php-in-php: superglobals_refresh.c FILES/multipart parser — migrate bracket parser to lib/Web (phase 2 of #5330)

[PurHur]

Category

php-in-php | runtime | bootstrap

Problem

lib/AOT/runtime/superglobals_refresh.c is ~2594 lines of hand-written C for CGI superglobal refresh (#5330 phase 1 tracks the umbrella). A large subset (~900+ lines) implements $_FILES nested bracket parsing and multipart field name decoding — logic that belongs in lib/Web/ + ext/standard PHP and compiles through the normal pipeline.

Keeping multipart/bracket parsing in C blocks M5 self-host honesty and duplicates php-src semantics already partially mirrored in VM web tests (test/aot/NestedSuperglobalsAotTest.php).

php-src reference

• php/php-src main/php_variables.c — php_parse_multipart_data
• php/php-src main/rfc1867.c — multipart form parsing
• This repo: lib/Web/CgiAotDriver.php, examples/006-FileUploadWeb

Repro (C surface today)

wc -l lib/AOT/runtime/superglobals_refresh.c
grep -n 'FILES\|multipart\|bracket' lib/AOT/runtime/superglobals_refresh.c | head
./script/docker-exec.sh -- bash -lc 'source script/php-env.sh
php vendor/bin/phpunit test/aot/NestedSuperglobalsAotTest.php --filter Nested
'

Today: nested $_FILES['user']['avatar'] refresh depends on C bracket parser linked from lib/AOT/Linker.php.

Scope (PHP-in-PHP)

1. Port bracket/multipart name parsing to lib/Web/MultipartParser.php (or extend existing web helpers).
2. AOT entry: thin C trampoline calling compiled PHP refresh helper, or pure PHP init when PHP_COMPILER_AOT_USER_SCRIPT path allows.
3. Delete corresponding C functions from superglobals_refresh.c; target ≥800 lines removed (PR body must cite wc -l delta).
4. Do not grow runtime/*.c with new parser tables.

Done when

• NestedSuperglobalsAotTest green without C bracket parser bodies
• examples/006-FileUploadWeb AOT smoke still passes
• superglobals_refresh.c shrinks to env/string refresh ABI only (progress note test ProgressNoteRuntimeStandaloneTest pattern)
• Pairs parent php-in-php: CGI superglobal refresh — delete superglobals_refresh.c (~2623 lines), populate in lib/Web/ #5330 — claim either php-in-php: CGI superglobal refresh — delete superglobals_refresh.c (~2623 lines), populate in lib/Web/ #5330 or this issue, not both in one run unless scoped

Verification

make bootstrap-selfhost-helloworld
./script/docker-exec.sh -- bash -lc 'source script/php-env.sh && php vendor/bin/phpunit test/aot/NestedSuperglobalsAotTest.php'

Related

• php-in-php: CGI superglobal refresh — delete superglobals_refresh.c (~2623 lines), populate in lib/Web/ #5330 (umbrella CGI refresh migration)
• AOT: Per-request superglobal refresh in native binary (not compile-time bake) #201, P0: Fix duplicate __compiler_getenv symbol breaking AOT link #1068, Stdlib: getenv() — local_only second argument (ext/standard/basic_functions.c parity) #3710 (superglobal refresh history)