Optimize render and stream hot paths

[cappuc]

Summary

Optimizes the render and stream hot paths. No behaviour changes, no public API removals.

The work was directed by call-count instrumentation of the storefront theme benchmark rather than by inspection — several of the biggest costs were invisible in the source. Most notably, 72% of RenderContext::evaluate() calls were pure pass-throughs, and a generator object was being allocated for every single variable reference to deliver a value the caller almost always took the first of.

Measured impact

PHPBench, base 9913b0e vs 79300f0, 10 iterations x 20 revs on ubuntu24:

PHP	benchRender	benchStream	benchParse	benchTokenize
8.2	+52.3%	+53.4%	+1.9%	+1.1%
8.3	+47.3%	+51.0%	+0.7%	+0.6%
8.4	+39.1%	+39.9%	+0.3%	+3.2%*
8.5	+36.7%	+39.8%	+4.7%*	+2.7%*

Throughput (ops/s), so positive is faster. No benchmark regressed on any version. RSD stayed at or below 2.7% on both sides, and total memory moved by less than 0.6% everywhere.

* Treat the parse and tokenize outliers on 8.4/8.5 as noise, not wins. Nothing here touches the tokenizer or the parser, and the base baseline drifted between runs on those versions (8.5 parse 298.74 → 265.04 ops/s, tokenize 917.51 → 762.32 ops/s for the same commit). The render and stream numbers are the trustworthy ones: large, consistent across all four versions, and tight RSD on both sides.

Changes

Variable resolution:

• Resolve variables against the scope chain lazily instead of eagerly scanning every scope, and fast-path array scopes.
• Replace the iterateVariables() generator with findVariable() (innermost value) and the existing findVariables() (all values). Scopes, env data and static variables are all plain arrays, so the lookup is now a bare array_key_exists instead of a trip through internalContextLookup()'s dispatch. Measured in isolation, the generator alone cost ~0.079µs per lookup.
• findVariables() is now only used for the outer-scope fallback, i.e. when a name resolves but its lookup chain breaks on the innermost value. That fallback skips the value already walked, so a broken chain no longer repeats side effects it triggered on the way.
• RenderContext::evaluate() is a loop rather than recursive, and the hot call sites guard with instanceof CanBeEvaluated instead of calling through for values that cannot be evaluated.
• normalizeValue() short-circuits on non-objects, and is skipped entirely for them.

Drops and filters:

• Generalize drop member resolution and cache it per class (DropMetadata, DropMemberResolution, DropMemberType).
• Optimize filter application.

Body rendering:

• Trim per-child overhead in BodyNode::render()/stream(). Text children — 57% of all children in the theme benchmark — append their value directly, skipping a method call, two instanceof checks and the try/catch.

API

RenderContext::findVariable() is new. findVariables() keeps its signature and semantics. Nothing was removed, so this is not a breaking change.

Testing

• vendor/bin/pest — 824 passed (2170 assertions), on PHP 8.2–8.5, prefer-lowest and prefer-stable
• vendor/bin/phpstan analyse --no-progress — no errors
• vendor/bin/pint --test — passed
• PHPBench regression threshold (5%) — PASSED on all four PHP versions

New regression coverage for the outer-scope fallback, drop member caching, and variable lookup typing.

[github-actions]

PHPBench comparison (PHP 8.2)

PHP 8.2 | Runner ubuntu24 | Base 9913b0e | PR 79300f0 | 10 iterations x 20 revs | 1 warmup

Positive ops/s is faster. RSD above 5% is marked high.

Benchmark	Base ops/s	PR ops/s	Delta ops/s	RSD (base / PR)	Delta memory	Memory %
ThemeBench::benchParse	406.13 ops/s	413.78 ops/s	+1.88%	0.97% / 1.01%	+13.94 KiB	+0.17%
ThemeBench::benchRender	157.38 ops/s	239.75 ops/s	+52.34%	1.11% / 1.46%	+8.91 KiB	+0.10%
ThemeBench::benchStream	157.19 ops/s	241.07 ops/s	+53.36%	1.42% / 1.68%	+7.28 KiB	+0.09%
ThemeBench::benchTokenize	856.32 ops/s	865.71 ops/s	+1.10%	1.66% / 1.37%	+23.20 KiB	+0.31%

• Improved (> +2%): 2 (median +52.85%)

• Neutral (within ±2%): 2

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: +0.16%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.4)

PHP 8.4 | Runner ubuntu24 | Base 9913b0e | PR 79300f0 | 10 iterations x 20 revs | 1 warmup

Positive ops/s is faster. RSD above 5% is marked high.

Benchmark	Base ops/s	PR ops/s	Delta ops/s	RSD (base / PR)	Delta memory	Memory %
ThemeBench::benchParse	292.89 ops/s	293.81 ops/s	+0.31%	1.36% / 2.15%	-3.59 KiB	-0.04%
ThemeBench::benchRender	121.52 ops/s	169.04 ops/s	+39.10%	1.36% / 1.23%	-26.05 KiB	-0.30%
ThemeBench::benchStream	122.94 ops/s	172.01 ops/s	+39.91%	1.49% / 1.79%	-61.44 KiB	-0.77%
ThemeBench::benchTokenize	882.37 ops/s	910.70 ops/s	+3.21%	1.91% / 2.09%	+260.94 KiB	+3.49%

• Improved (> +2%): 3 (median +39.10%)

• Neutral (within ±2%): 1

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: +0.52%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.3)

PHP 8.3 | Runner ubuntu24 | Base 9913b0e | PR 79300f0 | 10 iterations x 20 revs | 1 warmup

Positive ops/s is faster. RSD above 5% is marked high.

Benchmark	Base ops/s	PR ops/s	Delta ops/s	RSD (base / PR)	Delta memory	Memory %
ThemeBench::benchParse	400.11 ops/s	402.88 ops/s	+0.69%	2.51% / 0.87%	-3.73 KiB	-0.04%
ThemeBench::benchRender	159.98 ops/s	235.69 ops/s	+47.32%	0.57% / 1.17%	+8.91 KiB	+0.10%
ThemeBench::benchStream	158.81 ops/s	239.83 ops/s	+51.02%	1.04% / 2.05%	+7.28 KiB	+0.09%
ThemeBench::benchTokenize	847.54 ops/s	852.73 ops/s	+0.61%	1.36% / 1.49%	-3.57 KiB	-0.05%

• Improved (> +2%): 2 (median +49.17%)

• Neutral (within ±2%): 2

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: +0.03%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.5)

PHP 8.5 | Runner ubuntu24 | Base 9913b0e | PR 79300f0 | 10 iterations x 20 revs | 1 warmup

Positive ops/s is faster. RSD above 5% is marked high.

Benchmark	Base ops/s	PR ops/s	Delta ops/s	RSD (base / PR)	Delta memory	Memory %
ThemeBench::benchParse	265.04 ops/s	277.59 ops/s	+4.73%	1.50% / 2.66%	-309.91 KiB	-3.42%
ThemeBench::benchRender	122.67 ops/s	167.63 ops/s	+36.66%	1.14% / 1.28%	+110.45 KiB	+1.25%
ThemeBench::benchStream	122.80 ops/s	171.64 ops/s	+39.78%	1.53% / 1.36%	+156.47 KiB	+1.98%
ThemeBench::benchTokenize	762.32 ops/s	783.22 ops/s	+2.74%	2.05% / 2.34%	-3.57 KiB	-0.05%

• Improved (> +2%): 4 (median +20.70%)

• Neutral (within ±2%): 0

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: -0.14%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED