fix: more typescript performance improvements#6320
Merged
Conversation
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (36)
📝 WalkthroughWalkthroughThis PR adds TypeScript ChangesTypeScript Variance Annotations and Table Type System Restructuring
🎯 4 (Complex) | ⏱️ ~45 minutes Possibly Related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
View your CI Pipeline Execution ↗ for commit e4f2a3a
☁️ Nx Cloud last updated this comment at |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
TanStack Table — TypeScript Type-Performance Report
Tracks TypeScript type-checking cost across the alpha branch, the beta branch as released (beta.10), and the optimized working tree (beta.11 candidate). Primary metric is
Instantiationsfromtsc --extendedDiagnostics(deterministic — confirmed identical across repeated runs). Wall times and memory are supporting evidence only.Setup
table-alpha/table/alpha@4f855fe4ebeta@925358059(+ uncommitted beta.11 work)@tanstack/table-coreversionpnpm build(tsdown) fortable-coreandreact-tablein the measured repo before every consumer benchmark (react-table, kitchen-sink) so declarations are current.Instantiationswas identical on every repeat.Benchmark commands
packages/table-corepnpm exec tsc --noEmit --extendedDiagnosticspackages/table-corepnpm exec tsc -p tests/tsconfig.declaration-emit.json --extendedDiagnosticspackages/react-tablepnpm exec tsc --noEmit --extendedDiagnosticsexamples/react/kitchen-sinkpnpm exec tsc --noEmit --extendedDiagnosticsThese mirror the repo's own
test:typesscript (tsc && tsc -p tests/tsconfig.declaration-emit.json).Results — Instantiations (primary metric)
@tanstack/table-coretsc --noEmit@tanstack/table-coredeclaration emit@tanstack/react-tabletsc --noEmitexamples/react/kitchen-sinktsc --noEmitPer-step deltas: alpha → beta.10 was −59.8% / −66.4% / −68.7% / −61.2%; beta.10 → beta.11 adds another −41.8% / −50.8% / −25.7% / −12.9% (same benchmark order as the table).
All framework adapters — Instantiations
pnpm exec tsc --noEmit --extendedDiagnosticsper adapter package (src + tests), all packages rebuilt first in each repo. beta.11 includes the explicit-type-args fix applied toangular-tableandpreact-table(sameconstructTablespread-inference hot spot as react'suseTable; lit/solid/svelte/vue already passed alias-typed variables and needed no change).@tanstack/angular-table@tanstack/lit-table@tanstack/preact-table@tanstack/react-table@tanstack/solid-table@tanstack/svelte-table@tanstack/vue-tableWithin beta, the
constructTableexplicit-type-args fix alone took angular-table 59,190 → 49,948 (−15.6%) and preact-table 49,125 → 41,558 (−15.4%).All kitchen-sink examples — Instantiations
angular/kitchen-sinklit/kitchen-sinkpreact/kitchen-sinkreact/kitchen-sinksolid/kitchen-sinkvue/kitchen-sink(vue-tsc)react/kitchen-sink-hero-uireact/kitchen-sink-mantinereact/kitchen-sink-material-uireact/kitchen-sink-react-ariareact/kitchen-sink-shadcn-basereact/kitchen-sink-shadcn-radixMeasurement caveats:
svelte/kitchen-sinkis excluded: plaintscdoes not check.sveltemarkup (it reports only ~2.2k instantiations from the loose.tsfiles in both repos);svelte-checkdoes not expose--extendedDiagnostics. The svelte adapter package row above covers the svelte surface.vue/kitchen-sinkis measured withvue-tsc(plaintscskips.vueSFCs and returns an identical, meaningless number in both repos). The UI-kit react variants carry large third-party type surfaces (MUI, Mantine, etc.), which dilutes the percentage but the absolute deltas (−145k to −154k) match the plain kitchen-sink.solid/kitchen-sinkhas 3 pre-existing app-level JSX errors, byte-identical in alpha and beta — unrelated to table types.Check times (secondary, two runs each): table-core 1.74s/1.83s → 1.07s/1.04s; declaration emit 1.58s/1.56s → 0.80s/0.87s; react-table 0.29s/0.26s → 0.21s/0.22s; kitchen-sink 0.53s/0.50s → 0.46s/0.50s.
The react-table regression and the variance fix
Intermediate state (before variance annotations): the
Table_Internalinterface conversion alone regressedreact-table's package-internal check to 136,031 (+84.6%). Root cause: with type parameters flowing into conditional types, the compiler's measured variance for the interface was unreliable, so relatingTable_Internal<A>toTable_Internal<B>across react's many fresh generic contexts fell back to member-by-member structural comparison, expanding the heavy members per pair. Adding explicitin out(invariant) variance annotations — the pattern used throughoutformandrouter— tells the checker to relate instantiations by their type arguments directly:Table_Internalalone took react-table from 136,031 to 65,891, and annotating the rest of the generic interfaces took it to 54,732 — 26% below the beta.10 baseline, and turned kitchen-sink from neutral to −12.9%. Variance annotations don't cache anything; they short-circuit relation checks by eliminating both variance measurement and the unreliable-variance structural fallback.beta.11 kept changes
Table_Internal→ broad interface (types/Table.ts). WasTable<TF,TD> & {…}— an intersection wrapping the deferredExtractFeatureMapTypesconditional, re-expanded structurally at every internal call site. Now an interface extendingOmit<Table_Table, broadened-keys>plus all four core API interfaces and all 14 stock featureTable_*interfaces, with the internal slots (_rowModels,_rowModelFns,options,initialState,store,atoms,baseAtoms) redeclared in their broad*_Allforms (TData-bearing conditionals dropped; TFeatures-only conditionals kept — feature-set count is small, data-type count is large). PublicTable<TF,TD>is untouched, so user-facing inference is unchanged.TableOptions_All's feature-map intersection de-U2I'd (types/TableOptions.ts). WasUnionToIntersection<Map[keyof Map]>(via aTableOptions_FeatureMap_Allhelper) — re-distributed 13 members per (features, data) pair. Now a direct intersection of the 13 named stock option interfaces (inlined intoTableOptions_All; the helper type was removed as perf-neutral, see experiment 6) plusTableOptions_PluginFeatureMapTypes, which falls back toUnionToIntersectiononly for plugin keys declaration-merged beyond the stock set ([Exclude<keyof Map, StockKeys>] extends [never]guard). Plugin flow-through verified against the custom-plugin example (mergesTableOptions_FeatureMapand readstable.options.onDensityChange).makeStateUpdatertakes a minimal structural param (utils.ts) —{ options: { atoms?: object }, baseAtoms: object }instead ofTable<TFeatures, any>. It only reads those two slots; any table view (public, internal, plugin) now passes without relating full table types. Accepts strictly more than before — non-breaking.in outvariance annotations on generic interfaces (166 sites acrosstable-core/src, applied toTFeatures/TDataparameters of exported interfaces). Both parameters are structurally invariant already (used in parameter and return positions throughout), so the annotations change no assignability semantics — TypeScript verifies annotations against structure and reported no contradictions. The win: the checker relates two instantiations of the same interface by comparing type arguments directly instead of measuring variance (which came back unreliable due to conditional-type usage) and falling back to structural member expansion. Same pattern as TanStackform(FieldApi,FormApi) androuter(Router,Route). This is what makes theTable_Internalinterface conversion a win on all benchmarks instead of a core-vs-react trade-off.createCoreRowModelfactory takesTable_Internal(removing anas unknown ascast);constructTablereturns via one explicit cast; twomergeOptionscall sites casttable.optionsdown toTableOptions; one unit test casts its constructed table toTable_Internal.constructTablecalls — reactuseTable, angularinjectTable, preactuseTableall builtconstructTable({ ...options, features: { <reactivity>, ...options.features } }), forcing generic inference + structural relation of the anonymous spread type againstTableOptions(react: ~740ms of traced check time).constructTable<TFeatures, TData>(...)skips inference. Saved ~12k (react), ~9.2k (angular), ~7.6k (preact). lit/solid/svelte/vue already passed alias-typed variables.Rejected / neutral experiments (with numbers)
ColumnDef_FeatureMapinto static/dynamic halves + plugin-residual extractor, so the TData-independent half caches per feature setRow_Coreinstead ofRow(skip deferred-conditional apparent type)ColumnHelper.accessorconditional signature into two overloads (key-first)Table_Internalas interface, members kept asInherited & AllintersectionsOmit-extends + broad slot members + de-U2I'd optionsin outonTable_Internalonlyin outon all generic interfaces, 166 sites (kept)in outon the publicTablealiasTableOptions_FeatureMap_AllintoTableOptions_All(its only internal user)in outonTValue extends CellDataparams (16 interfaces)TValue → unknowncovariant widening —Header<TF,TD,TValue>no longer assignable toHeader<TF,TD,unknown>(buildHeaderGroups.ts:116)TValueis genuinely covariant in output-position types; invariance rejects valid library and user code for a marginal winValidation
All re-run after the variance-annotation round:
packages/table-core:pnpm test:types(tsc + declaration emit) ✓,vitest run322/322 ✓, eslint ✓, publint ✓, build ✓ — run uncached vianx --skip-nx-cache.packages/react-table: tsc ✓, eslint ✓, build ✓ (no unit tests in package).angular,vue,solid,svelte,lit,preact):test:types✓ uncached.examples/react/kitchen-sink✓ andexamples/react/custom-plugin✓ (plugin declaration-merging exercises the rewritten options path).Table_Internalno longer carries plugin-merged table API members (plugin options/state still flow). Core code doesn't use plugin APIs and the plugin example only reads merged options, so nothing observable broke; a plugin reading its own merged table APIs throughTable_Internal(rather than the publicTable) would now need a cast.in outdeclares invariance. TypeScript verified the annotations against the measured structure with no contradictions, so today they change nothing semantically — but if a future interface edit makes a parameter genuinely covariant, the annotation would keep it invariant silently. Convention: annotate new generic interfaces and trust the compiler's TS2636 error to catch mistakes.Comparison to the beta.3 checkpoint
tsc --noEmittsc --noEmittsc --noEmit¹ The beta.3 report's alpha baselines differ from today's alpha measurement by ~0.3% (minor alpha commits since that snapshot). react-table and kitchen-sink baselines match exactly.
Profiling notes (for future rounds)
tsc --generateTrace+types.jsonaggregation byfirstDeclarationis the fastest way to localize: top creator in every program isUnionToIntersection's distributed function types (type-utils.ts:15), driven byExtractFeatureMapTypes.createTableHook.tsx,useTable) is highly sensitive to relation-check cost on core types: it doubled the package's check whenTable_Internalrelated structurally, and dropped 60% once variance annotations restored argument-level relation. Watch this benchmark whenever core's central generic types change shape.in out) are only legal on interfaces/classes and on type aliases whose RHS is an object/function/mapped type — not on intersection-with-conditional aliases likeTable(TS2637).Summary by CodeRabbit