Runtime shrink: lazy HashTable strtol decl (peer #35626 / #31988)

[PurHur]

Summary

• Stop declaring libc strtol(3) on every HashTable::implement() full load — only tryLookupPackedIntFromStringKey() (Zend numeric-string → int key fallback, php-src zend_hash.c / Stdlib: array numeric-string keys — "1" matches int key 1 (Zend zend_hash.c parity) #3679) calls ensureLibcStrtol() before lookupFunction('strtol').
• Same lazy pattern as php-in-php: lazy strcoll/strnatcmp from HashTable::implement (#35626) #35626 (strcoll/strnatcmp) and AOT DOM: fix DTD isId regression and setIdAttributeNode(null) LLVM verify (#34821, #33758) #35648 (undef-key trigger_error): thin hello-world AOT skips the decl during HashTable init, avoiding the Fix: AOT link fails for EVERY binary — duplicate memset declaration #31894 / Release v1.1.0: fix cold-build-check compile_failed on hello world (LLVM Module.php line 180; re-#24302) #32122 *.1 mint class on cold-build.

Root cause

HashTable::implement() eagerly called ensureLibcStrtol() at entry even though strtol is only used on the numeric-string key lookup path. Every compile (including hello-world) paid for a libc extern that most scripts never touch.

Verification

./script/aot-smoke.sh
# aot-smoke: 8 passed, 0 failed

./script/phpunit.sh --filter 'HashTableLazyStrtolRuntimeShrinkTest|StrtolRuntimeShrinkTest'
# OK (7 tests, 32 assertions)

Not run: ./script/release-readiness.sh (pillar 1 already green on master this run), full differential sweep, ci-fast.sh (006 benchmark README drift pre-existing on master).

Made with Cursor