[ryansolid]

Code-size pass on the client entries ahead of 2.0.0-next.31. Every change is behaviour-preserving: the client (jsdom), server and type suites are identical before and after, dev-mode messages are untouched, and .shake-check/check.mjs still passes. Measured against next @ 9600479 (the rebase base), so the numbers reflect what ships.

Sizes

Tooling: tsc && rollup -c output, then terser + brotli (level 11) per entry — the same solid-condition bundling / isServer=false shim / DEV fold the repo's .shake-check scripts use. "Typical app" is a minimal Vite production build (createRouter with three routes incl. one lazy, paths() links, useNavigate, useParams, query, action, createMemo, render), with solid bundled; "router share" is the same app with solid-js / @solidjs/web externalised so only the router's contribution is counted.

entry	min before	min after	Δ	brotli before	brotli after	Δ
. (solid condition, dist/index.jsx)	37 850	37 465	−385	12 792	12 757	−35
. (default, dist/index.js)	37 846	37 461	−385	12 776	12 732	−44
. client-folded (isServer=false, DEV folded)	33 891	33 455	−436	11 607	11 546	−61
./fs	5 450	5 431	−19	2 309	2 306	−3
./server	5 547	5 466	−81	2 273	2 249	−24
typical app — total (solid bundled)	90 369	89 914	−455	30 142	30 030	−112
typical app — router share	28 983	28 525	−458	10 226	10 157	−69

scripts/check-fs-gate.mjs: 104.5 / 125.5 KB → 104.1 / 125.0 KB.

Brotli on inputs this small has roughly ±15 B of noise from context reshuffling, so individual steps are reported with that caveat; the cumulative numbers above are the ones that matter.

Changes (router share of the typical app, min / brotli)

1. keysFor() reuses pathnameOf(url) (routers/factory.tsx) — it duplicated the pathname derivation match() already does. −44 / −11
2. ?. for route preload, one-expression segment encoder, flattened paths proxy get (routing.ts, paths.ts) — component?.preload?.() replaces the nested ifs; s => /^[:*]/.test(s) ? s : encodeSegment(s); the symbol branches in paths collapse to two returns. −80 / −8
3. window.history / history.state aliases (routers/history.ts) — saveCurrentDepth and browserHistory.getSource read the chain once. Brotli-neutral (+2) but −118 min and easier to read; kept for that reason. (Merged with fix: don't throw when history.state is null after replaceState #627's state?._depth ?? length - 1.)
4. Memo name options behind DEV (routing.ts) — DEV && { name: "lazyRoutes" } etc. Production solid never reads name on memos/signals (only createContext does), so these strings were pure dead weight in prod. −41 / −26
5. matchSegment / filter lookup (utils.ts) — one ternary chain instead of early-return ifs; filters?.[key] instead of matchFilters && matchFilters[key]. −60 / −21
6. settle() helper in query() (data/query.ts) — the cached and fresh paths each spelled out the same then/non-thenable settle; now one local function. Brotli-neutral (+4), −21 min; kept because it removes a duplicated branch that has to stay in sync.
7. Micro folds (routing.ts, data/action.ts, routers/components.tsx, routers/scrollRestoration.ts) — dropped the constant key getter closure in createLocation, disposers[i]?.(), form alias in handleFormAction, response?.data / response?.error, destructured performance.getEntriesByType?.("navigation"). −68 / −13
8. Trivia (routing.ts, data/action.ts, data/events.ts) — new URL(mockBase) once instead of a separate origin const, fn.with returns toAction(...) directly, flightData keys loop tightened, redundant rel && before rel.includes(...). −11 / −13

Tried and reverted (brotli-negative)

Brotli already compresses literal duplication very well, so extracting shared helpers to dedupe identical code consistently cost bytes on the wire even when it saved minified bytes: a shared isExternalAnchor / isSvg / preloadAnchor for events.ts + claims.ts (−370 min, +18 br), a toggle attribute helper in claims.ts (+8 br), a shared commitEntry for browserHistory / hashHistory (+ on the browser-only typical app), Object.fromEntries in hashKey, an isAsyncIterable helper in liveQuery, spread-gated ...(DEV && { name: "location" }). None are in this PR.

Verification

Run on next @ 9600479 and on this branch — identical:

• client (jsdom): 40 files / 442 tests
• server: 6 files / 38 tests
• test:types: pass
• .shake-check/check.mjs: all 8 checks ok (outputs reverted, not committed)
• pnpm build incl. fs gate: pass

Public API changes

None.

Suggestions (not made — need an API decision)

• scrollRestoration is always bundled. routers/scrollRestoration.ts ships in every client bundle even for apps that pass scrollRestoration: false. Making it an opt-in adapter (like the history adapters) would save every such app its full cost, but changes the createRouter option shape.
• import("./serverForms.js") in the delegation fallback pulls Vite's __vitePreload helper (~1.1 KB min) into every router app, even ones that never render a server form. A static import or a documented opt-in would remove it; either is a behaviour/API decision, not golf.
• setupNativeEvents' curried signature and its preload option name (vs preloadLinks on the router config) force an intermediate options object in createRouter (~90 B min). Internal, but tests import it directly, so left alone here.

Deviations from the brief

• The brief's "typical app" listed Router, Route, A, createAsync. Those exports don't exist on next (v2), so the app uses their v2 equivalents: createRouter with a route array, paths() links on plain <a>, useNavigate, query, action, and createMemo for reads.
• The fs gate figure includes solid (it bundles solid-js), so it moves less in relative terms than the per-entry numbers.

— Claude via Cursor

[changeset-bot]

🦋 Changeset detected

Latest commit: 3fe292d

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@solidjs/router	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR