perf(compilers): emit getter keys as string literals, never computed by ryansolid · Pull Request #3514 · solidjs/solid

ryansolid · 2026-09-17T08:04:20Z

Finding 1 of #3511. Both compilers emit non-identifier getter keys in compiled props literals as computed accessors:

// today
Polymorphic(merge({
  get as() { return merged.as ?? "button"; },
  get ["aria-disabled"]() { return merged.disabled || void 0; },
  get ["data-loading"]() { return merged.loading ? "" : void 0; },
  …
}, others));

Babel does it via t.objectMethod("get", id, [], body, !t.isValidIdentifier(key)), and packages/compiler/src/shared/ast.rs mirrors it on purpose ("Babel: … non-identifier getter keys are computed"). get "aria-disabled"() is the same property, but a computed key in an object literal takes V8 off the literal boilerplate path into per-property runtime definition for the whole object.

Measured on ButtonRoot's seven-getter <Polymorphic> props literal from yak-bench's polymorphic-chain, build + read all 7 keys, min of 7 × 200k:

form	TurboFan	`--no-opt`
literal, computed keys (today)	966 ns	715 ns
literal, string-literal keys (this PR)	525 ns	485 ns
literal, all identifier keys (control)	482 ns	482 ns

So the string-literal form is within 10% of what the same object costs with plain identifier keys — the computed key was the cost, not the hyphen. Every aria-*, data-*, and class prop paid it (class is a reserved word, so isValidIdentifier rejects it too).

Change

id at every site is already an Identifier or a StringLiteral (from convertJSXIdentifier, or t.identifier for locked DOM properties), so computed is just false. Six Babel sites — shared/component.ts ×3, dom/element.ts, ssr/element.ts, universal/element.ts — and the single Oxc funnel object_getter_property_with_statements. Nothing else in the emitted code changes.

Fixtures regenerated in both trees (vitest -u for Babel, UPDATE_OXC_FIXTURES=1 UPDATE_PARITY=1 for Oxc, including the three cross-mode parity .diff files). Audited: every changed line is a getter key. Babel's generator prints a reserved-word string key bare (get class()), which is valid and the same property.

Verified

@solidjs/babel-plugin 242/242; @solidjs/compiler vitest 5785/5785 and cargo test across all three feature sets.
@solidjs/web DOM (825), hydrate (185), server (1001) suites run end-to-end through the rebuilt native compiler; harness artifacts unchanged.

Case-level SSR A/B in the yak-bench harness is pending a quiet machine — tonight's runs had a load average of 11 from a remote-desktop session and the raw numbers disagreed by 10× within a tag, so I'm not quoting them. The microbenchmark is deterministic and the emitted code is otherwise identical; I'll add the harness number when I have one worth trusting.

changeset-bot · 2026-09-17T08:04:29Z

🦋 Changeset detected

Latest commit: 246eeeb

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

This PR includes changesets to release 11 packages

Name	Type
@solidjs/babel-plugin	Patch
@solidjs/compiler	Patch
test-integration	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
@solidjs/signals	Patch
solid-js	Patch
@solidjs/universal	Patch
@solidjs/web	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

coveralls · 2026-09-17T08:11:26Z

Coverage Report for CI Build 35200204109

Coverage remained the same at 71.46%

Details

Coverage remained the same as the base build.
Patch coverage: No coverable lines changed in this PR.
No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

Coverage Stats


Relevant Lines:	1018
Covered Lines:	772
Line Coverage:	75.83%
Relevant Branches:	790
Covered Branches:	520
Branch Coverage:	65.82%
Branches in Coverage %:	Yes
Coverage Strength:	14.93 hits per line

💛 - Coveralls

codspeed · 2026-09-17T08:24:40Z

Merging this PR will not alter performance

✅ 172 untouched benchmarks
⏩ 3 skipped benchmarks¹

_{Comparing perf/compiler-getter-keys (246eeeb) with next (f0ee8a0)}

3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

Both compilers emitted non-identifier getter keys in compiled props literals as computed accessors — `get ["aria-disabled"]() {}` — Babel via `t.objectMethod("get", id, [], body, !t.isValidIdentifier(key))` and the Oxc compiler mirroring it deliberately. `get "aria-disabled"() {}` is the same property, but a computed key in an object literal drops V8 off the literal boilerplate path into per-property runtime definition for the whole object. Measured on ButtonRoot's seven-getter `<Polymorphic>` props literal from yak-bench's polymorphic-chain: 966 ns → 525 ns per object to build and read once (#3511, finding 1). Every `aria-*`, `data-*`, and `class` prop paid it. `id` at every site is already an Identifier or a StringLiteral (from convertJSXIdentifier or `t.identifier`), so `computed` is simply `false`. Six Babel sites (shared/component.ts ×3, dom/ssr/universal element.ts) and the single Oxc funnel (`object_getter_property_with_statements`). Fixtures regenerated in both test trees plus the three cross-mode parity diffs; every changed line is a getter key. Babel prints reserved-word string keys bare (`get class()`), which is valid and the same property. Co-authored-by: Claude via Cursor <noreply@cursor.com>

ryansolid · 2026-09-17T08:27:00Z

Harness measurement, second attempt. The machine is still not quiet (load average 17–19: Spotlight indexing the new worktrees plus a remote-desktop session), and both wall-clock and process.cpuUsage per render swing ±40% within a single tag, so no time-based number is quotable yet.

What is stable under load is allocation. Scavenge count over a fixed render batch (--max-semi-space-size=1, --trace-gc enabled at runtime after warm-up; two reps each, reps agree to ±2%):

case baseline this PR Δ

polymorphic-chain (n=400 × 100) 807 / 770 733 / 734 −7%

tabs (150 × 100) 874 / 865 862 / 854 −1.5%

multifile-composition (150 × 60) 507 / 505 514 / 513 +1.5% (noise)

btn-variant (1000 × 150) 459 / 461 460 / 461 0

Consistent with the key mix: polymorphic-chain is the case whose props are mostly aria-* / data-*; the others are mostly identifier keys and never took the computed path. Allocation is a lower bound here, not the effect size — in the microbenchmark the computed form's extra cost is mostly the per-property runtime definition, and only ~14% of it is bytes. Time-based harness numbers still to come when the machine is quiet.

— Claude via Cursor

ryansolid · 2026-09-17T16:16:09Z

Harness SSR A/B on a quiet machine (load 5.5, Spotlight and the remote session gone). Baseline next at #3509 vs this branch, solid-mprim lane, both compiled with the respective checkout's Babel plugin; 8 interleaved pairs per case, fresh process each, 1 s warm + 2 s measure, instances/sec:

case median vs median paired-ratio median pairs favoring PR

polymorphic-chain (400) +7.5% +8.3% 6 / 8

tabs (150) +2.6% +6.4% 6 / 8

multifile-composition (150) +2.2% +2.8% 4 / 8 — noise-level

btn-variant (1000) −0.9% +0.2% flat

Baseline spread within a tag is ±3% on polymorphic-chain, ±12% on tabs, so the first row is a real effect and the second is suggestive. Same shape as the allocation proxy above: the win lands where the props are aria-* / data-* / class, and there is nothing to gain where every key is an identifier. polymorphic-chain SSR moves from 0.29× React to roughly 0.31×; the rest of that gap is finding 2 in #3511.

— Claude via Cursor

ryansolid force-pushed the perf/compiler-getter-keys branch from 8a0c039 to 246eeeb Compare September 17, 2026 08:32

ryansolid merged commit 1e7ebe9 into next Sep 17, 2026
7 checks passed

This was referenced Sep 17, 2026

Compiler: props literals with accessors are the largest SSR self-time bucket on composition cases (computed keys + closure-per-getter) #3511

Open

feat(signals): ASYNC_STORE_SETTER — a store setter callback that returns a Promise throws in dev #3527

Merged

Navigation Menu

Uh oh!

perf(compilers): emit getter keys as string literals, never computed - #3514

perf(compilers): emit getter keys as string literals, never computed#3514
ryansolid merged 1 commit into
nextfrom
perf/compiler-getter-keys

ryansolid commented Sep 17, 2026

Uh oh!

changeset-bot Bot commented Sep 17, 2026 •

edited

Loading

Uh oh!

coveralls commented Sep 17, 2026 •

edited

Loading

Uh oh!

codspeed Bot commented Sep 17, 2026 •

edited

Loading

Uh oh!

ryansolid commented Sep 17, 2026

Uh oh!

Uh oh!

ryansolid commented Sep 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

ryansolid commented Sep 17, 2026

Change

Verified

Uh oh!

changeset-bot Bot commented Sep 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

coveralls commented Sep 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Coverage Report for CI Build 35200204109

Coverage remained the same at 71.46%

Details

Uncovered Changes

Coverage Regressions

Coverage Stats

💛 - Coveralls

Uh oh!

codspeed Bot commented Sep 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

ryansolid commented Sep 17, 2026

Uh oh!

Uh oh!

ryansolid commented Sep 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

changeset-bot Bot commented Sep 17, 2026 •

edited

Loading

coveralls commented Sep 17, 2026 •

edited

Loading

codspeed Bot commented Sep 17, 2026 •

edited

Loading