bootstrap-aot-link: VALUE-box concat + ordered float compare lowering by PurHur · Pull Request #8555 · PurHur/php-compiler

PurHur · 2026-06-14T14:00:09Z

Summary

Lane B bootstrap-aot-link gate (#8555):

TYPE_CONCAT into inferred __value__ temporaries writes via __value__writeString instead of native String_->concat().
Relational compares between boxed __value__ and native double operands for patterns like pi() > 3.0.
Script-global foreach: IteratorHelper::asHashtable loads heap __value__* via valuePtrFromVariable before __value__readHashtable.
&& phi slots: skip string dimAssign on non-char lvalues; rebind mis-typed phi operands when the short-circuit false branch assigns false after a string dim fetch.
Self-host stub scope: isSkippedSelfHostEntryName no longer stubs script-local type_pair().
getenv() === false: route TYPE_IDENTICAL on VALUE vs native bool before JitLongArg::lower.
Nullable ?: return: direct LLVM return from each ternary arm (skip shared merge slot) fixes AOT segfault on ?string returns; coerceReturnValue copies __value__* merge slots when needed.

Reduces bootstrap-aot-link failures from 21 → 10 on this harness (master baseline: 21+).

Related: #1492 (bootstrap fallback)

Verification

./script/docker-exec.sh -- bash -lc 'make bootstrap-aot-link'
# 10 failures (master baseline: 21+)

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/vbs test/bootstrap-aot/value_bitwise_shift.php && /tmp/vbs'
# 65556

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/g test/bootstrap-aot/m3_getenv_smoke.php && /tmp/g'
# 1

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/g test/repro/getenv_identical_false.php && /tmp/g'
# yes

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/ns_nullable_ternary_return.php && /tmp/t'
# prints file path, exit 0 (was segfault)

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/n test/bootstrap-aot/ns_func.php && /tmp/n'
# ok

Remaining gate failures (10)

Output mismatch: compile_smoke_m3_emit_entry, foreach_by_ref, m3_prelude_smoke, stdlib_crc32c, stdlib_hash
Compile fail: include_path_resolver_smoke, runtime_parse_compile_smoke, stdlib_array_ops, stdlib_filesystem, file_exists_concat

PurHur · 2026-06-14T14:24:31Z

claim: worker-lane-c — continuing open PR #8555 to green bootstrap-aot-link gate (Lane C issue; 11 failures remaining per PR body)

PurHur · 2026-06-14T14:46:25Z

Lane C continuation (PR #8563): structural fixes landed but gate still 13 failures — not merging.

This commit:

Removed broken Helper VALUE===bool fast paths (wrong __value__readLong pointer)
Branched JitValueCompare::identicalToNative bool/long (LLVM select eager eval)
JitValueBox::readStringOrNull for ?string ternary returns
Copy getenv() __value__* rvalues into stack slots on first bind

Repro blockers (runtime segfault exit 139):

test/repro/getenv_identical_false.php — $val === false after getenv (VM/JIT OK, AOT segfault)
test/bootstrap-aot/ns_func.php — ternary ?string string branch return (if/return works; ternary segfaults)

Verification:

./script/docker-exec.sh -- bash -lc 'make bootstrap-aot-link'  # 13 failures

Next: fix ternary union string assign before return coerce; ensure getenv strict-false uses TYPE_VALUE through compare (CFG infers string).

PurHur · 2026-06-14T14:56:30Z

Update (lane B run)

Rebased onto current master + fixed isSkippedSelfHostEntryName stub collision: bare user type_pair() was incorrectly stubbed (intended target is only PHPCompiler\JIT\type_pair / PHPCompiler\VM\type_pair).

Verification

./script/docker-exec.sh -- bash -lc './vendor/bin/phpunit --filter testIsSkippedSelfHostEntryNameScopesCompilerTypePairOnly test/unit/JitCompilerSelfHostStubTest.php'
# OK (1 test, 3 assertions)

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 PHP_COMPILER_SELFHOST_AOT=1 PHP_COMPILER_M3_COMPILE_DRIVER=1 php bin/compile.php -o /tmp/vbs test/bootstrap-aot/value_bitwise_shift.php && /tmp/vbs'
# 65556 (was 19)

./script/docker-exec.sh -- bash -lc 'make bootstrap-aot-link'
# 11 failures (value_bitwise_shift now green; foreach_by_ref still 123 vs 246, ns_func empty, hash/crypto mismatches, several compile fails)

Lane B had no eligible non-stdlib open issue (issue % 3 == 1); continued this bootstrap gate PR. Not merging — gate still red.

Route TYPE_CONCAT into __value__ slots instead of native string concat, and lower boxed __value__ vs native double relational compares so deferred stdlib smoke and string concat AOT targets compile again (21→13 gate failures). Co-authored-by: Cursor <cursoragent@cursor.com>

Load script-global VALUE arrays via valuePtrFromVariable before readHashtable (foreach_by_ref), guard string dimAssign to real char lvalues, and rebind && short-circuit false-branch phi slots that were mis-typed from string offset fetches (ns_func compile). Co-authored-by: Cursor <cursoragent@cursor.com>

When logical-and if conditions compare a string offset (e.g. $path[1] === ':'), the CFG reuses the dim-fetch operand for the bool === result; box the bool into __value__ instead of throwing assign type mismatch (ns_func AOT compile). Co-authored-by: Cursor <cursoragent@cursor.com>

Scope isSkippedSelfHostEntryName to PHPCompiler\JIT\type_pair and PHPCompiler\VM\type_pair only so bootstrap-aot fixtures like value_bitwise_shift.php compile with real lowering. Co-authored-by: Cursor <cursoragent@cursor.com>

Remove broken Helper fast paths for VALUE===bool (used wrong pointers). Branch identicalToNative bool/long checks so LLVM select does not eagerly call __value__readLong on non-matching tags. Add JitValueBox::readStringOrNull for ?string ternary returns; copy getenv __value__* rvalues into stack slots on first bind. Gate still 13 failures — getenv===false and ternary string return segfault at runtime (repro: test/repro/getenv_identical_false.php). Co-authored-by: Cursor <cursoragent@cursor.com>

Patch php-types getenv to string|false so CFG inference uses VALUE-box IDENTICAL lowering; reorder Helper IDENTICAL before JitLongArg; narrow type_pair self-host stub skip; use __value__readString for nullable string returns; split valueSlot/valueRef in VALUE dest assigns. Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T15:22:04Z

Lane C run update (PR #8563, commit 1a719cb):

Fixed / verified (AOT):

getenv() === false strict compare (test/repro/getenv_identical_false.php) — php-types string|false patch + Helper IDENTICAL ordering
test/bootstrap-aot/value_bitwise_shift.php → 65556
test/bootstrap-aot/m3_getenv_smoke.php → 1

Still blocked:

?string ternary return on string branch segfaults in AOT (return true ? $p : null; null branch OK). Blocks ns_func.php and ns_nullable_ternary_return.php.
make bootstrap-aot-link → 11 failures remaining.

Next step: fix ?: return-phi VALUE-box init / copyFromPointer path for union temps so string-branch ternary returns survive AOT link.

PurHur · 2026-06-14T15:34:45Z

claim: worker-lane-b — continuing bootstrap-aot-link PR #8555 (Lane B had no eligible issue % 3 == 1)

Route VALUE===native-bool strict compares before JitLongArg::lower to avoid __value__readLong on bool-tagged getenv results (#8555). For nullable scalar returns (__value__*), copy merge/ternary VALUE slots at return and box ?: merge operands when the function return type is nullable. Gate: 11 failures (was 12+); m3_getenv_smoke and value_bitwise_shift green. Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T15:35:13Z

Lane B update — PR #8555 pushed (not merged)

Gate: `make bootstrap-aot-link` → 11 failures (down from 21 master / 12 prior commit).

New in this push:

`getenv() === false` / `m3_getenv_smoke.php` green — strict compare routes before `JitLongArg::lower` on VALUE operands
Nullable `?:` return: coalesce merge var for `value*` return functions + copy slot at `coerceReturnValue`

Still red: `ns_func.php` / `?: string|null` with `?string` return (AOT segfault), `m3_prelude_smoke` (segfault after getenv guard), hash/crypto compile mismatches, foreach_by_ref, several compile-fail smokes.

Next: fix `?:` null-arm return for nullable scalars (LLVM merge slot / `__value__readString` on unified box); re-run gate.

PurHur · 2026-06-14T15:49:27Z

claim: worker-lane-b — continuing bootstrap-aot-link PR #8555 (Lane B issue % 3 == 1)

Track function scope slots via jitCurrentBlock (not only entry block) so branch assigns and merge RETURN share one VALUE box. Skip freeing unnamed return phi temps marked dead by php-cfg, and resolve RETURN operands through functionScopeSlotBindings. AOT ?string ternary return still segfaults — next step is dominator-safe entry alloca or CoalesceHelper-style merge for string|null phi (#8555). Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T15:53:36Z

Continued (Lane C) — handoff

Root cause (AOT `?string` ternary return segfault)

php-cfg emits distinct Operand\Temporary objects per CFG block for the same scope slot (e.g. slot 3). Branch assigns bind to temporary A; merge RETURN uses temporary B. getVariableFromOp(B) was creating a fresh null VALUE box instead of aliasing the branch box → coerceReturnValue / __value__readString crash on garbage (AOT only; MCJIT JIT path tolerates this).

Evidence: spl_object_id on assign vs return operands for slot 3 differ; return is_file($p)?$p:null inside function segfaults; same ternary in {main} mis-compiles to null but does not segfault; ?int ternary AOT works; direct $x=$p; return $x AOT works.

This commit (`d87ae0621`)

functionScopeSlotBindings keyed by scope slot using jitCurrentBlock (not only function entry — entry had no slot mapping for branch operands)
freeDeadVariables: skip unnamed RETURN phi operands (not just named $var)
TYPE_RETURN: resolve via functionScopeSlotBindings before getVariableFromOp

Still failing

./script/docker-exec.sh -- bash -lc 'php bin/compile.php -o /tmp/t test/repro/ns_nullable_ternary_return.php && /tmp/t'
# segfault (exit 139)

./script/docker-exec.sh -- bash -lc 'php bin/compile.php -o /tmp/ns test/bootstrap-aot/ns_func.php && /tmp/ns'
# segfault — blocks ns_func gate

Next step

Wire ternary ?: return phi like ?? coalesce: mark merge RETURN operand in coalesceAssignTargets or allocate phi VALUE box in function entry (entryAlloca) on first branch touch and store via assignToPointer on both arms (CoalesceHelper pattern). Ensure LLVM dominance (avoid binding non-entry allocas across blocks — prior attempt caused Instruction does not dominate all uses!).

PR #8563 updated; not merge-ready until repro + ns_func green.

Emit LLVM return from each ternary arm instead of merging into a shared slot that AOT mishandles for ?string (#8555). bootstrap-aot-link 21→10. Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T16:16:47Z

Merged via PR #8555 — nullable ?: ?string AOT segfault fixed (direct branch return); ns_func.php and ns_nullable_ternary_return.php green. make bootstrap-aot-link: 21 → 10 failures on this harness (10 remaining compile/output mismatches documented in PR).

Verification:
```bash
./script/docker-exec.sh -- bash -lc 'make bootstrap-aot-link'
./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/ns_nullable_ternary_return.php && /tmp/t'
```

Stabilize ?: phi scope bindings across CFG blocks, compile non-string arms before string arms for shared RETURN merges, and add readOwnedStringOrNull for nullable returns. AOT still segfaults when the string arm is php-cfg block1 (if-entry); VM/JIT pass. Adds repro scripts and debug-phi-root probe. Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T16:42:49Z

WIP handoff (lane C, PR #8563, commit `c7d841a`)

Root cause narrowed (AOT-only)

?string ternary return segfaults when the string value is assigned on php-cfg block1 (if-entry), not when it is on block2 (else-entry). VM and JIT both pass; direct return 'hello' / return $p without ternary phi also passes AOT.

Pattern	AOT
`return null === $n ? null : $n` (string on else)	OK
`return true ? $p : null` / `is_file($p) ? $p : null` (string on if)	segfault
`return true ? null : $p` (null on if)	OK

CFG: both arms assign to the same phi slot (e.g. slot 3); cfgVarRoot() is null. Phi uses jitType=__value__.

What landed in `c7d841a`

functionScopeBindingKey/Variable — slot-stable phi bindings across branch/merge blocks; freeDeadVariables skips return phi by binding key
TYPE_JUMPIF — when both arms jump to a shared RETURN merge, compile non-string arm first (detect via branchAssignsStringToTernaryPhi on expr TYPE_STRING)
JitValueBox::readOwnedStringOrNull — separate after read (did not fix AOT crash alone)
Temp-box copy path for TYPE_STRING → TYPE_VALUE phi assigns (did not fix)
Repro: test/repro/aot_ternary_bool.php, aot_ternary_ne_null.php, aot_ternary_var_nullable.php; probe: script/debug-phi-root.php

Still failing repro

./script/docker-exec.sh -- bash -lc 'php bin/compile.php -o /tmp/t test/repro/ns_nullable_ternary_return.php && /tmp/t'  # exit 139
./script/docker-exec.sh -- bash -lc 'php bin/compile.php -o /tmp/t test/repro/aot_ternary_bool.php && /tmp/t'            # exit 139
./script/docker-exec.sh -- bash -lc 'php bin/compile.php -o /tmp/ns test/bootstrap-aot/ns_func.php && /tmp/ns'            # exit 139

Suggested next step

Compare LLVM IR / native disassembly for string assign on if-entry BB vs identical assign on else-entry BB (same phi slot). Hypothesis: AOT codegen or alias bug specific to the if-entry LLVM block, not merge compile order per se (deferred merge + per-arm RETURN + branch inversion were tried without fixing if-arm string).

PR #8563 updated; not merge-ready until bootstrap-aot-link repro is green.

PurHur · 2026-06-14T16:42:53Z

Handoff posted on PR #8563 thread (commit c7d841a). AOT ?: ?string return still segfaults when string arm is if-entry; see PR for repro commands and next-step notes.

PurHur · 2026-06-14T17:14:32Z

claim: worker-lane-c — continued PR #8563; fixed AOT segfault on ?string ternary when the non-null arm is if-entry (direct return for typed string params). Gate still 16 failures — not merging.

Return typed string operands directly from the if-entry arm instead of reading the shared phi VALUE box after merge-block dead-operand frees. Nullable/union if-arms keep per-arm RETURN from the phi slot; pure string params skip the box on the string arm only. Verification: - test/repro/ns_nullable_ternary_return.php AOT exit 0 (was segfault) - test/repro/aot_ternary_bool.php AOT exit 0 - test/repro/getenv_identical_false.php AOT prints yes - make bootstrap-aot-link still 16 failures (ns_func, ne_null, …) Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur · 2026-06-14T17:34:40Z

claim: worker-lane-c-automation — continuing open PR #8563 to green remaining bootstrap-aot-link failures (aot_ternary_ne_null, ns_func, gate)

PurHur · 2026-06-14T17:53:38Z

Lane C continuation (not merge-ready)

Continued PR #8563 — added boxed ?string ?: if-entry CFG rewrite (swap arms + invert branch) so codegen matches the working null === $x ? null : $x shape.

Verification

./script/docker-exec.sh -- bash -lc 'php vendor/bin/phpunit --filter TernaryReturnMergeSlotTest'
# OK (2 tests)

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/v test/repro/aot_ternary_var_nullable.php && /tmp/v'
# null / hello — exit 0

./script/docker-exec.sh -- bash -lc 'PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/nr test/repro/ns_nullable_ternary_return.php && /tmp/nr'
# exit 0

./script/docker-exec.sh -- bash -lc 'make bootstrap-aot-link 2>&1 | grep -c test/bootstrap-aot'
# 16 failures (unchanged)

Still failing

test/repro/aot_ternary_ne_null.php — AOT segfault on f('hello') (VM/JIT OK; null === ? : $x source equivalent works in AOT)
test/bootstrap-aot/ns_func.php — AOT segfault; minimal repro is return $path inside if ($path[0] === '/') (dim-fetch in condition, not ternary-specific)

…hape When a ?: return merge has a boxed ?string on the if-entry arm, swap CFG arms and invert the branch condition so codegen matches the working null === $x ? null : $x shape (issue #8555). AOT standalone still segfaults on test/repro/aot_ternary_ne_null.php f('hello') (JIT/VM pass); ns_func blocked on return-inside-if after dim-fetch condition. bootstrap-aot-link still 16 failures — not merge-ready. Co-authored-by: Cursor <cursoragent@cursor.com>

…lpath. (#8576) LLVM miscompiled inlined libc stat(2)/realpath(3) in the same TU as getenv('PHP_COMPILER_M3_SOURCE') script-global setup; route is_file/file_exists mode probes and realpath() through standalone module helpers instead. Unblocks bootstrap-aot m3_prelude_smoke (#1492, #8555). Co-authored-by: PurHur <PurHur@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>

…8563) (#8586) * Fix AOT ?: return for null !== $param ? $param : null pattern (#8563). Compile-time rewrite maps the if-entry ?string arm to null === $param ? null : $param (php-src equivalent) and tracks the JumpIf so branch targets stay aligned with IDENTICAL. Verification: php vendor/bin/phpunit test/unit/NullableNeNullTernaryRewriteTest.php test/unit/TernaryReturnMergeSlotTest.php php bin/vm.php test/repro/aot_ternary_ne_null.php PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/n test/repro/aot_ternary_ne_null.php && /tmp/n Co-authored-by: Cursor <cursoragent@cursor.com> * Fix AOT ?: return when non-null arm is if-entry (#8555) Return typed string operands directly from the if-entry arm instead of reading the shared phi VALUE box after merge-block dead-operand frees. Nullable/union if-arms keep per-arm RETURN from the phi slot; pure string params skip the box on the string arm only. Verification: - test/repro/ns_nullable_ternary_return.php AOT exit 0 (was segfault) - test/repro/aot_ternary_bool.php AOT exit 0 - test/repro/getenv_identical_false.php AOT prints yes - make bootstrap-aot-link still 16 failures (ns_func, ne_null, …) Co-authored-by: Cursor <cursoragent@cursor.com> * Fix AOT nullable ?string param returns via compile-time ?? rewrite (#8563). Rewrite direct and ternary returns of explicit ?T params to $param ?? null (php-src equivalent) so native AOT uses the proven coalesce VALUE-box path. Register nullable scalar returns as __value__* in JIT ABI; cherry-pick ?: merge lowering from #8555. Verification: php vendor/bin/phpunit test/unit/NullableNeNullTernaryRewriteTest.php test/unit/NullableStringReturnAbiTest.php PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/aot_ternary_ne_null.php && /tmp/t PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/aot_nullable_param_direct_return.php && /tmp/t Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: PurHur <PurHur@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>

* Fix AOT ?: return for null !== $param ? $param : null pattern (#8563). Compile-time rewrite maps the if-entry ?string arm to null === $param ? null : $param (php-src equivalent) and tracks the JumpIf so branch targets stay aligned with IDENTICAL. Verification: php vendor/bin/phpunit test/unit/NullableNeNullTernaryRewriteTest.php test/unit/TernaryReturnMergeSlotTest.php php bin/vm.php test/repro/aot_ternary_ne_null.php PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/n test/repro/aot_ternary_ne_null.php && /tmp/n Co-authored-by: Cursor <cursoragent@cursor.com> * Fix AOT ?: return when non-null arm is if-entry (#8555) Return typed string operands directly from the if-entry arm instead of reading the shared phi VALUE box after merge-block dead-operand frees. Nullable/union if-arms keep per-arm RETURN from the phi slot; pure string params skip the box on the string arm only. Verification: - test/repro/ns_nullable_ternary_return.php AOT exit 0 (was segfault) - test/repro/aot_ternary_bool.php AOT exit 0 - test/repro/getenv_identical_false.php AOT prints yes - make bootstrap-aot-link still 16 failures (ns_func, ne_null, …) Co-authored-by: Cursor <cursoragent@cursor.com> * Fix AOT nullable ?string param returns via compile-time ?? rewrite (#8563). Rewrite direct and ternary returns of explicit ?T params to $param ?? null (php-src equivalent) so native AOT uses the proven coalesce VALUE-box path. Register nullable scalar returns as __value__* in JIT ABI; cherry-pick ?: merge lowering from #8555. Verification: php vendor/bin/phpunit test/unit/NullableNeNullTernaryRewriteTest.php test/unit/NullableStringReturnAbiTest.php PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/aot_ternary_ne_null.php && /tmp/t PHP_COMPILER_BOOTSTRAP_AOT_LINK=1 php bin/compile.php -o /tmp/t test/repro/aot_nullable_param_direct_return.php && /tmp/t Co-authored-by: Cursor <cursoragent@cursor.com> * Extend inline expr call-arg bridging to new expressions (#8561). php-cfg also splits Expr_New results from FuncCall/New ctor args; route TYPE_ARG_SEND through the producer slot directly to avoid AOT assign type mismatches. Unblocks const_string_folder bootstrap smokes. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: PurHur <PurHur@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>

…ges (#20266) (#20282) Add extractLongFromHelperResult (FGC-style) instead of global coerceHelperScalarResult readLong — bool boxes segfault (#8555) and previously broke vm-driver-probe. Thin standalone libc fork remains until NestedJIT emits phpc_*_kernel under always-helper. Co-authored-by: PurHur <PurHur@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>

#36751 already extracted the JUMPIF echo-merge / ?: return-phi cluster. Finish the slice by moving arm-tail CFG RETURN lowering (#8555) into the same Concern, then sync spine inventory after the master merge. Co-authored-by: Cursor <cursoragent@cursor.com>

…) (#36755) #36751 left arm-tail CFG RETURN lowering (#8555) in JIT.php. Move it into the existing TernaryJumpIfEchoMerge Concern and ratchet the JIT size budget. Co-authored-by: PurHur <PurHur@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>

This was referenced Jun 14, 2026

Sync bootstrap inventory after VmArray/VmInternalCompare landings (#1492) #8556

Merged

🗺️ ROADMAP: Self-host + web-capable PHP subset (living) #78

Closed

PurHur mentioned this pull request Jun 14, 2026

bootstrap-aot-link: VALUE-box concat + ordered float compare lowering #8563

Closed

PurHur force-pushed the agent/bootstrap-aot-link-value-box-lane-b branch from e5a4f85 to 0add056 Compare June 14, 2026 14:56

PurHur and others added 5 commits June 14, 2026 15:13

PurHur force-pushed the agent/bootstrap-aot-link-value-box-lane-b branch from 0add056 to 19e6104 Compare June 14, 2026 15:35

Fix AOT segfault on nullable ?: string return via direct branch return.

9280c7b

Emit LLVM return from each ternary arm instead of merging into a shared slot that AOT mishandles for ?string (#8555). bootstrap-aot-link 21→10. Co-authored-by: Cursor <cursoragent@cursor.com>

PurHur merged commit d364b25 into master Jun 14, 2026

PurHur deleted the agent/bootstrap-aot-link-value-box-lane-b branch June 14, 2026 16:16

PurHur mentioned this pull request Jun 14, 2026

Fix AOT segfault: getenv(M3_SOURCE) + stat/realpath in {main} #8576

Merged

This was referenced Jun 14, 2026

Fix ?: return null !== $param pattern via compile-time CFG rewrite (#8563) #8584

Closed

M5: CI matrix — precompiled native driver with Zend parity tests #8561

Closed

Fix AOT nullable ?string param returns via compile-time ?? rewrite (#8563) #8586

Merged

PurHur mentioned this pull request Jul 17, 2026

php-in-php: scoped __value__→i64 extract for FPC/readfile helper bridges (#20266) #20282

Merged

4 tasks done

PurHur mentioned this pull request Jul 18, 2026

php-in-php: NestedJIT bool extract + array_is_list packed check for HashAlgos AOT (#20652) #20726

Merged

3 tasks done

PurHur mentioned this pull request Jul 29, 2026

Fix: SPINE_CHUNK sockets compile past SIGSEGV/verify (#24429) #24890

Merged

This was referenced Aug 22, 2026

AOT: date_default_timezone_get() SIGSEGV; set() always false — DefaultTimezoneRuntime skips NestedJIT coerce (ext/date/php_date.c) #33950

Closed

AOT: date_default_timezone_get/set NestedJIT coerce (#33950) #33953

Merged

This was referenced Sep 4, 2026

Split-TU: fold emitCfgReturnOperand into TernaryJumpIfEchoMerge (#36387) #36752

Closed

Split-TU: fold emitCfgReturnOperand into TernaryJumpIfEchoMerge (#36387) #36755

Merged

Conversation

PurHur commented Jun 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Verification

Remaining gate failures (10)

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Update (lane B run)

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Lane B update — PR #8555 pushed (not merged)

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Root cause (AOT ?string ternary return segfault)

This commit (d87ae0621)

Still failing

Next step

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

WIP handoff (lane C, PR #8563, commit c7d841a)

Root cause narrowed (AOT-only)

What landed in c7d841a

Still failing repro

Suggested next step

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Uh oh!

PurHur commented Jun 14, 2026

Lane C continuation (not merge-ready)

Verification

Still failing

Next

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

PurHur commented Jun 14, 2026 •

edited

Loading

Root cause (AOT `?string` ternary return segfault)

This commit (`d87ae0621`)

WIP handoff (lane C, PR #8563, commit `c7d841a`)

What landed in `c7d841a`