Stacked on #27480. Even after #27480's cache, a reply that isn't cached yet still pays the full micromark + remark + rehype pipeline. On large threads that is about 80 ms of parsing when a thread opens. Most of it was GFM checking for syntax the text could never contain.
This keeps the same engine and plugins, and the rendered output stays the same:
chat/markdownSyntaxGates.ts replaces remark-parse's parser. GFM constructs whose trigger characters are missing are left out, so email links aren't tried at every word without an @ and tables aren't tried at every line when no delimiter row exists. Link and table checks also run per position, from micromark's own previous/tokenize hooks. GFM's second search for bare links in the parsed text only runs when some text outside a link could hold one.rehype-raw only when needed. It parsed every reply as HTML again. Now it runs only for raw HTML, or for nesting the HTML parser would repair, such as a footnote inside a link label. Otherwise the pipeline repeats the two property changes that round trip made.mdast-util-to-hast collects definitions with a plain loop instead of unist-util-visit.| Over 2,039 real replies (1.9 MiB), production build | Before | After |
|---|---|---|
| Full render, best | 2,080 ms | 930 ms |
| Full render, median | 2,284 ms | 1,089 ms |
| Parse only (CommonMark without extensions: 615 ms) | 1,491 ms | 706 ms |
| Real app, opening a 1,114-reply thread: markdown time | 91 ms | 40 ms |
Parity was checked against the unmodified code with unpatched dependencies:
hast-util-to-jsx-runtime drops blank text inside table elements.Tests compare the renderer with react-markdown (gates, escaped links, BOM, containers, CRLF, streaming through the incremental parser), and compare the shared pipeline with plain rehype-raw. Each fix fails its test when reverted. Codex (GPT 6.1 Sol) and Fable reviewed it in two rounds and found no remaining regressions.
Benchmark and parity scripts with their results: markdown-bench.tar.gz. The corpus is left out because it is private chat history; the README explains how to build one from your own database.
Opus 5.5 via Claude Code
🤖 Generated with Claude Code
Added 1 commit
Verdict: Approved at b8638a5
Macroscope's review found this PR approvable — This change reduces Markdown processing work while retaining HTML parsing where needed, sanitization, and URL transformation. Its effects are confined to rendering and Markdown-derived processing, with no product-default or persistent-data changes.
You can add or adjust custom eligibility rules. Learn more.
âś… Thread transfer remains within every enforced ceiling.
ℹ️ No successful
mainbaseline artifact is available yet. This run establishes the initial measurement.
| Provider | Metric | Main baseline | This PR | Impact | PR ceiling | |
|---|---|---|---|---|---|---|
| Codex | Total thread wire | — | 5.0 KiB | — | 6.8 KiB | ✅ |
| Codex | Thread snapshot wire | — | 3.8 KiB | — | 4.9 KiB | ✅ |
| Codex | Live turn WebSocket wire | — | 1.3 KiB | — | 2.0 KiB | ✅ |
| Codex | Live turn WebSocket decoded | — | 21.6 KiB | — | 24.4 KiB | ✅ |
| Codex | Live turn messages | — | 2 | — | 8 | ✅ |
| Claude | Total thread wire | — | 5.1 KiB | — | 6.8 KiB | ✅ |
| Claude | Thread snapshot wire | — | 3.8 KiB | — | 4.9 KiB | ✅ |
| Claude | Live turn WebSocket wire | — | 1.3 KiB | — | 2.0 KiB | ✅ |
| Claude | Live turn WebSocket decoded | — | 21.9 KiB | — | 24.4 KiB | ✅ |
| Claude | Live turn messages | — | 2 | — | 8 | ✅ |
Baseline: unavailable · PR result: b8638a5 · Source CI: success
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.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.
No actionable comments were generated in the recent review.
f9bc7384-f8c6-484d-9c15-3770579bfd76Reviewing files that changed from the base of the PR and between 9db57770d4e8b48f98835bac31be4fa5af0a0b27 and b8638a5ad233d518ac344921cf5a706625ad4c0a.
pnpm-lock.yaml is excluded by !**/pnpm-lock.yamlapps/web/package.jsonapps/web/src/components/chat/markdownRenderer.test.tsxapps/web/src/components/chat/markdownRenderer.tsapps/web/src/components/chat/markdownSyntaxGates.tsapps/web/src/textDirection.tspackages/shared/package.jsonpackages/shared/src/markdownListIndentation.tspackages/shared/src/markdownPipeline.test.tspackages/shared/src/markdownPipeline.tspatches/mdast-util-to-hast@13.2.1.patchpatches/micromark@4.0.2.patchpnpm-workspace.yamlIncluded review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.
The chat Markdown renderer now gates parser extensions based on document content. The shared rehype pipeline conditionally uses raw HTML parsing. The changes also update Markdown tree processing and text-direction handling.
Changes: Markdown Rendering
| Layer / File(s) | Summary |
|---|---|
Gated chat Markdown parsing apps/web/src/components/chat/markdownSyntaxGates.ts, apps/web/src/components/chat/markdownRenderer.ts, apps/web/src/components/chat/markdownRenderer.test.tsx, apps/web/package.json, patches/micromark@4.0.2.patch, pnpm-workspace.yaml | The chat renderer applies parser gates for selected GFM constructs and caches merged extensions. Tests compare parser output across Markdown samples and streamed reply prefixes. The workspace pins patched parser versions. |
Conditional shared HTML processing packages/shared/src/markdownPipeline.ts, packages/shared/src/markdownPipeline.test.ts, packages/shared/src/markdownListIndentation.ts, packages/shared/package.json, patches/mdast-util-to-hast@13.2.1.patch | The shared pipeline invokes rehypeRaw when raw HTML or specified nesting is present. Related changes update tree traversal and compare pipeline output with a version that inserts rehype-raw. |
Text Direction
| Layer / File(s) | Summary |
|---|---|
RTL-free text handling apps/web/src/textDirection.ts | The function returns "ltr" for text with Unicode letters but no RTL characters. It returns undefined when the input contains neither. |
Priority: External image Normal
Estimated code review effort: 3 (Moderate) | ~25 minutes
Change: Refactor
| 1 | sequenceDiagram |
| 2 | participant MarkdownRenderer |
| 3 | participant remarkParseWithoutUnusedSyntax |
| 4 | participant micromark |
| 5 | participant fromMarkdown |
| 6 | MarkdownRenderer->>remarkParseWithoutUnusedSyntax: applies the parser attachment |
| 7 | remarkParseWithoutUnusedSyntax->>micromark: parses with gated extensions |
| 8 | remarkParseWithoutUnusedSyntax->>fromMarkdown: parses with merged mdast extensions |
Possibly related issues
pingdotgg/t3code#16105: The PR gates GFM strikethrough parsing, but the available summary does not establish a change to the single-tilde rule described by the issue.Suggested reviewers: t3dotgg, yash-singh1, maria-rcks
b8638No actionable regression is established from the supplied evidence; the change appears mergeable after normal checks.
Failed Failed checks: 1 warning
| Check name | Status | Explanation | Resolution |
|---|---|---|---|
| Description check | Warning | The description follows the template well. It states the problem (uncached replies pay the full parsing pipeline, about 80 ms on large threads) and the change (syntax gates, conditional rehype-raw, tw… | Add a Scope and approval section. Link the triaged bug issue or the maintainer-approved discussion, with the approval comment. If no prior issue exists, explain why this performance change qualifies as a very small, focused fix. Also confir… |
| Check name | Status | Explanation |
|---|---|---|
| Title check | Passed | The title states the main change: a performance improvement to web chat Markdown parsing that keeps the same output. The changeset adds syntax gates to the Markdown parser, runs rehype-raw conditional… |
| 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. |
Explanation
The title states the main change: a performance improvement to web chat Markdown parsing that keeps the same output. The changeset adds syntax gates to the Markdown parser, runs rehype-raw conditionally, and adds fast paths, all aimed at faster parsing. The title is also written in conventional commit form with a scope, as the repository template asks. The title is 72 characters, so it is slightly longer than ideal but still readable.
Explanation
The description follows the template well. It states the problem (uncached replies pay the full parsing pipeline, about 80 ms on large threads) and the change (syntax gates, conditional rehype-raw, two upstream-style patches, and fast paths). It gives benchmark tables and parity methodology in the Verification area. It also names the model and harness used. Gaps remain against the template. The Scope and approval section has no link to a triaged issue or maintainer-approved discussion, and no explanation for why the change qualifies without one. The description also cites a prior stacked PR but does not show that PR's approval. The Verification section lists test coverage, but the tests and parity scripts sit in an external tarball, not in the repository. The description also does not include a before and after screenshot, though this change is not a UI change.
Resolution
Add a Scope and approval section. Link the triaged bug issue or the maintainer-approved discussion, with the approval comment. If no prior issue exists, explain why this performance change qualifies as a very small, focused fix. Also confirm that stacked PR #27480 is merged or reviewed, or state its status. Keep the verification notes that already exist. Confirm the test names or file paths in the repository so reviewers can run the focused tests without the external tarball.
Comment @coderabbitai help to get the list of available commands.