perf(plugins): skip dynamic-request AST parse for static-import-only modules — 5000-route build −21%

[hyf0]

No dependencies — reviewable and mergeable on its own. Touches only plugins/ (ast-utils + ignore-dynamic-requests + extensionless-dynamic-import).

What

Extracts a shared, documented pre-parse gate — DYNAMIC_IMPORT_PRESCAN / mayContainDynamicImport in ast-utils — and uses it as the "might this module contain a dynamic import(...)?" check in:

• ignore-dynamic-requests (in-handler prescan), and
• extensionless-dynamic-import (Rolldown transform.filter.code, which already used the same idea inline).

Why

ignore-dynamic-requests pre-filtered with:

if (!code.includes("require") && !code.includes("import")) return null;

The import substring matches almost every ESM module (every static import x from "…"), so the plugin ran parseAst + a full AST walk on essentially the entire module graph — then found nothing to do in the vast majority. It was the only one of vinext's AST-walking transform plugins without a tight code filter; the siblings already gate natively (require-context → /\brequire\b…\.context/, extensionless-dynamic-import → import(, typeof-window → typeof window, import-meta-url → import.meta…).

Static imports never put a ( (or a comment leading to one) right after the import keyword, so narrowing the import side to dynamic-call syntax skips static-import-only modules entirely. The require side stays a broad substring check (it must keep covering aliasing and comment-separated require/* … */().

Pulling the pattern into a named, documented helper makes the intent explicit — this is a deliberate, measured performance gate, not an incidental regex — and lets both plugins share one source of truth. extensionless-dynamic-import additionally gains correct handling of comment-separated dynamic imports (import/* … */("…")), which its old \bimport\s*\( filter missed.

Measured

On the 5000-route stress fixture from #2388 (same machine, clean builds, median of 3):

	Build (5000 routes)
before	~22.2 s
after	~17.5 s
	−21%

CPU profile of the build: jsonParseAst 1536 → 510 ms, forEachAstChild 1199 ms → off the chart, with matching GC relief — i.e. ~5000 redundant module parses removed. The saving is in AST parsing and is orthogonal to #2389 (the route-graph read cache); they stack.

Correctness

• The gate is intentionally over-inclusive: a false positive costs one redundant parse, a false negative would silently skip a real dynamic import — so \s*[(/] tolerates whitespace and block/line comments between import and (, and require stays broad.
• Dynamic-request / dynamic-import suites pass — dynamic-requests-build, extensionless-dynamic-import, import-meta-url: 100 tests. vp lint clean.
• Equivalence is asserted via these suites and unchanged transform behavior, not output bytes: vinext's production build is not bit-reproducible run-to-run.

[github-actions]

Performance benchmarks

Compared d9bd689 against base da086c0 using alternating same-runner rounds. Next.js was unchanged and skipped.

0 improved · 0 regressed · 6 within ±1.5%

Scenario	Framework	Baseline	Current	Change
Client bundle size (gzip)	vinext	122.0 KB	122.0 KB	⚫ -0.0%
Client entry size (gzip)	vinext	117.0 KB	117.0 KB	⚫ -0.0%
Dev server cold start	vinext	2.34 s	2.36 s	⚫ +0.6%
Production build time	vinext	2.77 s	2.73 s	⚫ -1.2%
RSC entry closure size (gzip)	vinext	94.3 KB	94.3 KB	⚫ +0.0%
Server bundle size (gzip)	vinext	160.1 KB	160.0 KB	⚫ -0.0%

View detailed results and traces

_{🟢 improvement · 🔴 regression · ⚫ change below 1.5% · paired base/head}