Build: helper-unit dedup part B — one shared runtime prologue (common.o), re-emit the 446 committed units, move the 790 MB of objects out of git (re-#36198) (lib/AOT/HelperRuntimeCache.php, script/emit-helper-runtime-object.php)

[PurHur]

Category

Foundation: · build architecture · child of #36188 · respin of #36198 (closed by PR #36238, which delivered part A only: per-function .text.<symbol> sections + --gc-sections + -s)

Problem

PR #36238's own summary: "Part B (shared common.o runtime prologue) and helper-cache refresh remain follow-ups — committed prelinked unit.o files still use monolithic .text until re-emitted." So today: every committed unit.o still embeds ~462 duplicate runtime symbols (prelinked/helper-runtime/x86_64-linux/units/ext_ctype_CtypeJitHelper_php/unit.o = 1.65 MB for 2 exported helpers), the cache is 446 units / 790 MB in git (.git 6.6 GB), and --gc-sections can only drop bodies from objects that have per-function sections — i.e. none of the committed ones.

PHP implementation target

1. Re-emit the helper corpus with AotGcSections active (emit-helper-runtime-object.php --prelink) so the committed units carry .text.<fn> sections; record hello-world size before/after.
2. Emit the shared runtime (__value__*, __string__*, __hashtable__*, __ref__*, memory manager, GC) once as prelinked/helper-runtime/<arch>/common.o; lower units with those functions as declarations (HelperRuntimeCache::declareExternFromBitcode path); fingerprint common.o with the units; keep the class-id/ABI hazard (emit-helper-runtime-object.php:353-357) in mind.
3. Publish the per-arch cache as a release tarball fetched by core_fingerprint/llvm_identity_token; keep only manifests in git; offline fallback = local emit.

Done when

• Median unit.o < 100 KB; hello-world stripped binary < 3 MB; script/aot-smoke.sh byte-identical output; AOT sweep unchanged
• git ls-files prelinked/helper-runtime | grep -c '\.o$' = 0; fresh clone < 500 MB; cold hello-world build time unchanged after fetch

[PurHur]

Cleanup bot: duplicate re-spin — parent issue already closed; work tracked on the original. See #36142 / #36188 priority list.

[PurHur]

Cleanup bot: reopened — these are intentional follow-ups / part-B items tracked on #36188, not status slop.

[PurHur]

claim: cursor-agent — starting this run (helper-unit dedup part B: common.o infrastructure + gc-sections re-emit)

[PurHur]

claim: runtime-migration-worker — re-emit prelinked helper units with AotGcSections + auto-enable common.o linking (Part of #36246)