Stdlib: grapheme_strsplit() — ICU grapheme cluster split missing (ext/intl/grapheme)

[PurHur]

Category

stdlib — implementation-ready (VM landed; JIT + compliance remain)

Status (2026-06-18 audit)

grapheme_str_split() is registered on VM (ext/intl/grapheme_str_split.php, VmGrapheme). function_exists('grapheme_str_split') → true on bin/vm.php.

Still open for this issue:

• JIT/AOT lowering (currently deferred / VM-only)
• Compliance PHPT vs Zend on cluster + $length chunking
• php-src-strict: enum-case string operands must TypeError (Stdlib: grapheme_*() — enum case string operands must TypeError (ext/intl/grapheme) #5914)
• Capability matrix JIT column

Problem

grapheme_str_split() splits on grapheme cluster boundaries (combining characters stay attached). Apps expect it alongside grapheme_strlen() / grapheme_substr() (#3352).

php-src reference

• php/php-src ext/intl/grapheme/grapheme_string.c — PHP_FUNCTION(grapheme_strsplit) (PHP name: grapheme_str_split)
• php/php-src ext/intl/grapheme/grapheme_util.c — cluster iteration

Repro — existence (VM green)

./script/docker-exec.sh -- bash -lc 'source script/php-env.sh
php bin/vm.php -r "var_export(function_exists(\"grapheme_str_split\")); echo \"\\n\";"'
# bool(true)

Repro — behavior (compare Zend + VM)

Use committed maintainer script test/repro-maintainer/grapheme_str_split_missing.php:

./script/docker-exec.sh -- bash -lc 'source script/php-env.sh
php test/repro-maintainer/grapheme_str_split_missing.php
php bin/vm.php test/repro-maintainer/grapheme_str_split_missing.php'

Expected Zend + VM (php-src-strict):

Case	Output shape
"e\xCC\x81" (one grapheme)	array(1) { [0]=> string(3) … }
'abc'	three single-char graphemes
'abcdef', 2	chunks of 2 graphemes

Repro — JIT gap

./script/docker-exec.sh -- bash -lc 'source script/php-env.sh
php bin/jit.php test/repro-maintainer/grapheme_str_split_missing.php 2>&1 | head -5'
# deferred / not implemented — must not segfault

Scope (this repo)

• ext/intl/grapheme_str_split.php — VM ✅; add JitGraphemeStrSplit lowering
• ext/intl/VmGrapheme.php — shared cluster iterator (reuse for strlen/substr)
• ext/intl/Module.php — registration ✅
• test/compliance/cases/stdlib/grapheme_str_split.phpt (new)
• Enum operand guard shared with Stdlib: grapheme_*() — enum case string operands must TypeError (ext/intl/grapheme) #5914

PHP-in-PHP: ICU/cluster logic in ext/intl/ PHP, not runtime/*.c.

Done when

• test/repro-maintainer/grapheme_str_split_missing.php matches Zend on VM (all three cases)
• test/compliance/cases/stdlib/grapheme_str_split.phpt green
• JIT path implemented or matrix documents VM-only with compile-time clear error (no segfault)
• Enum-case operand covered in compliance (Stdlib: grapheme_*() — enum case string operands must TypeError (ext/intl/grapheme) #5914)
• php script/capability-matrix.php regenerated

Verification

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

Related

• Stdlib: grapheme_* — Unicode grapheme cluster string API (ext/intl/grapheme parity) #3352 (grapheme umbrella) · Stdlib: grapheme_*() — enum case string operands must TypeError (ext/intl/grapheme) #5914 · Epic: Self-host critical path — compiler compiles itself (M3→M5) #1492

[PurHur]

Cleanup bot (Jun 30): Overlaps with #11825 (phantom `function_exists('grapheme_str_split')` when ext/intl is absent). Keep #6246 for ICU cluster-split semantics + JIT/AOT; use #11825 for reference-profile gating policy. Not a duplicate — cross-linked for implementers.

[PurHur]

claim: PHP Compiler STD Lib Advanced — batch (issues #6246, grapheme JIT cluster)

[PurHur]

Merged via PR #19962 — JitGrapheme::tryStrSplitFold + JIT PHPT + AOT lint unit test. Verified: vendor/bin/phpunit --filter GraphemeStrSplitJitCompileTest OK in Docker. Runtime JIT for non-literals still deferred (same as other grapheme_*).