[PHP 8.6] Partial Function Application (PFA)

[lisachenko]

Refs #224 (part of epic #219)

Full PHP 8.6 Partial Function Application support

PFA lets any call use a ? placeholder for a single open argument, evaluating to a Closure:

$makeSlug = str_replace(' ', '-', ?);

This PR started as interim behavior-documenting coverage while parsing was blocked upstream. nikic/PHP-Parser#1159 merged on 2026-09-05 and shipped in php-parser 5.9.0 on 2026-09-13 (ArgPlaceholder node), so the follow-up planned in this PR's history is now implemented here:

• composer.json: nikic/php-parser bumped ^5.4 → ^5.9; rector/rector bumped ^2.0 → ^2.6.7 (see below).
• Tests flipped from "raises a parse error" to "reflects cleanly": the engine and the public ReflectionFile entry point reflect the PFA stub (functions, classes, signatures) on every host runtime — confirming ParserFactory::createForNewestSupportedVersion() picks the 5.9 grammar up on a PHP 8.5 host with no engine change; every placeholder position parses (trailing, leading, multiple, named f(name: ?), method/static/new contexts) with the expected ArgPlaceholder count; method bodies keep their ArgPlaceholder nodes with isPartialFunctionApplication() true.
• Resolver contract (decided per the interim plan): a placeholder argument in a constant-expression position degrades into ReflectionException — the same contract user-defined first-class callables already have. The message is generalized ("Cannot statically resolve a placeholder argument in a function/constructor call") since the existing instanceof Arg guard covers ArgPlaceholder and VariadicPlaceholder alike; dedicated tests pin all three shapes.
• Cache hygiene re-pinned with a genuinely broken source (PFA no longer fails to parse).
• Guarded PHP 8.6 test includes the stub, compares parsed vs native signatures, and executes a partial application (Closure returned, ('a b') → 'a-b').
• FCC regression guard kept; note that php-parser 5.9 deliberately reports a first-class callable as a special case of PFA (isPartialFunctionApplication() === true for foo(...)), which the test documents.
• The stub stays out of AbstractTestCase::getFilesToAnalyze() — it now parses everywhere but is still a compile error when included on PHP 8.5.

Why the rector floor bump is needed

Rector's package bootstrap.php preloads its bundled, unprefixed php-parser copy whenever PHPUnit ≥ 12 is running — which the PHPUnit 13 bump (#236) activated, silently shadowing the project's own php-parser in every phpunit run. Rector 2.6.7 (2026-09-13) is the first release bundling php-parser ^5.9; with an older rector, the lowest-dependencies CI job would run the suite against a shadowed pre-PFA parser and fail. Worth keeping in mind generally: under PHPUnit ≥ 12, the test suite exercises rector's bundled php-parser build, not the locked nikic/php-parser package.

Local validation

Check	Result
vendor/bin/phpunit (PHP 8.5.9, php-parser 5.9.0, PHPUnit 13.3.4)	OK — 13810 tests, 16139 assertions, 135 skipped, 2 incomplete, 0 failures
php8.6 vendor/bin/phpunit (PHP 8.6.0beta1)	OK — 15041 tests, 30385 assertions, 142 skipped, 2 incomplete, 0 failures
PFA test class on 8.6	OK — 20 tests, 63 assertions, 0 skipped
vendor/bin/phpstan analyse src --no-progress (level 10)	[OK] No errors

🤖 Generated with Claude Code

https://claude.ai/code/session_01Sy8BcM8ivUEpu8uVm7wADn

[lisachenko]

@claude can you have a look here for feature request to nikic/php-parser or create it there?

[lisachenko]

Checked upstream — a feature request already exists and is actively moving, so I did not create a duplicate:

• Issue: nikic/PHP-Parser#1158 — "PHP 8.6: Partial function application" (opened 2026-07-18 by @sebastianbergmann; motivated by code-coverage breaking in the PHPUnit ecosystem when any covered file uses PFA).
• Implementation PR: nikic/PHP-Parser#1159 — "Implement support for partial function application" (open). It introduces an ArgPlaceholder node (named after review feedback from @nikic, who is actively reviewing) alongside the existing VariadicPlaceholder, plus an isPartialFunctionApplication() helper treating first-class callables as a special case. Last activity 2026-08-24.

So the follow-up here is exactly what this PR's body anticipates: once #1159 ships in a PHP-Parser release, bump the nikic/php-parser constraint, flip this PR's "raises a parse error" assertions to "reflects cleanly", and decide the NodeExpressionResolver contract for ArgPlaceholder in constant-expression positions. The new node names (ArgPlaceholder with a name property) match the graceful-degradation tests already landed here.

Worth subscribing to #1159 for release timing — I could not watch it from this session (repository outside my access scope).

---

Generated by Claude Code

[lisachenko]

Upstream update: nikic/PHP-Parser#1159 (partial function application support, ArgPlaceholder node) was merged into master on 2026-09-05 (commit 23c8e36, all checks green), closing upstream #1158.

No tagged release contains it yet — latest is still 5.8.0. My recurring check keeps watching; as soon as a release ships PFA parsing, I'll do the full follow-up on this branch (constraint bump, flip the parse-error assertions to positive reflection tests, ArgPlaceholder handling in NodeExpressionResolver, full gate on 8.5 + 8.6) and report here. Merge stays with @lisachenko.

---

Generated by Claude Code

[lisachenko]

php-parser 5.9.0 shipped PFA support on 2026-09-13, so this PR is now the full implementation — the "On Hold" reason is resolved. Pushed in two commits (33d0d4c, b2ef543), PR body updated to match:

• nikic/php-parser ^5.4 → ^5.9; parse-error assertions flipped to positive reflection (all placeholder positions incl. named f(name: ?); ArgPlaceholder in method bodies; isPartialFunctionApplication()), no engine change needed — createForNewestSupportedVersion() picks the grammar up on a PHP 8.5 host, as asserted.
• Resolver contract: placeholder arguments in constant-expression positions throw ReflectionException (same as user-defined FCC), message generalized for both placeholder node kinds.
• Guarded 8.6 test executes the stub natively (partial application returns a working Closure).
• One finding worth your attention: since the PHPUnit 13 bump (build(deps-dev): update phpunit/phpunit requirement from ^11.0.7 to ^13.3.1 #236), Rector's bootstrap preloads its bundled unprefixed php-parser into every phpunit run, shadowing the project's own copy. Rector 2.6.7 is the first release bundling php-parser ^5.9, so I also raised rector/rector to ^2.6.7 — otherwise the lowest-deps CI job would test against a shadowed pre-PFA parser and fail.

Local gate green: 13,810 tests on 8.5 / 15,041 on 8.6 (PFA class 20/20 on 8.6, partial application evaluates to 'a-b'), PHPStan level 10 clean. I've stopped the recurring upstream check; merge decision is yours as always.

---

Generated by Claude Code