[cappuc]

Summary

size, first and last resolve as pseudo-properties only with dot notation, like Shopify Liquid.

 {{ list.size }}        → 3            (unchanged)
-{{ list["size"] }}     → 3
+{{ list["size"] }}     → ""           (undefined variable in strict mode)
 {{ hash["size"] }}     → hash's "size" key   (unchanged)
 {{ list[key] }}        → ""           (unchanged: dynamic keys never were pseudo-properties)

ExpressionParser / VariableLookup::fromMarkup
  ["size"] | ["first"] | ["last"]  → new BracketKey('size')   # other keys stay plain strings
walkLookupParts
  BracketKey → plain key lookup, skips the pseudo-property fallback
RenderContext::findPathOverride
  BracketKey → plain key when matching override paths   # set('a.size') also overrides a["size"]

BracketKey is exportable, so compiled templates rebuild it as new BracketKey('size'). VariableLookup::toString() writes it back as ["size"], so RenderContext::get((string) $lookup) round-trips.

Evidence

• Before: {{ value["size"] }}|{{ value["first"] }} with value = "Hello" → 5|H
  After: |, like Shopify Liquid 5.13. In Shopify's variable_lookup.rb, command flags are set only for non-bracket lookups.
• The tests that asserted the old behaviour are updated, and the new cases cover a hash with a real size key, dynamic keys and the RenderContext::get() round-trip. Parsed, compiled and strict variants all run. 3,818 tests pass.
• Before (after rebasing on fix(context)!: path overrides instead of outer-scope lookup fallback #100): with a = {size: "original"} and set('a.size', 'override'), a["size"] → original, while a.size → override.
  After: both → override. Covered by dotted keys override bracket pseudo-property keys.
• The serialize, var-export and compiled template caches all render {{ v["size"] }}|{{ v.size }}|{{ h["size"] }} as |2|9.
• No Liquid template in our sites, CMS themes or laravel-cms uses ["size"], ["first"] or ["last"].
• PHPBench A/B against main (measured before the rebase on fix(context)!: path overrides instead of outer-scope lookup fallback #100): about −1% on ThemeBench render and stream; OperationBench ranges from −1.3% to 0%.

Merge Danger

Door: two-way

Blast Radius: templates

Templates using a["size"], a["first"] or a["last"] on arrays or strings now get an empty value instead of the pseudo-property. Replace them with dot notation. The README entry documenting the old difference is removed.

Rebased on main after #100: the changelog conflict is resolved and findPathOverride() treats a BracketKey lookup as its plain value.

[github-actions]

PHPBench comparison (PHP 8.3)

PHP 8.3 | Runner ubuntu24 | Base e2abc5f | PR f789070 | 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::benchCompile	203.40 ops/s	202.13 ops/s	-0.63%	0.98% / 0.72%	-224.00 B	-0.01%
ThemeBench::benchParse	652.09 ops/s	653.72 ops/s	+0.25%	1.28% / 1.16%	-192.00 B	-0.00%
ThemeBench::benchRender	553.09 ops/s	543.82 ops/s	-1.68%	1.00% / 0.74%	-256.00 B	-0.00%
ThemeBench::benchRenderCompiled	666.10 ops/s	650.44 ops/s	-2.35%	1.34% / 1.07%	-272.00 B	-0.00%
ThemeBench::benchStream	467.99 ops/s	469.37 ops/s	+0.30%	1.89% / 0.72%	-256.00 B	-0.00%
ThemeBench::benchStreamCompiled	585.20 ops/s	576.27 ops/s	-1.53%	1.12% / 0.74%	-272.00 B	-0.00%
ThemeBench::benchTokenize	1,364.30 ops/s	1,364.65 ops/s	+0.03%	2.06% / 1.64%	-192.00 B	-0.00%

• Improved (> +2%): 0

• Neutral (within ±2%): 6

• Regressed (< -2%): 1 (median -2.35%)

• Worst throughput regression: ThemeBench::benchRenderCompiled (-2.35%)

• Total memory change: -0.00%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.5)

PHP 8.5 | Runner ubuntu24 | Base e2abc5f | PR f789070 | 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::benchCompile	132.96 ops/s	133.92 ops/s	+0.72%	0.74% / 0.44%	-224.00 B	-0.01%
ThemeBench::benchParse	429.89 ops/s	430.20 ops/s	+0.07%	0.80% / 0.57%	-192.00 B	-0.00%
ThemeBench::benchRender	351.39 ops/s	348.46 ops/s	-0.83%	1.08% / 0.32%	-256.00 B	-0.00%
ThemeBench::benchRenderCompiled	419.16 ops/s	413.16 ops/s	-1.43%	0.41% / 0.96%	-272.00 B	-0.00%
ThemeBench::benchStream	296.48 ops/s	294.62 ops/s	-0.63%	0.31% / 0.51%	-256.00 B	-0.00%
ThemeBench::benchStreamCompiled	365.38 ops/s	364.64 ops/s	-0.20%	0.41% / 1.26%	-272.00 B	-0.00%
ThemeBench::benchTokenize	785.50 ops/s	785.86 ops/s	+0.05%	0.93% / 0.62%	-192.00 B	-0.00%

• Improved (> +2%): 0

• Neutral (within ±2%): 7

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: -0.00%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.2)

PHP 8.2 | Runner ubuntu24 | Base e2abc5f | PR f789070 | 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::benchCompile	106.80 ops/s	106.56 ops/s	-0.23%	0.51% / 1.39%	-224.00 B	-0.01%
ThemeBench::benchParse	362.80 ops/s	363.92 ops/s	+0.31%	0.99% / 1.61%	-192.00 B	-0.00%
ThemeBench::benchRender	278.96 ops/s	274.48 ops/s	-1.61%	0.99% / 1.32%	-256.00 B	-0.00%
ThemeBench::benchRenderCompiled	326.42 ops/s	320.89 ops/s	-1.70%	1.19% / 1.05%	-272.00 B	-0.00%
ThemeBench::benchStream	230.71 ops/s	229.27 ops/s	-0.62%	0.84% / 0.76%	-256.00 B	-0.00%
ThemeBench::benchStreamCompiled	289.17 ops/s	287.03 ops/s	-0.74%	0.76% / 0.74%	-272.00 B	-0.00%
ThemeBench::benchTokenize	727.20 ops/s	723.06 ops/s	-0.57%	1.12% / 0.81%	-192.00 B	-0.00%

• Improved (> +2%): 0

• Neutral (within ±2%): 7

• Regressed (< -2%): 0

• Worst throughput regression: n/a

• Total memory change: -0.00%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED

[github-actions]

PHPBench comparison (PHP 8.4)

PHP 8.4 | Runner ubuntu24 | Base e2abc5f | PR f789070 | 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::benchCompile	168.43 ops/s	165.80 ops/s	-1.56%	2.29% / 1.83%	-224.00 B	-0.01%
ThemeBench::benchParse	521.00 ops/s	525.99 ops/s	+0.96%	1.16% / 1.63%	-192.00 B	-0.00%
ThemeBench::benchRender	418.40 ops/s	426.86 ops/s	+2.02%	2.23% / 1.72%	-256.00 B	-0.00%
ThemeBench::benchRenderCompiled	480.31 ops/s	505.63 ops/s	+5.27%	1.44% / 1.92%	-272.00 B	-0.00%
ThemeBench::benchStream	340.87 ops/s	359.82 ops/s	+5.56%	1.98% / 1.91%	-256.00 B	-0.00%
ThemeBench::benchStreamCompiled	448.73 ops/s	438.87 ops/s	-2.20%	1.53% / 1.60%	-272.00 B	-0.00%
ThemeBench::benchTokenize	948.08 ops/s	954.25 ops/s	+0.65%	2.26% / 1.92%	-192.00 B	-0.00%

• Improved (> +2%): 3 (median +5.27%)

• Neutral (within ±2%): 3

• Regressed (< -2%): 1 (median -2.20%)

• Worst throughput regression: ThemeBench::benchStreamCompiled (-2.20%)

• Total memory change: -0.00%

• Regression threshold (PHPBENCH_MAX_REG): 5.00%

• Threshold status: PASSED