scroll position through 40 updates. Targeted lint reports two warnings in unchanged code.

This affects streaming Markdown in web and Electron's shared web renderer. Mobile and file/diff viewers are unchanged. Local and remote clients use the same rendering path; no transport, provider or contract changes.

Before, desktop:

After, desktop:

Before navigation and streaming, narrow viewport:

https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cf9292628b42deda/parser-before-demo.webm

After navigation and streaming, narrow viewport:

https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/45f147cf61401e0c/parser-after-demo.webm

Recordings show visual states, not wall-clock speed: the recorder compresses dropped-frame intervals. The table uses separate, unrecorded timing runs; 0.8 seconds of recorder startup was trimmed. Only synthetic fixture data is included. Probes and evidence are not committed.

Integration follow-up: rebased on current main; the integrated stack passes 94 focused Markdown/highlighting tests and 159 timeline tests, plus web typecheck. The timing table remains pinned to the original measured builds.

Model: GPT-6. Harness: Codex.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.0 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	57.0 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	8	—	21	✅
Claude	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.1 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	57.8 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	8	—	21	✅

Baseline: unavailable · PR result: a660ad9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.9 KiB
• Claude decoded thread snapshot: 114.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Bugbot is paused — on-demand spend limit reached

Bugbot uses usage-based billing for this team and has hit its on-demand spend limit.

A team admin can raise the spend limit in the Cursor dashboard, or wait for the next billing cycle to continue.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR introduces stateful AST caching, parser wrapping, position shifting, and automatic fallback decisions in the production streaming Markdown path. Focused parity tests reduce risk, but the new parsing strategy and automatic processing gate are substantial runtime changes.

No code changes detected at a660ad9. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: c89c418c-586f-4eb9-8adf-91e9ddc98f7e

📥 Commits

Reviewing files that changed from the base of the PR and between 3b9cf6e and a660ad9.

⛔ Files ignored due to path filters (1)

• pnpm-lock.yaml is excluded by !**/pnpm-lock.yaml

📒 Files selected for processing (4)

• apps/web/package.json
• apps/web/src/components/ChatMarkdown.tsx
• apps/web/src/markdown-incremental.test.tsx
• apps/web/src/markdown-incremental.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 8 remain after this review.

---

📝 Walkthrough

Walkthrough

The web app adds an incremental Markdown parser for eligible streaming messages. It caches completed fenced-code prefixes, integrates the parser into ChatMarkdown, and adds tests for parsing equivalence and fallback cases.

Changes

Incremental Markdown streaming

Layer / File(s)	Summary
Incremental parser and plugin apps/web/src/markdown-incremental.ts, apps/web/package.json	Adds cached-prefix parsing, AST cloning, position shifting, full-parse fallbacks, and the exported incremental plugin. Adds required development dependencies.
Streaming renderer integration apps/web/src/components/ChatMarkdown.tsx	Enables incremental parsing during streaming when fenced code is present and no extra remark plugins are configured.
Parser behavior validation apps/web/src/markdown-incremental.test.tsx	Tests cached prefixes, streaming positions, line endings, definitions, edits, repeated renders, transformed nodes, and code-fence edge cases.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~25 minutes

Change: Refactor

Sequence Diagram(s)

sequenceDiagram
  participant ChatMarkdown
  participant IncrementalMarkdownPlugin
  participant MarkdownParser
  ChatMarkdown->>IncrementalMarkdownPlugin: enable for eligible streaming text
  IncrementalMarkdownPlugin->>MarkdownParser: parse current Markdown document
  MarkdownParser-->>ChatMarkdown: return Markdown AST for rendering

Loading

Suggested reviewers: t3dotgg

Merge Risk: ⚪ Minimal · up to a660a

No merge-blocking risk was identified.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 33.33% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 6 functions across 3 files. (1 skipped: 1…	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly and concisely describes the main change: reusing completed Markdown prefixes during streaming for web performance.
Description check	✅ Passed	The description explains what changed, why it changed, scope, performance results, testing, and UI verification. It includes screenshots and videos for the rendering changes. It does not use the templ…

Full details: Docstring Coverage

Explanation

Docstring coverage is 33.33% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 6 functions across 3 files. (1 skipped: 1 unsupported.)

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch legend-perf/incremental-markdown

---

_{Comment @coderabbitai help to get the list of available commands.}