Stdlib: mb_internal_encoding() / mb_language() / mb_http_input() — encoding defaults (ext/mbstring parity)

[PurHur]

Category

stdlib · php-src-strict

Problem

Encoding configuration builtins from ext/mbstring are missing. Frameworks read/set internal encoding and language at bootstrap (mb_internal_encoding('UTF-8'), mb_language('uni')). Without these, multibyte pipelines cannot mirror Zend defaults.

Verified 2026-06-19: function_exists('mb_internal_encoding') → false on bin/vm.php. mb_strlen / mb_substr / mb_trim exist but cannot follow Zend encoding state until this lands.

Blocks (downstream)

Issue	Symptom without mb_internal_encoding
#9977	mb_trim($s, characters: '-') after mb_internal_encoding('UTF-8')
#9208	Full mb_ltrim/mb_rtrim UTF-8 mask repro
#6548	mb_send_mail charset negotiation

php-src reference

• php/php-src ext/mbstring/mbstring.c — PHP_FUNCTION(mb_internal_encoding), mb_language, mb_http_input
• php/php-src ext/mbstring/mbstring.stub.php
• Internal state: MBG(internal_encoding) in mbstring.c

Repro (failure today)

<?php
var_dump(function_exists('mb_internal_encoding'));
echo mb_internal_encoding(), "\n";
mb_internal_encoding('UTF-8');
echo mb_internal_encoding(), "\n";
var_dump(mb_language());
var_dump(mb_http_input());

./script/docker-exec.sh -- bash -lc 'source script/php-env.sh
php repro_mb_internal_encoding.php
php bin/vm.php repro_mb_internal_encoding.php
'

Runtime	First mb_internal_encoding()
Zend	e.g. UTF-8
This compiler	undefined function ❌

Encoding propagation probe (done-when guard)

<?php
mb_internal_encoding('UTF-8');
$s = '  üñîçø∂é  ';
var_dump(mb_strlen($s));
var_dump(mb_trim($s));

Scope (PHP-in-PHP first)

Path	Work
ext/mbstring/VmMbstring.php	Store/retrieve internal encoding + language in PHP state
ext/mbstring/mb_internal_encoding.php	Builtin registration (getter/setter)
ext/mbstring/mb_language.php, mb_http_input.php	Stub parity subset
ext/mbstring/Module.php	Wire functions
ext/mbstring/mb_strlen.php, mb_trim.php, …	Read encoding from shared state (not hard-coded UTF-8)
JIT/AOT	VM-first v1; JIT calls same PHP helpers

No new runtime/*.c — thin FFI only if ICU/libc charset probe unavoidable.

Done when

• function_exists('mb_internal_encoding') true; getter returns Zend-default encoding string
• mb_internal_encoding('UTF-8') setter succeeds and changes mb_strlen / mb_trim results on non-ASCII sample above
• mb_language() get/set subset matches Zend for uni / Japanese names
• mb_http_input() returns expected string or false per Zend when no HTTP input
• test/compliance/cases/stdlib/mb_internal_encoding.phpt green
• ./script/ci-fast.sh --filter mb_internal_encoding green
• Unblocks Stdlib: mb_trim()/mb_ltrim()/mb_rtrim() — PHP 8.4 named characters param + new builtins (ext/mbstring/mbstring.c) #9977 encoding probe (comment cross-link when green)

Verification

./script/docker-exec.sh -- bash -lc 'source script/php-env.sh && vendor/bin/phpunit --filter mb_internal_encoding'

Related

• Stdlib: mb_trim()/mb_ltrim()/mb_rtrim() — PHP 8.4 named characters param + new builtins (ext/mbstring/mbstring.c) #9977 mb_trim characters · Stdlib: mb_convert_encoding / mb_detect_encoding (ext/mbstring parity) #3075 convert/detect · Stdlib: mb_ereg*() — multibyte POSIX regex API (ext/mbstring/mbregex.c parity) #4635 mb_ereg · Epic: Self-host critical path — compiler compiles itself (M3→M5) #1492

[PurHur]

Spec expanded 2026-06-19 parity architect run: added blocking table (#9977/#9208), encoding propagation probe, VmMbstring scope, verification commands. Ready to claim.

[PurHur]

claim: PHP Compiler Ext — I/O & Crypto — starting this run

[PurHur]

Merged via #17870

What landed:

• mb_language() getter/setter with MbstringLanguageRegistry (neutral/uni/Japanese/… aliases)
• mb_http_input() type probe (G/P/C/S → false; L/I → UTF-8 list; invalid → ValueError)
• Completes encoding-metadata slice alongside existing mb_internal_encoding()

php-src: ext/mbstring/mbstring.c

Verification:

source script/php-env.sh
php bin/vm.php test/repro/maintainer_gap_mb_language_http_input.php   # ok
./vendor/bin/phpunit --filter mb_language_http_input                   # OK (2 tests, VM+JIT)
php script/check-selfhost-spine-coverage-sync.php                      # OK (4414/4414)

Remaining on #4636: mb_send_mail charset negotiation (#6548) and deeper detect-order-per-language tuning are out of scope for this slice.