Bun parity: remaining open items to reach 100% test-suite pass

Native Bake full-suite checkpoint: source ccecdcf9b and 1,185 Git-verified evidence artifacts. All 24 unchanged Bake files pass 177 cases with zero failures/zero skips and two original TODOs; native output, exact identities and source inventory validate. Genuine frozen Home locks replace the empty React marker. The suite budget preserves per-test deadlines; spawned fixture cwd and production source-map retrieval are repaired. Canonical graph identities fix changed-dependency path aliases. Coordinator controls retain 29 complete journals; development/HMR/production pass. The fresh 157-file regression has 396 passing cases, 142 process checks, zero failures, 37 case skips and three file skips. Build, validators, ABI, lint and corpus integrity pass. The initial 175-pass/two-failure run remains retained. #902 is complete for this tranche; #901 keeps optimized native/platform gates and #564 keeps Linux TODO coverage. TODOs/skips receive zero pass credit. The full Bun feature/suite/all-platform objective remains open.

Tracking issue for the remaining work to reach 100% parity with Bun's official test suite (~/Code/bun; current native artifact source SHA 4982b91e3702094330f3be3883354c52b8c01323, original historical pin fd0b6f1a27), runnable here via home test. Each item has the root cause traced during investigation. We tackle them top-to-bottom; check off as landed (small fix:/feat: commits, pushed).

Current main checkpoint (2026-10-13)

Native console methods, buffered replay and cleared-realm lifetime (2026-10-13) landed in source 56df99d7f, merged integration be4364972, and verified evidence de6aea239. Canonical SDK records provide count/reset, typed timeLog/timeEnd and per-session buffered replay with actual stacks/timestamps. Owner attachment differs from VM console state; Runtime.disable suppresses notifications while preserving connected state. groupEnd/clear arguments match the native contract. Clearing the buffer and remote group now retires idle realm injected-script roots; actual still-bound handles, pending calls and paused debugger state prevent retirement. Unrelated object handles and an in-flight async call remain usable. Last frontend detach retires injected-script state, and the non-owning console registry retires across finalization threads. Both source and merged builds pass 56/56 steps; fresh merged validation passes all nine focused commands, six broad drivers, 33 native groups, seven Node/seven Home realm controls, 14 ABI tests/453 assertions and 32 report-validator tests. All 64 report outcomes match expected counts/errors/exits, including twelve negatives. Twenty unchanged originals retain 53 passed cases, one unchanged unsupported-method expectation failure and 17 passed process checks, with zero skips/TODO/filter/retries; #886 receives no pass credit. Failed Node oracle assumptions, the Home post-clear lifetime failure and resource stops remain immutable; temporary tracing is removed. Full schema, independent remote-object/session ownership, evaluated-realm retirement outside explicit console clearing, module/export/callback/owner-shutdown acceptance and full suite/platform validation remain open under #888/#428/#885/#66.

Native profiler sharing at 5cd57e8ea with verified evidence: CLI and inspector owners consume actual JSC samples through detached native records, preserving CLI JSON/Markdown history and independent repeated session windows. Shortest active owner interval is selected; session stop/disconnect preserves CLI profiling and active-session process exit exports successfully. Debug 56/56, all 28 native groups, 14 ABI controls/453 assertions and 32 report-validator controls pass. Twenty unchanged originals retain 53 passed cases, one failed unsupported Runtime.evaluate expectation and 17 process checks; no skips/TODO/filter/retries and no #886 pass credit. Evidence preserves the first inlined-workload assertion failure and 64 replayed report outcomes including 12 negatives; earlier eight resource stops remain in investigation evidence. Full Bun parity, complete realms, worker-to-main inspector attachment, coverage, precise cadence measurement, all ownership schedules, Release/ASAN/all-platform remain open under #428/#66.

Historical publication checkpoint (2026-09-20)

Native phase/context work landed in 64481fe09 with evidence publication b83c54f4d. Vendor checkout and install/build became separately verified phases, with revision checks before hooks and live remap/Docker settings carried into children. The build passed 23/23 steps and 21/21 harness tests, plus launcher, service/lifetime and validator controls. Fresh pinned Elysia setup matched all 283 expected source hashes; all 129 selected original files completed once with 1,516 registered passes, zero failures/skips/TODOs, retaining the one original WebSocket exclusion. This checkpoint did not claim complete primary execution, Linux Docker verification, or full parity. Boundary and API · Evidence

Previous publication / remaining work (2026-09-08)

Full logical Bun parity remains incomplete. Published to origin/main: native service/preparation source a2e2863fc, verified evidence c9fc9bf63. Home SHA256 91d6b1024f9d4c02c6b7f9089b476d8a0646294f4ed7392e5e3ce549e52db763.

Native service lifetime and explicit vendor preparation are implemented. home test --bun-corpus-remap launches the original installed remap service with its original arguments and 5,000 ms startup deadline, drains output throughout its lifetime, and owns shutdown until parent stdin closes or the service exits unexpectedly. home test --bun-corpus-prepare-vendor elysia selects and journals clone/fetch/checkout/revision/install/build, retains separate Git/Home executable identities, and preserves the original 180-second Git/install and 60-second build deadlines. A failed install leaves build visibly unstarted. These operations earn zero registered-case or passing-process credit.

23/23 native build steps and 21/21 harness tests pass, along with four service lifetime units, four journal units, 18 validator controls and five private CLI controls. The private CLI evidence retains invalid readiness, the five-second deadline after output EOF, unexpected service exit with parent stdin open, install failure and build failure. The real pinned remap service passes five HTTP protocol checks through the native command, with complete captures and unchanged dependency hashes. Fresh native Elysia preparation completes all seven steps in 49.8 seconds; its 283 resulting source hashes match the earlier independently verified setup/lockfile normalization. Formatting, targeted Pickier and 50 pinned setup-input checks pass.

The latest complete Elysia run is not green: all 129 selected files complete once with 1,515 registered passes, one failure, zero skips/TODOs. handle distinct union in test/validator/response.test.ts calls .listen(3000); the unrelated PredictHQ development server (PID 28404 at diagnosis) owns that port. Pinned Bun fails the same unchanged case in a separate one-file diagnostic (27 passes, one failure). The listener and original tests were not changed, and no retry or alternate port was used to obtain a pass. Both full-run source integrity and the original selection match. The original WebSocket exclusion remains separate. The first diagnostic verifier expected the same error literal; a separate retained-capture analysis confirms the conflict without rerunning either runtime. A private error-object comparison additionally found message/errno/address/port differences, now tracked in #726; Home's existing preservation of real OS failures must be retained while auditing compatibility.

572 committed compressed artifacts have verified decompressed hashes. Restored journals reproduce successful service/preparation, failed private controls, the failed complete Elysia run and failed pinned-Bun diagnostic separately. The first build was stopped below the unchanged 1 GB critical disk floor after macOS allocated more swap; verification did not run. A cleanup preflight and refused admission are retained too. Corrected cleanup preserved protected binaries and linked native inputs; the completed rebuild required 5 GB free to start. Removed 6,083,408,581 logical bytes of generated caches/dependencies, old vendor data and unused build outputs. Source and other active work are preserved.

Still open: connect these operations with the original CI ordering and selection gates, Docker coordination, live service settings and default unified coordinator (#724); free the original fixed-port prerequisite before claiming a successful new Elysia run; resolve the missing optional legacy DuckDB and earlier lifecycle failures (#725); audit bind-error compatibility (#726); execute the whole applicable primary/vendor suite and finish remaining runtime features, native Linux/Windows/ASAN verification and independent native ownership. Earlier root/test setup process success and the separate 18-case N-API DuckDB run remain evidence on their recorded earlier binary; they are not new execution on this one.

API and scope · Native service/preparation evidence · Earlier native setup evidence

Earlier prepared Elysia execution checkpoint

Published to origin/main: implementation 9d6815afa, evidence 176f685f7. Home SHA256 19e41223c7fc2d2dcda763470fcc84570900db08b4d547340bdae5286d42bc98.

Complete pinned-selected Elysia execution: 129/129 files completed, 1,516 registered cases passed, zero failed, zero registered skips, zero TODOs. The run used no positional filter or start/limit and one attempt with original per-file deadlines. The selection matches pinned Bun exactly, including the separately recorded original ws/connection.test.ts exclusion; that excluded file earns no passing or skipped-case credit. All 129 captures reached EOF. Raw output and JUnit hashes, registered cases, selection completeness, all 283 post-install vendor source hashes and the Home executable hash validate.

The production home test --bun-corpus-prepared-vendor elysia command verifies Git HEAD against the manifest tag and executes each selected vendor file in its own project's cwd with ordinary config discovery, forced vendor test mode, correct validation paths and serial environment. Ordinary failures remain recorded and later files continue. This command explicitly requires prior checkout/install/build and journals setup_performed: false. The preceding Home 716f39bd… build performed the independently verified native install/build; the new 19e41223… build executed the suite. This is not a fresh end-to-end CI setup claim.

23/23 build steps, 17/17 native harness tests and 11/11 validator controls pass. Private controls exercise project config, custom test paths/extensions, serial environment and a real assertion failure followed by a successful file. Targeted Pickier and whitespace checks pass. 409 committed compressed artifacts have verified decompressed hashes; restoring the archived reports reproduces all 129 complete outcomes and 1,516 passing cases. Cleanup removed another 371,467,160 bytes of obsolete owned generated objects, preserving source work.

API and verification boundary · Complete execution evidence

Earlier outcome and lifetime batch at 61670c72b (#722/#723): 23/23 build steps and 12/12 native corpus harness tests pass, plus 2 child-ownership units, 3 journal units, 9 validator units and targeted Pickier. The original per-file deadline now covers a child that stays alive after both output pipes reach EOF. Actual libc close through Home FFI reproduced the pre-fix 30-second supervisor failure; the corrected native runner times out at the unchanged 20-second Node deadline and preserves early successful exit. Child status remains unreaped until final cleanup. The wait helper also cross-compiles for x86_64 Linux/musl and Windows/GNU; native Linux/Windows execution remains unverified, so #723 stays open for that work. Schema 2 records verified stream completeness; legacy schema 1 remains readable with completeness marked unknown.

That earlier six-file original batch validates with 402 passing cases, 1 original platform skip, 10 TODOs, one successful Node script, one verified expected-failure process check and one comment-only file. All six captures reached EOF. The original negative fixture retains its one-failure JUnit case and contributes no passing test credit. Every selected original source/snapshot hash and the executable identity remain unchanged. Private controls verify mixed outcomes, invalid negative contracts, overwrite refusal, both open-pipe and actual closed-pipe timeouts, SIGTERM, interruption, and failed writes. Original failures and first unsuccessful controls remain retained. Final evidence; baseline and remaining scope.

#724 remains the full-suite prerequisite: selection/API and Elysia setup are now verified as described above; native host discovery, complete root/test installation and registry/service setup, vendor execution, and complete applicable-suite outcomes remain unfinished. The pre-exclusion discovery inventory alone is not the complete CI suite. Excluded, skipped, TODO, unstarted and unsupported files never count as passing implementation coverage.

Previous native launch batch (#703): 23/23 native/harness build steps and 9/9 harness tests pass. Five focused units pass. All 19,016 launch-rule selections match pinned Bun over 4,754 discovered files and four CI/local/ASAN contexts; these are rule comparisons, not passing corpus files. Production capture now applies original ordinary/integration/Node/script deadlines, CI GC/integrity/transpiler-cache settings, validation-list rules, fresh per-file storage and real Home command aliases. Real cache/child/cleanup regressions verify the corrected allocation ownership and initially empty install cache. Launch baseline, 64 final artifacts.

Prior grouped launch verification: seven native private regression files pass. Unique selected original cases are Home 71 pass / 1 fail; pinned Bun 72 pass: HTML 45, link/binlink 11, audit 16. Two additional original Node files complete real process checks, and one original comment-only file contributes no passing cases. All selected original sources and snapshots remain unchanged. The generated fresh-cache control passes in both runtimes. The first build's three allocator crashes and Home-only cache-control failure are retained, with their fixes independently verified.

#721 retains the one failing original audit snapshot assertion. Audit remains Home 15/16 versus pinned Bun 16/16. The only differing line expects express › connect › csurf › csrf › uid-safe › base64-url. Its pinned fixture resolves that UID package to csrf/uid-safe@2.1.4, which depends on random-bytes and has no base64-url edge. Home reports the actual express › connect › express-session › uid-safe@2.0.0 › base64-url@1.2.1 path. No nonexistent edge or passing assertion was fabricated. #720/#721 remain open; native protocol/authentication, exact production reachability, affected-version/workspace paths, cycles and lock preservation are covered by the passing native audit regression.

#722 retains the remaining full-inventory execution and outcome-verification requirement, following the verified storage/JUnit/skip/TODO implementation above. The inventory contains 4,754 pre-exclusion entries (1,863 test-runner entries and 2,891 scripts); a completed full-suite execution is still missing. Synthetic corpus rewriting/bootstrap fallback is removed. Full integration/vendor/service setup, native ASAN and other platforms, default CLI fallback removal, and independent native build ownership remain open under #703/#709/#202.

Previous patch checkpoint a4b73596f passed 58 original cases per runtime (link 4, native-binlink 7, patch 29, install-patch 18), six private files, and build 46/46; evidence. #719 is complete for native patch activation and tested lifecycles. The original patch aggregate was not rerun for this launch batch; all six older private files plus audit were. #718 bin ownership is fixed on macOS, with native Windows filesystem verification still open. #717 remains complete for the tested native link lifecycle.

Previous package milestones #713/#714/#715/#716 remain verified and closed: local dependency identity/deduplication, long-prerelease frozen lock equality, native update/outdated and cache-freshness regression. The preceding selected original update/outdated/interactive/semver batch passes 78 cases per runtime, excluding filtered-out cases. Create entrypoints and required native parser restoration are verified. #190 self-upgrade and #710 fixture validity are now closed; #438 JSX failures and #701/#694 unexplained deadline failures remain open.

Current compressed build/test artifacts are committed under docs/docs/bun-port-evidence/. Generated fixtures and aged compiler cache data were cleaned after preserving evidence; source changes and installed binaries were retained. Earlier raw logs lost when the previous worktree disappeared are not claimed retained; their committed summaries/hashes remain historical evidence.

Earlier checkpoints

Earlier verified checkpoint: 1cbac2921, with implementation e1a5d0e20 and scanner correction 6448d628a, all pushed to origin/main. #702 is complete for native HTML activation and removal of fabricated FormData child output/RSS. 100% logical Bun parity remains incomplete.

• Ordinary and single-native-VM HTML directory routes each pass 145 tests / one original Linux-only skip / zero failures. The native aggregate's 530 assertions match pinned Bun. Home's real 100-iteration file-error child measures 1.78 MB growth, within the unchanged 10 MB limit. Build 46/46; focused harness 23/23 steps / 3/3 tests, with independent registration metadata.
• Adjacent routes pass: Blob 65/65, body/stream 9,467 passes / four upstream skips, Headers/Response 170/170, Request 24/24, bootstrap FormData 5/5, microtasks 2/2. #701 remains unexplained and open; a passing run is not a repair. #694 remains open.
• #703 is the next broad migration. Exact routing predicates show 175 native test-runner paths, 266 native scripts and 4,267 bootstrap paths among Home's 4,708 classifications. Exact pinned CI discovery selects 4,754 tracked paths before exclusions, exposing 78 missing and 32 extra Home entries, plus mode/config differences. These are inventory counts, not suite passes. The audit command and full path sets are now published in the repository.
• The native triage scanner no longer extends original test deadlines or retries timeout-only failures as slow. Its negative control retains one attempt and a failure. It remains a triage tool, not the complete acceptance gate; discovery and full outcome preservation remain in #703.
• Pinned-object integrity remains 12,990 identical regular files, five raw-identical symlinks, zero missing paths and zero mode differences, with only historical expectations metadata different. Remaining substitutes, faithful full-suite execution, other platforms and independent native build ownership remain unfinished under #202/#703. Logs: zig-out/bun-port-702/; verified Home SHA-256 4a937ded48d609b76e336fdc192dd96439653a228073a234aaafcba176a5e783.

Previous Blob publication: f491cacd7 closed #700 with 65 native passes, preserved the original allocation/prototype/ownership workloads and four snapshots, and removed UTF-16 child and array-prototype descriptor substitutes.

Previous body/stream publication: 04644b2d0 closed #699 with 9,467 native passes and four original skips, removed three clone-specific substitutes, and retained the full HTTP/3 matrix and child/GC workloads. Its adjacent Request run exposed the original 5-second deadline failure now tracked in #701; subsequent diagnostic passes were not counted as a repair.

Previous Headers/Response publication: 649c3c48f closed #698 with 170 native passes, corrected mirror cwd/startup environment and absolute-path CLI rewriting, and preserved the original Response snapshot. Its build passes 46/46 and focused harness 6/6; the initial 4/5 run exposed the additional CLI root defect and remains recorded as historical evidence.

Previous Request publication: 5e45c58a3 closed #697 for native Request activation and removed unused bootstrap VM initialization from native corpus dispatch. Its complete four-file matrix passes 24/24. Direct native stress peak footprints were 22,594,376 bytes for clone stress and 194,872,232 bytes for method-allocation stress; original loops and limits remain intact. The older bootstrap implementation is not claimed repaired.

Previous paired repair: e148e06c1, published to origin/main. #695 and #696 are fixed: complete TypeScript conformance is 1,419/1,419; both optimized Bake integration scripts and gzip decoding with both libdeflate settings pass. 100% logical Bun parity remains incomplete.

The preceding main commit 2129e4a82 enters the macOS WebView host before CLI/JSC initialization and restores native execution of the shell and both complete WebView matrices. All three remaining whole-file TODO substitutions and their helper are removed. #693 is complete for that integration.

• Unchanged pinned WebKit: 57 pass / 1 upstream TODO / 0 fail / 141 assertions. Shell: 378 pass / 83 upstream TODOs / 0 fail / 1,035 assertions.
• Four ordinary native corpus routes: 487 pass / 89 upstream TODOs / 0 fail / 0 unsupported registrations. Chrome WebSocket contributes zero passes and five upstream TODOs, not implemented coverage.
• The first strict Chrome run still fails its original first-navigation deadline (51 pass / 1 fail); pinned Bun passes 52/52. Later probes and a full Home run pass but do not erase that failure: #694 remains open. No retry, warm-up, sleep, timeout extension, assertion change, or workload reduction was added.
• Optimized native build: 46/46 steps. Earlier feature-flag units: 6/6 with defaults and in each explicit 0/1 fresh-process run; unit build 23/23 steps. Both complete Bake production/development integration scripts pass; absent/0/false opt-in is correctly rejected. Unchanged gzip/fetch is 24/24 / 43 assertions with each libdeflate setting.
• Nested-rest checker regressions: 4/4; complete TypeScript conformance 1,419/1,419; standalone reproduction exits zero. The original regression now requires zero diagnostics, and new negatives retain element-type and tuple-order errors.
• Earlier source-restoration regression: 1 pass, zero logged errors/leaks. Earlier completed native runtime units: 1,826 pass / 19 skip / 0 fail; that entire unit artifact was not rerun for the latest paired repair.
• The earlier optimized aggregate was interrupted (exit 143) and is still incomplete. Its nested-rest and Bake/feature-flag failures are repaired and verified separately above (#695, #696); no completed aggregate or whole-suite pass is claimed. The default Debug aggregate was not established by that run.
• The earlier bootstrap Request stress test reached 13.7 GB sampled footprint / 14.4 GB peak and was stopped. Its obsolete execution path is replaced by verified native coverage in #697 above; the bootstrap shim used by other adapters is not claimed repaired. The earlier full aggregate was not resumed, and no completed corpus aggregate or whole-suite pass count is claimed.
• Pinned Bun 4982b91e3702094330f3be3883354c52b8c01323: 12,996 tracked paths present; 12,990 identical regular files, five identical symlinks, one differing expectations.txt, zero missing. The expectation delta removes four historical FAIL entries; executable test sources are unchanged.
• Changed-file formatting, documentation Pickier and whitespace checks pass. Repository Pickier retains 59,302 findings (31,238 errors / 28,064 warnings). The README's Settlers checkout is absent. External Bun C++ build artifacts, other adapters (#202), upstream TODOs/skips, other platforms and full source/build ownership remain unfinished.

Publication correction: all 18 commit URLs in this issue's historical body were verified as ancestors of main at this checkpoint. Historical statements below saying those commits were not merged describe their original publication state and are superseded by this audit.

How to run / verify

• Native debug build: zig build debug -Denable_jsc=true -j2 --summary all from the Home repository. External native dependencies and ownership limits are documented in docs/docs/NATIVE_RUNTIME_BINDINGS.md.
• Native regression: HOME_NATIVE_VM=1 ./zig-out/bin/home-debug tests/runtime/native-worker-inbox-shutdown.test.mjs (substitute the desired regression file).
• Strict native corpus: HOME_NATIVE_VM=1 HOME_CORPUS_FULL_VM=1 ./zig-out/bin/home-debug test <corpus-file>; both overrides are required to bypass the pragmatic corpus adapter.
• Outer corpus CLI: unset HOME_NATIVE_VM and HOME_CORPUS_FULL_VM; any nonempty value, including 0, enables those overrides. Keep native execution and the pragmatic adapter's results separate.
• Use bounded subprocesses and per-child timeouts; limit concurrency to 2–3 small fixtures. Clean only owned temporary artifacts. Do not use blanket pkill commands.
• Leak measurements must retain their original workloads and thresholds; Debug and ReleaseSafe results are not interchangeable.
• Skips, disabled bodies, timeouts, unsupported cases, external dependencies, and failing negative controls do not count as completed parity.

Historical publication checkpoints (through 2026-08-28)

• Published HTTP/3 verification-isolation checkpoint: 0fd884f59, pushed on codex/runtime-port, not merged to main. #539 is closed after a controlled environment correction: the complete unchanged adversarial file passes 12/12 serialized Home processes and 12/12 serialized pinned Bun processes, all 27/27 / 163 assertions. Three eight-way waves reproduce the same first-64-KiB deadline sequence in both binaries (Home 11/12 process passes; pinned Bun 5/12) while all sixteen UDP ports are distinct and the shared eleven-CPU host is saturated by native tests, a ReleaseFast build, and system work. No runtime source, test deadline, workload, retry, sleep, skip, or protocol fallback changed. Heavy HTTP/3 files must be verified serially and away from builds. #530 retains platform/integration and native dependency ownership; all other explicit suite gaps remain. 100% logical Bun parity is not achieved.
• Previous native TLS SNI/ALPN checkpoint: a9a95f6a4, pushed on codex/runtime-port, not merged to main. #542 now ports per-connection synchronous/asynchronous SNICallback, pending-certificate resumption, original callback error propagation, same-host precedence, offered-protocol ALPNCallback validation, callback-safe teardown, and ALPN selector handoff across SNI contexts. The complete unchanged server file is 38/38 / 147 assertions, exactly matching pinned Bun. The new combined lifecycle regression passes 20/20 independent processes plus VM destruction. The seven-file TLS sweep is 120 pass / 1 skip / 3 TODO / 2 fail / 380 assertions; both failures are live bun.sh OCSP checks reproduced in pinned Bun. Build 17/17; units 1,828 pass / 19 skip / 0 fail (19/19 steps); sequential regressions 63/63, including full spawn and child-process. Home's vendored C source compiles with reconstructed pinned release flags, while the debug build still links the pinned Bun object. Darwin only; #530 platform/integration, other open subsystems, skipped/TODO behavior, dependency ownership and whole-suite parity remain incomplete. 100% logical Bun parity is not achieved.
• Previous native TLS session/keylog checkpoint: 04569d8f9, pushed on codex/runtime-port, not merged to main. #543 replaces the no-op native dispatch with retained FIFO session/keylog delivery for direct TCP TLS and upgraded Duplex TLS, preserving session-before-data ordering across destructive callbacks. The complete unchanged connect file is now 28 pass / 2 fail / 1 skip / 106 assertions, exactly matching pinned Bun; both remaining live OCSP assertions also fail in Bun. The native lifetime/GC regression passes 20/20 independent processes. Build 17/17; units 1,828 pass / 19 skip / 0 fail (19/19 steps); sequential regressions 62/62, including 31 native files. Adjacent TLS is 47 pass / 2 skip / 7 fail / 424 assertions; all seven failures are SNI/ALPN work in #542, now next. The checked-in Home uSockets source has the logical queue implementation and compiles with the linked release object's exact flags, but the debug build still links that pinned Bun object. Darwin only; platform/integration, skipped/TODO behavior, native dependency ownership and full parity remain incomplete. Whole-suite parity is not achieved.
• Previous native TLS context checkpoint: fbdfbf890, pushed on codex/runtime-port, not merged to main. #532 now provides private contexts/CA isolation, binary PFX, detailed certificate chains, hostname authorization and TLS min/max propagation. Six unchanged core files pass; context 7/7, cache 9/9, cert 28 pass / 3 TODO, hostname 5/5. Native normal/destruction and original leak checks pass. Native build 17/17; runtime units 1,828 pass / 19 skip / 0 fail, 19/19 steps; sequential regressions 61/61, including 30 native files. Complete compression 32/32, HTTP/2 client 60/60, HTTP/3 client 52/52, both proxy files, full body-stream 9,086/9,086 / 135,984 assertions, five EOF/fetch files and both abort files pass. The initial HTTP/3 adversarial run failed 24 pass / 3 fail / 160 assertions, followed by three 27/27 / 163 serial replays. #539 remains unresolved; passing retries do not erase the final artifact’s initial failure or prior Home/Bun failures. 249 TLS sources, 48 HTTP/regression sources, and 13 TLS implementations match the pin. Full TLS server/connect files still fail: #542 tracks SNI/ALPN callbacks, #543 session/keylog delivery; two live certificate assertions also fail in pinned Bun. Failed attempts and all skips/TODOs are retained. Darwin only; integration, platform and full ownership remain incomplete. Whole-suite parity is not achieved.
• Previous compression/CONNECT checkpoint: 2c5df8cce, pushed on codex/runtime-port, not merged to main. #531 ports all four request-compression encodings and file/buffer ownership; #535 repairs CONNECT writer/envelope handling. Complete unchanged compression 32/32 / 78 assertions, HTTP/2 client 60/60 / 197, HTTP/3 client 52/52 / 241, and both proxy files pass against pinned controls. Full body-stream 9,086/9,086 / 135,984; build 17/17, units 1,827 pass / 19 skip / 0 fail (19/19 steps), regressions 60/60, including 29 native files; all five EOF/fetch and both abort files pass. HTTP/3 reliability remains unresolved in #539: initial adversarial failure, then three passing Home replays; failures also reproduce in published Home and pinned Bun. TLS context work remains #532. Forty-eight test/harness/fixture sources and eleven implementations match the pin. Failed attempts remain recorded. Darwin only; skips/TODOs, platform/integration and full source ownership remain incomplete. Whole-suite parity is not achieved.
• Previous HTTP/2 checkpoint: fd239c3ac, pushed on codex/runtime-port, not merged to main. #533 now preserves the idle-session Host hash and rejects invalid response-name tokens while consuming complete HPACK blocks. Complete unchanged HTTP/2 adversarial 15/15 / 33 assertions (parent 13 pass / 2 fail); lifetime 7/7 / 21 assertions with original 200-request workloads. Full body-stream 9,086/9,086 / 135,984 assertions. Build 17/17, units 1,825 pass / 19 skips / 0 fail (19/19 steps), sequential regressions 59/59 including 28 native files. HTTP/2 client remains 58 pass / 2 fail / 2 errors; HTTP/3 client 50 pass / 2 fail, both compression #531. HTTP/3 adversarial retains 27/27. Newly exercised CONNECT-proxy crashes reproduce in both parent and candidate; pinned Bun passes both files, tracked separately in #535. TLS context work remains #532. Forty measured test/harness/fixture sources and six consulted implementations match the pin. Darwin only; platform/integration, skips/TODOs and full source ownership remain unverified. Whole-suite parity is incomplete.
• Previous TLS/HTTP transport checkpoint: 29c196835, pushed on codex/runtime-port, not merged to main. TLS ABI offsets, HTTP/3 trust eligibility and duplex progress are repaired. Full unchanged body-stream 9,086/9,086 / 135,984 assertions; HTTP/3 adversarial 27/27; client 50 pass / 2 fail (compression #531). HTTP/2 client 58 pass / 2 fail / 2 errors (#531); adversarial 13 pass / 2 fail (#533 verification-host isolation/header validation). TLS server validation 16/16; remaining context/CA/version work is #532. Build 17/17, units 1,824 pass / 19 skips / 0 fail (19/19 steps), sequential regressions 58/58 (27 native files); full spawn 126 passes, child-process 32 passes. Five complete fetch files and both abort files retain their passes. 72 test/harness sources and six implementation sources match the pin. Earlier timeout and disk-exhaustion attempts are retained; complete sequential reruns pass with original workloads/deadlines. See #530 for detailed evidence. Shared Darwin checkout; platform/integration, skipped/TODO behavior and full ownership remain unverified. Whole-suite parity is incomplete.
• Earlier body-abort checkpoint: fb161314a, pushed on codex/runtime-port, not merged to main. #529 ports terminal response-error settlement before upload cancellation; unchanged abort file 2/2 / 6 assertions (parent 1 pass / 1 fail), queued-abort 1/1, and native 24-scenario normal/destruction/reentry checks pass. Build 17/17; units 1,824 pass / 19 skips / 0 failures (19/19 steps); regressions 57/57 (26 native files); 64 measured upstream sources match the pin. Five body files pass, including clone 25/25 and body 346 passes / 4 skips; five fetch files retain their passes including gzip 24/24. At that earlier checkpoint, the full body-stream matrix was not green: parent 4,883 pass / 4,203 fail / 2 errors, candidate 4,883 pass / 4,203 fail / 10 errors, pinned Bun 9,086/9,086. #530 tracks the HTTP/3 server path: parent client/adversarial files are 6 pass / 46 fail and 5 pass / 22 fail; both clients fail against Home's server before its JS handler, both succeed against Bun's server. Original assertions/workloads/deadlines retained. Historical HTTP/3 completion claims below are corrected. Shared Darwin checkout; unrelated compiler edits excluded. Platform/integration, skips/TODOs and full source ownership remain unverified. Whole-suite parity is incomplete.
• HTTP/installer checkpoint: 90ced0ed6, pushed on codex/runtime-port, not merged to main. #526 repairs buffered EOF completion and fragmented tar-format selection; native real-install, streaming-handshake, VM-destruction and three 404 transport controls pass. #527 advances unchanged fetch-gzip from 22/2 to 24/24 / 43 assertions, preserving the 1,025 MiB workload. #528 advances unchanged streaming-extract from 5/1 to 6/6 / 180 assertions, enforcing the pinned 2 GiB buffered limit against the original 2.25 GiB payload. Four further complete fetch files pass; full zlib retains 378 passes / 2 skips / 434 assertions. Retains bun-run 291/291, native hard-link-PATH bunx 33 pass / 1 Windows-only skip / 0 fail, workspaces 58/58, eval 33/33, tsconfig 6/6, resolver 39 pass / 2 skips / 1 TODO plus its spawned 9 pass / 1 TODO, and 21 config controls. Build 17/17; units 1,824 pass / 19 skips / 0 failures (19/19 steps); regressions 56/56 (25 native files); 46 measured upstream test sources match the exact pin. No upstream assertion/workload/deadline changes. #526–#528 retain platform/integration scope. Shared Darwin checkout; unrelated compiler edits excluded. Skips/TODOs, other platforms, integration and full source ownership remain unverified. Whole-suite parity is incomplete.
• Resolver checkpoint: 9b3a5c1ad, pushed on codex/runtime-port, not merged to main. #525 ports canonical resolver diagnostics and fallible, retained installer initialization errors; readable-file dispatch now precedes cwd scanning. Unchanged tsconfig-override 6/6 / 17 assertions (parent 5/1), resolver 39 pass / 2 skips / 1 TODO / 0 fail / 59 assertions plus spawned file 9 passes / 1 TODO / 41 assertions (parent's unreadable-cwd failure resolved under the original 5,000 ms deadline). Native repeated-error and VM-destruction checks pass. Retains bun-run 291/291 / 416 assertions, native hard-link-PATH bunx 33 pass / 1 Windows-only skip / 0 fail / 161 assertions, workspaces 58/58 / 192 assertions, eval 33/33 / 70 assertions, and 21 config controls. Build 17/17; units 1,820 pass / 19 skips / 0 failures (19/19 steps); regressions 56/56 (25 native files); 37 sources match the exact pin. No upstream assertion/workload/deadline changes. #526 remains a reproduced failure: EOF-delimited registry 404s trigger a pending-task assertion in parent and final candidate; length-delimited controls pass. Other platforms, branch integration, skips/TODOs and full source ownership remain unverified. Shared Darwin checkout; unrelated compiler edits excluded. Whole-suite parity is incomplete.
• CLI config checkpoint: 828ecc729, pushed on codex/runtime-port, not merged to main. #517 optional-config and entrypoint scanning now match all 21 pinned-control boundary cases; native normal/destruction regression passes and published parent fails the new assertion. Full unchanged bun-run 291/291 / 416 assertions, native hard-link-PATH bunx 33 pass / 1 Windows-only skip / 0 fail / 161 assertions, workspaces 58/58 / 192 assertions, and eval 33/33 / 70 assertions (declared React dependency installed by Home). Build 17/17; units 1,820 pass / 19 skip / 0 fail (19/19 steps); regression commands 56/56 (25 native files); 35 measured sources byte-identical to pin. No upstream assertion/workload/deadline changes. New #525 tracks the unchanged tsconfig-override failure: Home parent/candidate 5 pass / 1 fail, pinned Bun 6/6, due to auto-install 404 diagnostics. #517 now retains platform/integration and help-rendering verification. Shared Darwin checkout; unrelated compiler edits excluded. Skips/TODOs, other platforms and full source ownership remain unverified. Whole-suite parity is incomplete.
• Native installer checkpoint: 3be72ca5b, pushed on codex/runtime-port, not merged to main. Native bunx and the real installer chain (#518) now resolve all three remaining bun-run failures: 291 pass / 0 fail / 416 assertions. Full unchanged bunx passes 33 tests / 1 Windows-only skip / 0 failures / 161 assertions with isolated PATH hard links named node/bun to the exact Home binary (no installed JS runtime delegation). Ordinary PATH remains 32/1/1 because of an installed claude binary; pinned Bun reproduces it. An earlier native symlink PATH run's old CLI-basename failure is separately recorded. Native registry credentials, actual installation, both analyzers, auto-install HTTP initialization and VM destruction checks pass; Angular's original 300,000 ms timeout is resolved without changing the deadline. Workspace suites 58/58 / 192 assertions; regression commands 56/56 (25 native files); build 17/17; units 1,820 pass / 19 skips / 0 failures (19/19 steps). All 33 measured upstream sources match the pin; generated test-tarball bytes are archived and restored. No assertion/workload/deadline changes. #508 now retains platform/integration scope; #517 optional-config scanning and #519 platform/integration verification remain open. Shared Darwin checkout; unrelated compiler edits excluded. Skips/TODOs and other platforms are unverified. Whole-suite parity is incomplete.
• Compression checkpoint: 095f90396, pushed on codex/runtime-port, not merged to main. #516 ports Zlib/Brotli/Zstd async buffer pins, GC roots and fresh write-result resolution. The complete unchanged zlib file now passes 378 tests / 434 assertions, with 2 skips and 0 failures; bounds 9/9, reentrancy 3/3, native regression commands 55/55 (24 native files), and runtime units 1,819 pass / 19 skip / 0 fail. Twenty lifetime stress runs plus VM destruction pass. No upstream assertions/workloads/deadlines were changed. This supersedes the prior zlib failure linked from #513. Darwin only; #514's four remaining stream-file gaps, broader CLI/runtime gaps and whole-suite parity remain open.
• The 65 previously local port commits through 6de9852ea8443c3b5d4a58dcdbcf11944456fd2e are now published on codex/runtime-port. Their original hashes were recovered from the existing local codex-git-shadow.git metadata; no replacement history or force push was used. They are published on the branch, not merged into main.
• Worker checkpoints on the published branch: descendant shutdown 41f10cbebe6d65340304f4b93eb4fd3fa10d0e19 (#470), native parentPort 3b0721c32b2d237c38380900ed449217ef1af853 (#477), and snapshot cancellation c6a4e4a3513dfb2a05a13357c06f598ad9f1d6ce (#471). Not merged into main. Snapshot tasks carry IDs only; the parent thread owns promise handles and completed payloads through success, worker exit, and parent-VM shutdown.
• Snapshot checkpoint: 20/20 native regression files, 16 byte-identical upstream Node fixtures, and native build/cache 16/16 pass. The six-scenario snapshot suite passes four complete runs, including two with VM destruction enabled. The unchanged upstream snapshot-GC test passes its full 15 × 300 = 4,500 snapshot workload (1 test / 15 assertions). No sanitizer claim.
• Strict native MessagePort corpus: 15 pass / 3 existing debug-sanitizer guard skips / 0 fail / 529 assertions. Both native execution flags were set. Skips are not completed parity.
• The full worker_threads corpus file is not green: at the snapshot checkpoint, 33 pass / 1 timeout / 1 between-test error, although its three snapshot cases pass. The test-name filter failure is now fixed in #481; the current filtered snapshot run is 3 pass / 31 filtered out / 0 fail. This does not resolve the full-file timeout. The unchanged 100 MiB eval-source case exceeds its original five-second deadline (#482); its standalone memory-threshold check passes in native debug in 20.22 s versus 2.20 s in Bun. Debug/optimized timing and the full-file deadline remain distinct from snapshot cancellation.
• Native test CLI checkpoint 1340cebd5 (#481), published on codex/runtime-port, not merged into main: real argument and bunfig parsing, fatal diagnostics, and buffered/direct output semantics. Build/cache 16/16, all 20 native regression files, and 50 unchanged upstream tests / 119 assertions pass across filtering, configuration, retry/repeat hooks, concurrency, path ignores, and full coverage. The snapshot selection filters out 31 cases; they are not counted as passes. No tests or snapshots were weakened.
• Newly exposed test-runner feature gaps: isolated global/module records #484 (1 pass / 3 fail, Bun control 4 pass) and parallel coordinator/worker IPC #485 (0 pass / 3 fail, Bun control 3 pass). Both retain their unmodified upstream reproducers and remain open. Other unverified CLI/runtime features are not implied complete by #481.
• Remaining shutdown work: generic queued task cancellation #465 and work-pool quiescence #468. Snapshot request ownership #471, parent-port lifecycle #477, and descendant shutdown #470 are implemented and verified on the published branch, with integration context retained in their issues. Related ownership: #461, #462, #466, #469.
• Other active parity gaps include the core corpus exclusions in #456, reduced-context realms in #454, and Node-API/FFI follow-ups #459 and #460. Historical checkbox sections below are not a current whole-suite completion report.
• 100% logical Bun parity is not achieved.

A. node:tls socket layer

• SNI/ALPN callback handling and asynchronous suspension — implemented in a9a95f6a4 (#542); the complete unchanged server file passes 38/38 / 147 assertions, matching pinned Bun.
• Session/keylog events and teardown ordering — implemented in 04569d8f9 (#543); the complete unchanged connect file matches pinned Bun, with two live OCSP failures retained in both.
• getCipher() cluster works — was a false diagnosis: the null cipher/session came from the broken (cert-less) handshake caused by the Buffer-cert gap, now fixed (b3492d3). Verified locally: getCipher {TLS_AES_128_GCM_SHA256, version:"TLSv1/SSLv3"} (matches pin test), getProtocol TLSv1.3, getSession 1106-byte Buffer, getEphemeralKeyInfo/getSharedSigalgs/exportKeyingMaterial/isSessionReused all correct. The pin test connects to bun.sh (crashes on this macOS 27 box) so it can't be run here, but the behavior is confirmed.
• Detailed getPeerCertificate at the current pin — implemented in fbdfbf890 (#532): peer/trust-store issuer chains, self-issued roots, checked conversion and native reference ownership. Complete cert suite 28 passes / 3 TODO, original leak workload and memory threshold pass. The earlier claim that the detailed TODO matched the pin was historical and does not describe the current pin.
• ALPN — fixed in 428aa2c (hand-mirror SSLConfig.fromJS never parsed ALPNProtocols).
• Static SNI server.addContext() selection — the complete context suite and most-recent-context Node program now pass (#532). Dynamic SNI/ALPN callbacks and asynchronous handshake resumption remain #542.
• Cert-chain verification — "unable to verify the first certificate" / "unable to get issuer certificate" on incomplete-chain + intermediate-CA scenarios (basic single-CA verify works).
• Config validation — tls.createServer({passphrase: wrong}) should reject but listens; passphrase IS passed to us_ssl_ctx_from_options, so it's uws-level validation timing/error-surfacing.
• getPeerCertificate '\0' handling (CVE-2009-2408 test).

B. Bun Shell ($)

• Bun Shell template commands and command substitution — closed by #73. Basic template execution and the mixed-output/error command-substitution stress case run through real Home execution. Follow-up 31e2dc603 forwards explicit spawn environments and pins the installed Home executable for the fixture-free corpus path.

C. fetch / HTTP client

• redirect + gzip — fixed in c508e00 (NativePromiseContext wiring; see below). fetch-gzip.test.ts 2/3 → 5/0.
• streamed gzip with delay — fixed in c508e00 (same root).
• body-stream.test.ts — full unchanged matrix passes 9,086/9,086 / 135,984 assertions at 2c5df8cce, including HTTP/3. Original workloads/deadlines preserved. Abort settlement remains green (2/2 plus queued 1/1, #529). This file-level result does not close HTTP/3 reliability/platform work in #530/#539.

D. HTTP/2 & HTTP/3

• Native HTTP/2 completion — complete unchanged client 60/60 / 197, adversarial 15/15 / 33, lifetime 7/7 / 21 at 2c5df8cce (#531/#533). Earlier client compression failures are repaired. Platform/integration remain unverified; file-level passes do not establish complete HTTP/2 parity.
• HTTP/3 — platform and integration verification, #530: 2c5df8cce brings client 52/52 / 241 and full body matrix 9,086/9,086 / 135,984. Request compression is implemented (#531). #539 is closed by 0fd884f59 as a verified host-contention artifact: serialized Home and pinned Bun each pass 12/12 processes, while intentional eight-way waves reproduce the same first-upload deadlines in both. Original tests/deadlines retained. Other platforms/integration remain unverified.

E. Deep/unported subsystems

• napi-FFI cc.test.ts crash — resolved: exact upstream file is 5 pass / 10 upstream skips / 0 fail in Home; the extended mirrored file is 8 pass / 10 upstream skips / 0 fail. Pinned Bun at 4982b91e3 currently reports 4 pass / 10 skips / 1 fail on the upstream file because napi_main is not exported from the fixture compile.
• S3Client remaining platform/external integration (#597) — deterministic native matrix is 66 pass / 3 upstream credential-gated skips / 0 fail in 5c2ac4542; the stream-cancel crash, task/body ownership, proxy resolution, early-response multipart reuse, malformed ListObjects parsing, and eight-file production routing are fixed. Linux/Windows and credentialed S3/R2 integration remain open.
• Bake/HMR remaining optimized/platform verification (#901) — native production/development/HMR is functional (#551/#553/#555). The complete native suite checkpoint validates all 24 unchanged files (177 pass / 2 upstream TODO / 0 fail / 0 skip), genuine frozen locks, complete journal identities and original test deadlines (#902 closed). Spawned fixture cwd, production source maps and changed-path aliases are repaired. Debug ownership and release frontend checks pass; optimized runtime and wider platform/SDK acceptance remain open. #564 tracks Linux TODO coverage.
• Worker corpus remaining platform/performance verification — e1523b1ee and eae013190 promote 16 exact native files: 264 pass / 3 upstream release-mode skips / 1 upstream TODO / 0 fail. The corpus runner applies its real preload and preserves sentinel path ownership. Web Worker messaging, transfer, structured clone, Blob, termination, leak, heap-snapshot GC, unchanged 15787.test.ts, and the complete worker-destruction matrix are green; #598 is closed after preserving cached exit code 0. The complete worker_threads.test.ts is green in ReleaseFast under #482 but retains its documented Debug-only five-second eval-source deadline gap. Linux/Windows verification, the upstream TODO/skips, and whole-suite worker parity remain open.

F. Memory-leak tests (triaged on ReleaseSafe 2026-07-07 — build target un-rotted in 55f7140)

• setTimeout-clear-in-callback — ✅ FALSE ALARM: all 3 variants (clear/refresh/repeat) pass on ReleaseSafe; debug-build overhead as suspected.
• text-decoder utf16le output-buffer leak — ✅ FALSE ALARM: both utf-16le and utf-16be pass on ReleaseSafe.
• streams-leak — ✅ FALSE ALARM: passes on ReleaseSafe.
• serve-body-leak — ✅ CLOSED (4ee746c): root cause was the libc mimalloc shim (allocator page retention, not a leak — see forensics comment). Real mimalloc re-attached (Phase 12.2); 7 pass / 0 fail on ReleaseSafe.
• inspect-error-leak — ✅ CLOSED (4ff003d): ZigException.Holder.deinit was a stub that dropped the pin's deferred resetArena — every inspected error left one full transpile in the module loader arena (74MB/100k iters). Now passes with 'RSS increased by 4 MB', exactly real Bun's number.
• node:util test-aborted gc-cleanup — ✅ CLOSED (d459e0d): FinalizationRegistry callbacks never fired AT ALL — JSCScheduler.zig was a no-op placeholder and the two event-loop exports were noop stubs, severing JSC's DeferredWorkTimer delivery chain. Pin's scheduler ported; test-aborted now 5/5 (gc-cleanup passes in 2.5ms).
• (new) HTMLRewriter handler leak — reclassified, NOT a leak: all 58 lol_html_* symbols are noop stubs (lol-html Rust crate unported — link_bun_rust_archive = false); the child panics on the first .on() (else => unreachable on a garbage return at lolhtml_sys/lol_html.zig:253) and the test fails on stderr. Belongs with section E as an unported-library item: link lol_html_c_api (rlibs exist in the Bun build tree), delete the stubs, and restore the pin's HTMLString.toString free (lol_html.zig:623 — latent leak once real lolhtml links).

G. Environmental / infra (mostly resolved)

• Test runner: home test defaults to native bun:test runner (078c659)
• bun:internal-for-testing gated in release (13deb08)
• node v24 + bun v1.3.14 installed via pantry (clears differential-spawn "node not found" failures)
• Build recovery recipe after /tmp wipe (memory: home-build-portability.md)

Done this session (context)

signals (db0df4d), native-VM file runner (f296bb6), Worker cross-thread + teardown (dcb6db6, 85aba46), native test runner default (078c659), translateUVErrorToE (188268e), Bun.markdown renderer (2768d9a), MD4 (1279c18), Bun.CSRF (96b8f05), node:tls SecureContext ~40 tests (6cf4a6b), release build + build-recovery (dff7bfa, df83921, 13deb08).

Native isolation checkpoint — 6e092777a

Published on codex/runtime-port; not yet merged to main. #484 now passes the entire unchanged isolation file: 19 tests / 67 assertions, plus three consecutive repeat runs. The real JSC module-record bridge replaces the parked panic path. File-boundary cleanup releases watcher, server, timer, watch-file scheduler, and AbortSignal GC references; server callback protections are balanced across reload and destruction.

Verification: native build and unchanged cached rebuild 17/17; generator/ABI tests 8 tests / 135 assertions; all 20 native regression files pass, including expanded isolation probes. Fourteen selected unchanged upstream commands total 92 passing tests / 242 assertions, including the 50-test CLI regression set. Two platform skips and 231 filtered cases in those selections are not passes. The existing AbortSignal spawn-glyph-only check alone is not proof of child assertion execution; direct native lifecycle and timeout checks provide separate evidence. Workloads, deadlines, and upstream assertions were preserved. Full details and control limitations are in docs/docs/NATIVE_RUNTIME_BINDINGS.md.

Separate newly tracked failures: fake clocks #490 (16 pass / 14 fail, also reproduced before this checkpoint; Bun control 30 pass) and deferred imports #491 (0 pass / 1 fail / 19 filtered out in the selected parser reproduction). Parallel orchestration #485 and the full worker eval deadline #482 remain open. Installed Bun 1.3.14 fails four newer isolation cases itself, so no passing full-file Bun control is claimed. The full port and 100% logical Bun-suite compatibility are still incomplete.

Native fake-clock checkpoint — d0ed7c906

Committed and pushed on codex/runtime-port; not yet merged to main. #490 now passes the unchanged core fake-timer file: 30 tests / 83 assertions, up from 16 pass / 14 fail. Installed Bun 1.3.14 passes the same 30 / 83 control. Mockable timer deadlines now share the fake clock, fractional advancement retains nanoseconds, real option/getter validation runs before clock mutation, and both reset paths release timer/AbortSignal references outside the heap lock.

Verification:

• Native build and cached rebuild: 17/17 steps.
• All 21 native regression files pass; the new fake-clock test also passes three consecutive repeats and fails against the cached pre-change Home binary. It checks actual Home execution, inherited/throwing getters, Date/performance agreement, fractional nested callbacks, retained cancelled handles, natural exit and bounded GC counts after allocating 3,072 timeout/interval wrappers plus 1,536 AbortSignal timeouts across resets.
• Entire unchanged fake-timer directory: 61 reported pass / 438 TODO / 0 fail / 115 assertions in Home and Bun. 14 reported passes are it.failing cases, not implemented features; all 438 TODOs are excluded from logical completion. There are 47 ordinary passing tests, including the core 30.
• Unchanged isolation: 19 tests / 67 assertions. Unchanged timers/promises: 4 / 7. Twenty-four selected corpus files match the pinned Bun checkout byte-for-byte. Focused Pickier, timer-file Zig formatting and whitespace checks pass.

The wider Node timer file still hangs after printing 19 passes on its final caught-immediate-exception fixture. This happens in the pre-change Home binary too. The standalone fixture times out in both Home binaries and exits successfully in Bun; the original-checkout Bun control passes the full 20 / 31. Tracked separately in #493; neither incomplete runs nor expected failures count as logical parity.

Deferred imports #491, parallel coordinator/IPC #485, the full worker deadline #482, remaining native ownership, other platforms and full-suite parity remain open. The full port and 100% logical Bun-suite compatibility are not complete. See docs/docs/NATIVE_RUNTIME_BINDINGS.md for the implementation and verification boundary.

Native POSIX spawn checkpoint — 6d797186c

Published on codex/runtime-port; not yet merged to main. #493 is now implemented and verified. The timer fixture hung during import, before scheduling immediates: Node's common helper spawned a missing absolute openssl-cli, and Home misread libc's direct positive errno as success and watched PID 0. The fix propagates spawn and file-action/attribute setup errors, preserves executable paths, and releases failed registration lists. No immediate dispatch code was changed.

Verification: 28 selected commands pass, including all 21 native regression files, the unchanged complete Node timer file (20 tests / 31 assertions), its standalone fixture, isolation (19 / 67), core fake clocks (30 / 83), timers/promises (4 / 7), PATH selection (1 / 2) and the selected missing-command/invalid-fd cases (2 / 9; 32 filtered out excluded). Native build and cached rebuild 17/17; repeated native spawn failures preserve descriptor counts and recover to a valid spawn without re-entering user microtasks. Eight relevant upstream files are byte-identical to source pin 4982b91e3702094330f3be3883354c52b8c01323.

New linked follow-ups keep the failing boundaries explicit: extra stdio pipes #495, real native counter snapshots #496, and package-script passthrough #497. Full child-process remains 30 pass / 2 fail / 1 skip / 1 TODO / 70 assertions; spawnSync remains 4 pass / 1 fail / 1 platform skip / 11 assertions. Installed Bun 1.3.14 passes those three failing scenarios but fails the newer invalid-fd case itself. Deferred imports #491, parallel test orchestration #485, full worker timing #482 and the other open work remain. The full port and 100% logical Bun-suite compatibility are still incomplete.

✅ A. ALPN — fixed in 428aa2c

Root cause (broader than ALPN): home.zig's jsc.generated namespace is a hand-mirror stand-in for the un-emitted bindgenv2 codegen. Its SSLConfig.fromJS hand-parses cert/key/passphrase/etc but explicitly skipped ALPNProtocols (left alpn_protocols = .none). The real C++ bindgenConvertJSToSSLConfig (which does handle it, incl. typed-array views via IDLArrayBufferRef) is present in the pinned obj but not called — Home routes through the hand-mirror instead.

Consequence: client sockets never called SSL_set_alpn_protos, servers never registered the select callback in time → alpnProtocol === false, no negotiation. The onOpen setup timing (socket.zig 512-519) was fine all along; the config just never carried the protos.

Fix: parse ALPNProtocols in the hand-mirror (string or wire-format buffer/typed-array via asArrayBuffer), store the buffer variant as a borrowed ArrayBuffer copied out synchronously in fromGenerated.

Verified locally: client negotiates http/1.1; multi-proto client ["h2"] vs server ["h2","http/1.1"] → h2; no-ALPN path still connects with alpnProtocol=false.

Note for remaining A items: this hand-mirror is likely the root of other TLS config gaps too — buffer/array cert forms are still .none (see SSLConfigFile.fromJS), which is probably why some cert-chain / addContext scenarios fail. getCipher/getPeerCertificate are separate (post-handshake SSL-state, not config).

Machine caveat: the reference bun (system + pin, both fd0b6f1) crashes with index out of bounds: len 4095 on TLS network connects on this macOS 27 box, so bun.sh-based tests (e.g. the connect-suite ALPN case) can't be validated here — used local server+client instead.

✅ Buffer/typed-array cert/key/ca — fixed in b3492d3

Follow-up to the ALPN root cause: the same hand-mirror's SSLConfigFile.fromJS only accepted PEM strings and Bun.file() Blobs — a Buffer/typed-array (the common readFileSync(path) result) fell through to .none. So a server given Buffer key/cert had no certificate (handshake timed out) and a client given a Buffer ca couldn't verify the peer.

Now accepts typed-array/ArrayBuffer via asArrayBuffer (view-resolved). BufferValue carries the resolved ArrayBuffer value (borrowed, copied out synchronously) rather than a *JSCArrayBuffer a view can't cast to.

Verified: server w/ Buffer key+cert serves data; client w/ Buffer ca → authorized=true; Buffer certs + ALPN → http/1.1. JS-array cert forms (multiple PEMs in an array) still .none — a smaller remaining gap.

This likely unblocks a broad swath of node:tls tests that pass certs as readFileSync Buffers.

✅ Array-form cert/key/ca — fixed in b9551ec

Completes the hand-mirror config-parse thread. SSLConfigFile.fromJS now parses JS-array forms (ca: [pem1, pem2], mixed string/Buffer/Bun.file elements) into an owned GeneratedList; the consumer already supported .array.

Config-parse gaps in jsc.generated (home.zig hand-mirror) are now closed: ALPNProtocols ✅, Buffer/typed-array certs ✅, array certs ✅. Verified: client with ca:[Buffer, string] → authorized=true.

Remaining section-A items are not config-parse — they're post-handshake SSL state (getCipher/getSession/getPeerCertificate) and separate APIs (server.addContext SNI, incomplete cert-chain verification, wrong-passphrase validation).

Session progress: config-parse thread closed, getCipher cluster cleared

Three pushed fixes this session — all traced to the same root cause (home.zig hand-mirror jsc.generated dropping unparsed TLS config): ALPN (428aa2c), Buffer/typed-array certs (b3492d3), array certs (b9551ec).

Ripple win: the Buffer-cert fix cleared the entire getCipher/getSession/getProtocol/getEphemeralKeyInfo/getSharedSigalgs/exportKeyingMaterial/isSessionReused cluster — those were never bugs, just symptoms of a cert-less (failed) handshake. All verified locally. Real-pin node-tls-cert.test.ts now runs 25 pass / 3 todo locally (remaining fails are external NODE_EXTRA_CA_CERTS + the machine's bun.sh crash).

New real gap isolated — getPeerCertificate (abbreviated): returns {} (0 own props) even though the cert is valid and the modern getPeerX509Certificate() works. The legacy encoder Bun__X509__toJSLegacyEncoding → JSX509Certificate::toLegacyObject(ncrypto::X509View, globalObject) (src/jsc/bindings/JSX509Certificate.cpp:706) produces an empty object with no exception — a C++/JSC-level discrepancy needing C++ debugging, not a config/Zig fix. Parked for a future session.

Machine caveat reminder: reference bun (system + pin, both fd0b6f1) crashes index 4095 on TLS network connects on this macOS 27 box, so bun.sh-based tls tests can't run here — local server+client tests used throughout.

Section A (node:tls) — local-test survey after the config-parse fixes

Ran the local (non-external) node:tls suites. Strong state:

Remaining local A failure — ssl-ctx-cache: "SSL_CTX is freed once no owners remain (weak cache, not strong)" — a unique-ciphers SecureContext is retained after GC (live count before+1, consistently). SecureContext.finalize is wired to SSL_CTX_free; intern() stores the wrapper in the pin's Bun__SecureContextCache (weak in the pin). So it's a GC/weak-map reclaim subtlety (or Bun.gc(true) not fully collecting in this build), not a config-parse gap — pre-existing, section-F-adjacent. Parked.

The larger node-tls-context / node-tls-connect suites can't be fully scored here — their remaining cases connect to bun.sh/badssl, which crash the reference bun too on this macOS 27 box.

Net section A: ALPN ✅, Buffer certs ✅, array certs ✅, SNI/addContext ✅, getCipher cluster ✅ (parity), getPeerCertificate ✅ (parity). Remaining: the ssl-ctx-cache GC leak + wrong-passphrase validation + detailed-cert chain (shared pin TODO).

🚩 Landmark fix — async Bun.serve handlers (c508e00)

Re-diagnosis of section C: the fetch-gzip failures were not a gzip/redirect/connection-lifecycle bug. Root cause: NativePromiseContext.create()/take() (RequestContext.zig's server-side .then() bridge) were soft-linked fn-ptrs that permanently pointed at local stubs — create → .zero, take → null — and the real Bun__NativePromiseContext__create/take in the linked Bun object were never wired.

Effect: when a Bun.serve handler returns a Promise that settles on a later event-loop tick (any timer or async file/network I/O — i.e. every async handler except one resolved by a synchronous microtask), onResolve fired but take() returned null, so the Response was silently dropped and the request hung until the client timed out. Sync/microtask handlers took the .fulfilled fast-path and masked it.

Isolation path: fetch-gzip redirect hang → not redirect (plain redirect OK) → not gzip → not the ArrayBuffer → await of any macrotask in a handler → handler runs to completion but Response never sent → onResolve fires but NativePromiseContext.take returns null → the create/take stubs.

Fix: point the fn-ptrs at the real externs under -Denable_jsc (stubs kept for the JSC-less test gate). This is broad — it unblocks any async handler doing real I/O, so expect improvements well beyond fetch-gzip.

Verified: handlers awaiting setTimeout/fs.promises.readFile/Bun.file().arrayBuffer() now respond; fetch-gzip 5/0; body-stream no longer hangs (remaining fails are a separate body-stream clone bug).

⚠️ Methodology: debug-build startup makes subprocess-spawning tests false-fail

Measured: a single home -e (or home run) on the debug build takes 3–4 s to start (160 MB binary + JSC init; release ≈ 250 ms). Many tests do Bun.spawn([bunExe(), …]) and assert within a 5 s timeout — so spawn+startup alone eats the budget, and test.concurrent (10× at once) blows past it.

Confirmed false-fails from this, not real bugs:

• http-server-chunking 10 timeouts — the chunked parser (incl. split \r\n terminators / split chunk-size CRLF) works: the exact subprocess script returns Got: Wiki in ~50 ms via both home run and home -e.
• bun-serve-body-json-async — also Bun.spawn(bunExe()).

Consequence: the debug binary is only a reliable signal for in-process tests. Subprocess-spawning suites (a large fraction of test/js/bun/http and CLI tests) need a ReleaseSafe build to score honestly. Building one now; will re-survey on it. This also unblocks the section-F leak tests, which require release.

🔎 New root cause: subprocess stdout not streamed incrementally (Bun FilePoll dispatch)

Investigating spawn-based test failures (#9222, and the pattern "spawn a server fixture → read its URL from stdout → connect") surfaced a real bug, distinct from the debug-startup artifact.

Symptom: proc.stdout from Bun.spawn delivers nothing via for await while the child stays alive — data only arrives when the child exits (pipe closes). Confirmed parent-side: even spawning sh -c 'echo X; sleep 5' (flushes immediately), Home's parent reads nothing until close. So it's not child buffering.

Root: the subprocess pipe reader (SubprocessPipeReader → PosixBufferedReader) registers a FilePoll on the uWS loop's kqueue (registerWithFd does a real kevent64 EVFILT.READ | EV.DISPATCH), and onTick is exported as Bun__internal_dispatch_ready_poll. But the dispatch never fires for these polls — while uSockets primitives (sockets and timers, incl. the await setTimeout handlers fixed in c508e00) work. So Bun FilePoll-based I/O (pipes, fs watch, DNS c-ares) isn't being pumped by the running loop, even though us_poll/us_timer are. A stale comment in SubprocessPipeReader.start ("Home's io.BufferedReader is currently a stub… once the real posix_event_loop FilePoll is wired") points at incomplete FilePoll↔uSockets-loop integration.

Impact: broad — any test that reads a live subprocess's stdout/stderr incrementally (large slice of test/js/bun/http, CLI, and node child_process suites). High priority, but a deep event-loop fix (need to confirm the loop the FilePoll registers on is the one being run, and that the pinned uSockets dispatches tagged Bun FilePolls to Bun__internal_dispatch_ready_poll).

Landed this session regardless: tls-option validation (ba67d6c) — Bun.serve({tls: <non-object>}) now throws TLSOptions must be an object.

🔬 Refined: subprocess-stdout streaming — FilePoll dispatch works; break is in the reader→ReadableStream layer

Instrumented registerWithFd + onTick in a debug build and ran a sh -c 'echo X; sleep 5' child:

• registerWithFd succeeds (rc=0 errno=SUCCESS, real kevent64 on the uWS loop's kqueue).
• onTick fires (is_filepoll=true) when the pipe becomes readable, and onUpdate matches the BufferedReader owner tag → calls onPoll → read.

So the FilePoll dispatch is not the bug (my earlier hypothesis). Traced the Zig side end-to-end and it all looks correct: SubprocessPipeReader → ReadableStream.fromPipe transfers the reader into a FileReader.Source (reader.from(..) rebinds parent, keeps the Source's FileReader-typed vtable, which does have onReadChunk; FilePoll ownership transfers via setOwner).

Remaining suspect (deep): the data read by onPoll isn't reaching the JS for await (proc.stdout) until EOF — the break is in the final FileReader.Source ↔ C++ ReadableStream onPull/enqueue propagation (or the read is buffered and only flushed on onReaderDone). The observed tight re-arm/re-fire loop on one fd suggests data isn't being consumed on the streaming path. Needs a probe of whether FileReader.onReadChunk is actually invoked and whether the C++ stream propagates enqueued chunks incrementally — a deep webcore/streams fix, deferred.

Net: still a real, high-impact bug (server-fixture spawn tests), now narrowed to the webcore streaming layer rather than the event loop.

✅ FIXED — subprocess stdout streaming (b20f4fa): macOS MSG_DONTWAIT was 0x40

Cracked the deep bug from the prior comments. Root cause is a one-line constant: bun.c.MSG_DONTWAIT was hardcoded to 0x40 (the Linux value). On Darwin, 0x40 is MSG_WAITALL and MSG_DONTWAIT is 0x80.

So every nonblocking recv() on macOS was silently passing MSG_WAITALL — asking the kernel to block until the entire read buffer was filled. On a nonblocking socketpair that returns EAGAIN whenever fewer bytes are available than the (large) buffer. The subprocess stdout/stderr reader therefore got EAGAIN despite kqueue reporting bytes ready, spun re-arming the poll, and only flushed on EOF.

Diagnosis chain (debug-instrumented): FilePoll dispatch fires → onPoll runs with fileType=socket, size_hint=9, streaming=true → recv returns errno 35 (EAGAIN) → swapping to read() delivered the 9 bytes and onReadChunk fired → traced the EAGAIN to MSG_WAITALL.

Fix: MSG_DONTWAIT = if (isMac) 0x80 else 0x40. Verified: for await (proc.stdout) streams incrementally from both a sh child and a live Home child; socket receive unaffected; bun-serve-body-json-async (#9222) 0/1 → 1/0.

Broad impact: fixes all nonblocking recv/send on macOS — unblocks the large "spawn a server fixture → read its stdout → connect" test category across test/js/bun/http, CLI, and node child_process suites. Also retroactively explains a chunk of the earlier http-server-chunking-style spawn timeouts.

Release-build survey after the MSG_DONTWAIT fix — strong state

Re-ran the spawn/http suites on a fresh ReleaseSafe binary:

http-server-chunking resolved as environmental: all 11 tests pass when run sequentially. The failures only appear under test.concurrent (10 subprocesses at once), where 10 × ~83 MB home spawns + server + client saturate this machine's 5 s budget. Not a Home correctness bug — the chunked parser (split \r\n terminators, split chunk-size CRLF, invalid-byte rejection) is correct. This is the "resource-exhaustion → false hang" class; it'll pass on CI.

Net: the runtime is in strong shape on release. The MSG_DONTWAIT fix (subprocess streaming) + NativePromiseContext (async handlers) cleared the two broad event-loop-ish blockers. Remaining real (non-environmental) items are narrower and deeper: HTML bundling (bun-serve-html 404s), body-stream typed-array clone edge case, ssl-ctx-cache GC reclaim, HTTP/2 protocol, shell segfault, napi.

🔴 CRITICAL (release-only): flaky segfault in in-process serve+fetch

Found while surveying on the ReleaseSafe binary. The debug build is stable; the release build intermittently segfaults (null+0xb0 deref) on the basic in-process Bun.serve + fetch(server.url) pattern.

Isolation (release binary):

• serve + sync handler + in-process fetch → crashes ~1 in 3 runs (flaky)
• serve + async handler (micro or macro) + in-process fetch → OK
• serve + async body stream + in-process fetch → crashes
• serve + external curl`` (no in-process client) → always OK
• response.test.ts (no serve/fetch roundtrip) → 14/0

So it's a timing-sensitive race when the fetch client and the serve server share one event loop, exposed by release-build speed. Real pin tests hit it: fetch-gzip, fetch-args, body-stream segfault on release (all were green on debug). bun-serve-static (34/0) and serve-listen (27/0) happen to dodge it.

Not attributable to this session's fixes: the crash is on the synchronous render path + in-process client; my changes touch config-parse (TLS), NativePromiseContext (async only — and async works here), and MSG_DONTWAIT (recv, which fetch doesn't use). All 8 fixes are debug-verified. Most likely a pre-existing latent event-loop race that only manifests under release timing.

Why it matters: this is probably the #1 reliability blocker for scoring the pin suite on release — a large fraction of tests do in-process serve+fetch. Needs an ASAN / debug-symbol release build to get a real stack trace (the optimized trace only shows a mis-unwound us_internal_dispatch_ready_poll). Flagging as critical; deferring the fix since it needs the ASAN toolchain, not a source guess.

Correction: the release race is general (not sync-specific), ~50-60% repro

Ran 12× each on the release binary:

• sync handler + in-process fetch → 7/12 crash
• async handler + in-process fetch → 6/12 crash

So both crash at the same rate — my earlier "sync crashes, async works" was flaky luck. This conclusively rules out the NativePromiseContext change (were it the cause, async would crash more, not equally). It's a general ~50-60% flaky segfault whenever a fetch client and its Bun.serve server run on the same event loop (loopback re-entrancy), only under release-build timing. External-client (curl) never crashes; debug never crashes.

Confirmed pre-existing and independent of this session's 8 fixes. Top reliability blocker for scoring the pin suite on release; needs an ASAN release build for a real trace. Practical workaround for now: score correctness on the debug build (stable), which is how all 8 fixes this session were verified.

✅ Primary release crash fixed (545c904) — HTTP Decompressor null-deref

Cracked the flaky release segfault to an exact line via lldb: ZlibReaderArrayList.deinit(this=0x0) on the HTTP Client thread. The Decompressor union's first tag is .zlib, so all-zero memory reads as .zlib with a null reader; AsyncHTTP.client is HTTPClient = undefined until init, and release-mode undefined can be zero, so InternalState.reset → decompressor.deinit over the not-yet-initialized state derefs null.

Guarded deinit against the null reader (teardown-only, correctness unaffected; a live reader is never null). Verified on release: rc_sync 0/15 crashes (was ~8/15), fetch-gzip 5/0, fetch-args 47/0 — all previously segfaulted.

A separate release crash remains: body-stream-excess now faults at address 0x0 (vs the decompressor's 0xb0) — different bug, investigating next. The deeper root (why HTTPClient teardown runs over undefined/freed state — a cross-thread lifecycle issue) is follow-up; the guard makes it non-fatal.

✅ Bun Shell core crash fixed (4166e25) + blob resolved

blob (was 15/1): the failure was an environmental leaked bun process holding port 3000 (from earlier differential runs), not a Home bug — 16/0 after killing it.

body-stream-excess / section B (shell): the JSC get_by_id-on-null crash was the Bun Shell, not body-stream. Cracked it via step-probing to JSArrayIterator's slow-path JSObject.getIndex, a stub returning error.JSError with no pending exception → the exception machinery deref'd the absent exception → segfault. A tagged template's strings.raw array is frozen (CopyOnWrite, skips the fast contiguous-vector path), so every $cmd`` crashed in createParsedShellScript → shellCmdFromJS → arrayIterator.next(). Wired the real jsc.JSObject.getIndex under -Denable_jsc.

Now working: $echo hello, substitutions (`$`echo ${42}), multi-arg builtins — the whole tagged-template + arg path. This also fixes any test that shells out via $.

Still broken (separate, deeper):

• External commands ($ls, `$`curl) crash with 0xdeadbee0 (use-after-free poison) in the shell subprocess-execution path — distinct from the builtin path. Blocks body-stream-excess (uses $curl``) and most test/js/bun/shell/*.
• $.braces returns a cyclic object (native binding mis-wired), not the expected expansion array.

Net: the fundamental shell (parse + builtin exec) is unblocked; external-command exec and the $.braces/lexer APIs are the next shell targets.

Shell external-command UAF — localized (deep, deferred)

$ls / `$`curl (any external command; builtins like echo work) crash with 0xdeadbee0 (bun.destroy poison). lldb (debug, preload-symbols off) pins it to:

thread #6 'Bun Pool 0', EXC_BAD_ACCESS address=0xdeadbee0
frame #0: event_loop.ConcurrentTask.from(this=0xdeadbee0, of=0x10af71660, .manual_deinit) at ConcurrentTask.zig:113

So a shell WorkPool task's embedded concurrent_task is freed (poisoned) before the pool thread posts its completion to the main loop — the receiver (this.concurrent_task) is poisoned while the payload of is a live allocation. The candidate sites all follow X.concurrent_task.js.from(ctx, .manual_deinit) on a pool thread: interpreter.zig:1768 (InnerShellTask WorkPool completion), states/Cmd.zig:74 (ShellAsyncSubprocessDone), IOReader.zig:282, IOWriter.zig:881. It's a cross-thread lifecycle race: the owning struct (Ctx/Cmd/IOReader) is torn down on the main thread while the pool thread is still finishing.

Pin's shell works, so it's a Home-specific lifecycle divergence (a missing keep-alive/refcount on the task owner across the WorkPool→main-loop hop). Unlike this session's bounded 'stub never wired' fixes, this needs careful multi-thread lifecycle analysis — deferring rather than guessing. Blocks external-command shell tests + body-stream-excess. The builtin shell path (parse + echo etc.) is unblocked by 4166e25.

Bun Shell — external commands unblocked (two more fixes)

Building on the tagged-template fix (4166e25), cracked two more layers of the shell:

1. External-command UAF — ConcurrentTask.from(this=0xdeadbee0) (2d9e47e). EventLoopTask's variants were *ConcurrentTask pointers initialized to a 0xdead_bee0 poison sentinel, whereas upstream holds the tasks by value. The shell's state.concurrent_task.js.from(ctx, ...) then wrote to the sentinel address on the WorkPool thread → every external command ($ls, `$`curl) crashed. Fixed: hold by value; enqueueTaskConcurrent takes the EventLoopTaskPtr pointer union (for allocate-then-enqueue callers like node_fs). Now $ls, `$`ls ${dir} work.

2. argv/env arena panic — attempt to use null value (c3e15f7). argv (Cmd.initSubproc) and env (fillEnv/spawnMaybeSyncImpl) reserved capacity for exactly the entries, then appended a null terminator; the grow called Zig 0.17's ArenaAllocator.resize, which does arena.loadFirstNode().? and panics on a foreign/empty arena (the old stdlib returned false + allocated fresh). Fixed: reserve the +1 terminator slot up front. body-stream-excess went panic→0/4, lex.test.ts crash→6/23.

Shell status: parse ✅, builtins (echo) ✅, simple external (ls, ls ${x}) ✅. Remaining: in-process $curl ${localServer}`` hangs (loopback: the shell awaits curl while the same-loop server must answer it — same class as the release in-process serve+fetch race), $.braces returns a cyclic object (mis-wired native binding), and lexer/other API tests. The core shell went from '100% crash on any $' to a working parse+exec path.

Three Home-vs-pin divergences in a row (stub pointer sentinel, Zig-0.17 stdlib behavior, JSArrayIterator stub) — none logic bugs.

Session update: cleared the reachable garbage-returning bindgen-dispatch stubs

Found and fixed a cluster of the "stub left in place after the real surface landed" class — this time in the bindgen v2 dispatch layer. The pinned-obj C++ wrappers (bindgen_*_js*) marshal JS args and call a Zig bindgen_*_dispatch* function to reach the real impl. Home had these dispatches noop-stubbed in native_stubs.zig, so the C++ wrapper called a void-returning noop → the return value / out-param was uninitialized garbage (a leftover register or stack slot, usually a pointer-sized number).

Each of these has a real, already-ported impl in Home; only the dispatch wire was missing. Real exports now live in jsc/js2native_workarounds.zig, mirroring the pin's generated dispatch shape.

Also un-stubbed the shell TestingAPIs.{shellLex,shellParse,disabledOnThisPlatform} host functions (same native_stubs cleanup). These work in the real runtime; the corpus harness still shadows bun:internal-for-testing with a JS shim lacking lex/parse, so lex.test.ts/parse.test.ts remain native-TODO until a JSC-C-API→internal-CallFrame bridge lands in the harness (documented inline in corpus_runner.zig).

bindgen_Fmt_jsc_dispatchFmtString1 was left stubbed on purpose: fmt_jsc.zig is still a genuine leaf (only the Formatter enum, no impl), it's only used by Bun's internal-for-testing.ts, and the corpus harness already provides highlightJavaScript/escapePowershell JS shims — so it's "not yet ported," not a stub-over-landed-surface.

Commits (all pushed to main): 44231fe73, 0d140e701, 9942474eb, 723213a4f.

Finding: using/await using disposal dropped in the corpus harness (js_parser reprint)

js/web/explicit-resource-management.test.ts fails on Expected none to be disposed. Root cause is not the runtime — Stage 3 explicit resource management works correctly under both home run and home -e:

home run  using.ts   → disposed ✓
home -e   'using…'    → disposed ✓   (await using → async-disposed ✓)

It fails only under home test. The corpus harness type-strips each .ts test by reprinting it through the Bun js_parser/js_printer port (transpileSourceWithBunParser in adapters/jsc_bootstrap.zig) before eval. That reprint turns

{ using u = a; }        // →
{ const u = a; }        // using qualifier + disposal silently dropped

Traced it down: the js_printer is correct (emits using/await using for k_using/k_await_using), and parse/parse.zig has the correct k_using branches (byte-identical to upstream Bun). But instrumentation shows a using-led statement in the reprint parse reaches neither parseExprOrLetStmt nor parseStmtFallthrough (both of which a let-led statement does reach), so it's parsed as a plain const — losing the k_using kind before it ever gets to the printer. using is a contextual keyword (correctly absent from the lexer Keywords map), so it should route through the t_identifier → parseStmtFallthrough → parseExprOrLetStmt path like let. The exact interception point in the js_parser reprint dispatch needs a deeper differential trace vs upstream — filing as a standalone follow-up.

Impact: any using/await using in a .ts corpus test loses disposal. Only the harness reprint path is affected (home run uses JSC's native parse). No runtime fix needed once the reprint parser preserves k_using.

Separately landed a small prerequisite: toBeSymbol() was missing from the harness expect (5 tests hit toBeSymbol is not a function before their real assertions) — added in 255d089d4.

Resolved: using/await using disposal (root cause found — it was the harness, not the js_parser)

Follow-up to the earlier finding. My initial suspicion (js_parser reprint routing) was wrong — the js_parser is fine. The real culprit: the corpus harness type-strips TS with a blanket needle/replacement table in corpus_runner.zig that replaced "await using " → "const " and "using " → "const " before the reprint parser ran. So the parser only ever saw const, and [Symbol.dispose]() disposal was silently dropped.

Found it by dumping the reprint input (not just output) — it already contained const u = a. Removing the two needle rows lets the js_parser parse using→k_using and lower it faithfully (try { const u = __using(...) } finally { … } with hygienic temps + working runtime dispose helpers).

js/web/explicit-resource-management.test.ts: fail → 4/0 (commit 502d6c9dd).

Verified no regression: abort/streams-leak (the other using-containing tests that now surface) already fail/hang on this box via await using Bun.spawn(...) independent of this change — confirmed against a stashed baseline. The two other using→const rewrites (minimum-release-age, archive) are test-scoped and left intact.

Separate follow-up gap: await using x = Bun.spawn(...) throws Object not disposable — Home's Subprocess lacks [Symbol.asyncDispose]. Worth wiring so subprocess resource-cleanup tests can pass once the spawn+pipe hang on this box is also addressed.

jest-extended matcher matrix: todo → 57/0 (commit bb95539)

Un-todo'd js/bun/test/jest-extended.test.js and closed the matcher gaps it exercises. Two kinds of fix:

Added missing matchers (faithful to Bun): toBeOdd, toBeEven (bigint-aware, Euclidean %), toBeNil, toBeBoolean, toBePositive/toBeNegative (Infinity/NaN excluded), toBeValidDate, toSatisfy (Bun-strict boolean result, wraps a throwing predicate).

Fixed bugs in existing matchers that the matrix surfaced:

• toBeArrayOfSize — used __home_fail on non-arrays, breaking every .not case
• toBeFinite — Number(value) coerced strings ("" → 0); now requires an actual number
• toBeWithin — end was inclusive; jest-extended is [start, end)
• toBeString — now accepts boxed String objects
• toIncludeRepeated — rejects -0 as the count (mirrors JSC isAnyInt)
• toEqualIgnoringWhitespace — validates string args instead of coercing

Debugging note: several matchers already existed later in the same prelude object literal (toIncludeRepeated/pass/fail), so my first-pass additions were dead duplicates (later key wins) — the real bug lived in the pre-existing copy.

expect.test.js (the sibling matrix) still needs .rejects/.resolves async-matcher infrastructure — left native-todo as a larger follow-up.

Systemic finding: corpus harness Engine ≠ home run VM (bare JSC context)

Several "works in home run, fails in home test" bugs share one root cause: the corpus harness evaluates test source in a bare JSGlobalContextCreate(null) context (packages/runtime/src/jsc/engine.zig) with Bun's environment reconstructed in JS (the harness_prelude), rather than a full Bun ZigGlobalObject. So it's missing the process-global JSC::Options that ZigGlobalObject.cpp sets — useSharedArrayBuffer, useShadowRealm, useV8DateParser, etc.

Concrete impact found this session:

• js/web/atomics.test.ts — SharedArrayBuffer is undefined in the harness, so the prelude's JS polyfill (backs it with a plain ArrayBuffer) kicks in, and Atomics.wait/notify reject the fake with "Typed array … must wrap a SharedArrayBuffer." (Real SAB + Atomics.wait work fine under home run.)

Tried calling jsc.initialize(false) (Bun's JSCInitialize, which sets these options) before the Engine context — it crashes the harness: JSCInitialize also installs crash/terminate handlers + main-thread init that assume the full Bun bootstrap. The right fix is a narrow C-ABI shim that sets only JSC::Options::useSharedArrayBuffer() = true (+ friends) before JSGlobalContextCreate, added alongside Home's existing napi_weak_home_dups.cpp build entry. Bounded but needs WebKit includes + ordering care — filing as follow-up.

Follow-up: async matcher generalization

expect.test.js (the full matcher matrix, sibling to the now-green jest-extended) needs .resolves/.rejects to forward all matchers. Today __home_make_resolves_expectation hardcodes 3 (toBe/toBeDefined/toBeUndefined) and .rejects hardcodes toThrow/toMatchObject with pending-count async tracking. Generalizing means a shared await+forward helper that preserves the __home_bun_tests.pending / __home_record_async_failure pattern — doable, medium risk to existing .rejects.toThrow tests.

Progress + a systemic observation (runtime is faithful; corpus blockers are harness shims)

Rebuilt the pinned Bun obj after a /tmp wipe (LLVM 21 + cmake + Bun C++ compile, ~1475 objects) so build+verify + differential vs the pinned bun binary work again.

Landed:

• Async matchers: .resolves/.rejects now forward ALL matchers (was 3/2 hardcoded) — 8d50a8eed.
• events.on: faithful async-iterator with AbortSignal listener cleanup added to the harness events shim; regression/issue/014187 unsupported → 1/0 — 4abf92f13.

Systemic finding (via differential vs pinned bun): Home's runtime is faithful on the paths I probed — Bun.inspect(Response/Blob), Blob size formatting (0 KB/N bytes/X.XX KB), and Subprocess[Symbol.asyncDispose] all match the pinned bun byte-for-byte under home run. The corpus failures for these are in the harness's JS reimplementations (the corpus evals in a bare JSGlobalContextCreate with Bun's env rebuilt in JS), e.g. the JS Bun.inspect shim (corpus_runner.zig:15108) diverges from native, and the SharedArrayBuffer polyfill can't back Atomics.

Implication: a large class of remaining corpus reds are harness-fidelity gaps, not runtime bugs. Wiring native Bun primitives into the harness context is blocked the same way using/SAB were — the bare context has no home VirtualMachine, so jsc.initialize (which would enable the real JSC options + native objects) crashes in RunLoopBun's VirtualMachine.get(). The higher-fidelity path is the HOME_NATIVE_VM=1 runner (real VM), which already runs many real Bun tests; the corpus-harness path needs per-API JS shims.

Real-runtime test flips via the native-VM sweep (noop-stub bindings)

Continuing to sweep real Bun tests with HOME_NATIVE_VM=1. The reliable tell for the fixable class is a Received: GlobalObject {...} in a failure — a bun:internal-for-testing binding that's noop-stubbed in native_stubs.zig and returns garbage. Fix = export the real host fn from js2native_workarounds.zig + drop the noop.

Flipped this pass:

• escapeRegExp.test.ts 0/2 → 2/0 (85ffd6480) — ported the jsEscapeRegExp{,ForPackageNameMatching} bridges from the pin's bun_string_jsc.zig.
• toUTF16Alloc.test.ts 0/6 → 6/0 (52e103cf4) — the UnicodeTestingAPIs.toUTF16AllocSentinel impl already existed; just wired it.
• regression/issue/014187 (corpus) → 1/0 (4abf92f13) — implemented events.on async iterator.

Wireable vs not: the fix only works when the underlying impl compiles+is correct. Un-wireable right now (impls have unported deps, so wiring forces a compile error or wrong output): ini/IniTestingAPIs (home.ini missing), patch/TestingAPIs (std.process.Child.init is old-Zig API), InternalSourceMap fromVLQ/toVLQ/find (wire cleanly but return wrong-shaped values — deeper impl bug), Counters (bridge "omitted"), highlightJavaScript (needs the unported fmt_jsc leaf). Others (getUseSystemCA, hosted_git_info TestingAPIs) have complete impls but no real test to flip.

Reinforces the earlier finding: Home's runtime is broadly faithful; the fixable reds cluster into (a) these noop-stub bindings with complete impls, and (b) genuinely unported leaves. Env artifacts (spawn hangs, missing npm devDeps, no bun-in-PATH) confound some native-VM results on this box.

ini subsystem wired + localized the top shell blocker

Landed: ini IniTestingAPIs wiring (16d606dc8) — the impl existed but didn't compile because home.zig never exposed bun.ini and strings.zig never re-exported escapeRegExp{,ForPackageNameMatching} (needed transitively by PnpmMatcher via ini.loadNpmrc). Exposed both + exported the host fns. iniInternals.parse("a = 1\nb = 2") → {a:"1",b:"2"} verified (1/0 in isolation).

Highest-leverage blocker, now precisely localized: Bun.$ with an external command HANGS. Bun.$/bin/echo x`` (any non-builtin) hangs — under home run with .text(), .quiet(), and inherited stdout. The child RUNS (output appears) but the shell's ShellSubprocess.onProcessExit NEVER FIRES, so the promise never resolves.

Ruled out, one by one: not the reader (inherited stdout hangs too); event_loop tag = js and it's the VM's main loop (interpreter.zig:811 globalThis.bunVM().event_loop, same instance Bun.spawn uses); is_sync=false; watch() + setExitHandler both called. And Bun.spawn + await p.exited works fine with the same bun.spawn.Process machinery. So the divergence is a subtle poll-dispatch/keep-alive detail in the shell path — prime suspects: the one-shot FilePoll not re-firing, or the shell disabling keep-alive before the child exits. Needs an lldb breakpoint on the kqueue/onWaitPid dispatch.

Why it matters: this single bug blocks essentially every test that shells out to a subprocess (ini, which, and the many tests whose fixtures use Bun.$${bunExe()} -e …``). Fixing it likely unblocks a large batch at once — the best next investment toward parity.

Shell external-command hang FIXED (616814377)

The long-standing hang on Bun.$<external command>`` is resolved. Root cause: home.zig's hand-rolled runtime.shell namespace defined ShellSubprocess as a no-op stub whose onProcessExit did nothing. process.zig binds bun.shell.ShellSubprocess into ProcessExitHandler, so every external command's EVFILT_PROC/NOTE_EXIT exit event dispatched to the empty stub — cmd.onExit never ran, the Cmd never advanced to .done, and the $ promise hung forever. Bun.spawn was unaffected because it uses the real Subprocess type.

Fix: alias the stub to the real type — pub const ShellSubprocess = subproc.ShellSubprocess;.

Verified (native-VM, Bun 1.3.14 runner):

• test/js/bun/ini/ini.test.ts — 52 pass / 0 fail (previously hung)
• test/js/bun/util/which.test.ts — 4 pass / 1 fail (previously hung; the 1 is which("bun") returning null because the harness binary is named home, not bun — an environment artifact, not a runtime bug)
• Bun.$/bin/echo …`` with .text(), .quiet(), and inherited stdout all complete cleanly

This unblocks the entire class of Bun tests that shell out to external commands via Bun.$.

Bun-shell external commands + brace expansion fixed

Follow-up to the exit-dispatch fix — a chain of shell bugs is now resolved, taking bunshell.test.ts from a hard crash to 352 pass / 3 fail:

1. Env null-terminator arena panic (d525e5f) — the env array's null terminator was appended on spawnAsync's freshly-created arena; Zig 0.17's ArenaAllocator.resize panics (loadFirstNode().?) growing an arena that hasn't allocated yet. Now copies into a fresh null-terminated buffer when there's no spare slot.
2. SystemError.getErrno stub enum (bad6ec1) — SysE was a placeholder enum { SUCCESS = 0, _ }; the shell IOWriter passes getErrno() to @tagName, which panics in debug on any unnamed errno (e.g. EPIPE=32 on a broken pipe). Aliased to the real bun.sys.E.
3. Brace expansion buffer poison (eff6913) — echo {a,b,c}{d,e,f} produced garbage argv. Expanded words are moved into argv by pointer, then the source list was reset with clearRetainingCapacity(); Zig 0.17 changed that to @memset(items, undefined) first, poisoning the aliased buffer with 0xAA. Re-init the list instead. All 11 brace-expansion tests pass.

Verified (native-VM): bunshell-instance.test.ts 16/0, ini.test.ts 52/0, bunshell.test.ts 352/3. Remaining 3: deno_task broken-pipe subproc, deno_task stacktrace, subshell 'sharp' — separate issues.

fmtString binding wired (highlightJavaScript / escapePowershell)

Another noop-stub-garbage fix (8a5cc41). bun:internal-for-testing's highlightJavaScript threw fmtBinding is not a function ... is an instance of GlobalObject — both the lazy binding (js2native_bindgen_fmt_jsc_fmtString) and the dispatch (bindgen_Fmt_jsc_dispatchFmtString1) were noop-stubbed, so $bindgenFn('fmt_jsc.bind.ts','fmtString') returned globalThis. Exported both real symbols in js2native_workarounds.zig (lazy binding → NewRuntimeFunction over the pin's C++ jsFmtString; dispatch → bun.fmt.fmtJavaScript + inlined escapePowershell). The Formatter enum values must match the pin's generated order (escape_powershell=0, highlight_javascript=1) since the linked C++ marshals the string arg to those numeric values.

Verified (native-VM): highlighter.test.ts 1/0, test/internal/powershell-escape.test.ts 1/0 — both were failing.

Session tally: 5 runtime fixes (shell exit dispatch, env terminator, SysE errno, brace-expansion poison, fmtString binding).

More noop-stub test bindings wired — big batch

The native-VM 'Received: GlobalObject {…}' tell keeps finding noop-stubbed bun:internal-for-testing bindings whose real impls already exist in Home. Un-stubbing them (export real host fn via host_fn.toJSHostFn, drop the noop):

• hostedGitInfo fromUrl/parseUrl (db43e8c) — hosted-git-info/from-url.test.ts 0/649 → 649/0
• npm Architecture/OperatingSystem isMatch (77703ed) — architecture-match.test.ts 0/30 → 30/0
• InternalSourceMap fromVLQ/toVLQ/find (02ac692) — internal-sourcemap-roundtrip.test.ts 0/3 → 2/3
• fmt_jsc fmtString (8a5cc41) — highlighter.test.ts + powershell-escape.test.ts now green

patch TestingAPIs stay stubbed — patch.zig's makeDiff uses Zig-0.16 std.process.Child.init (removed in 0.17), so the impl doesn't compile; needs a port first.

Session total: 8 runtime fixes; ~680 tests flipped green. The recipe (native-VM run → grep for GlobalObject/not-a-function → check impl exists → wire → drop noop) is highly productive; more candidates remain (dependency, css_internals, counters).

patch apply/parse wired (2/18 → 7/13)

Continuing the deep-subsystem work. patchInternals.{apply,parse} were noop-stubbed; both are pure file/parse ops, so wiring them (d1e5d22) took patch.test.ts 2/18 → 7/13.

The remaining 13 use patchInternals.makeDiff, which shells out to git diff. Home's makeDiff still uses Zig-0.16 std.process.Child.init (removed in 0.17). Home already has the right building blocks (patch.zig spawnOpts → bun.spawn.sync.Options, diffPostProcess), and I rewrote gitDiffInternal onto bun.spawn.sync.spawn — but that entry point (spawnWithArgv) is itself an unported Zig-0.17 chunk: it cascades through std.posix.kqueue (gone → std.c.kqueue), crash_handler.resetOnPosix (missing), and std.c.SIG becoming an enum (+likely more) in the sync signal-forwarding path. It's dead code today (never compiled), so porting it is low-risk but a bounded task of its own — deferred rather than rushed on the critical spawn path.

Session tally: 9 fixes (shell cascade ×4, noop-stub bindings ×4, patch apply/parse). Remaining blocked-until-subsystem-ported: makeDiff (sync-spawn), css_internals (css.StyleSheet), dependency (semver.toJS), Home-lang TS named imports.

css_internals wired — color.test.ts 916/1 (~920 tests flipped)

Deep-subsystem work paying off. Wired the 8 cssInternals test hooks (92fcaea): needed home.css to re-export the generic StyleSheet+DefaultAtRule+LocalsResultsMap+StyleAttribute (all in css_parser.zig; only BundlerStyleSheet was exposed) and to un-stub css/error.zig's toErrorInstance (re-export the real bridge from css_jsc/error_jsc.zig). Result: color.test.ts 916/1, regression/27598 0/5→5/0, css-system-color-contexts 0/1→1/0, color-mix un-crashed.

Localized the remaining css.test.ts (~1029 fail): the css PARSER is fine (.a{color:red} → rules.len=1), but css/rules/style.zig StyleRule.toCss is a no-op stub. I ported the pin's real toCss/toCssBase (~120 lines; all direct deps exist and work). It's blocked one layer deeper: a set of css value serializers (values/{css_string,time,number,resolution}, properties/{svg,shape}, rules/unknown) still @import a css_parser_stub.zig (separate stub Printer type) with @compileError toCss — a deliberate dependency-cycle break. The selector-serialize path references them at compile time → Printer-type mismatch. Un-stubbing that value-serialization layer (switch imports to the real css_parser + port their toCss) is the next lever (~1029 tests).

Session tally: ~10 fixes, ~1600 tests flipped (hostedGitInfo 649, css color 916, patch apply/parse, arch-match 30, + the shell cascade).

css style rules now emit — css.test.ts 3 → 574 pass (+571)

The css value-serializer stubs turned out to be stale (Wave-7 vintage; the deps they waited on are now ported), so the deep css port was tractable. Fixed three combined gaps (91ade3b) that made every css style rule print empty:

1. Parser threw away its own work — css_parser.zig QualifiedRuleParser.parseBlock built StyleRule{ .selectors={}, .declarations={}, .rules={} }, discarding the parsed selector prelude + declaration block + nested rules. (The smoking gun.)
2. StyleRule.toCss was a no-op stub — ported the pin's real toCss/toCssBase.
3. css/values/css_string.zig imported the stale css_parser_stub (separate Printer type, @compileError toCss) — un-stubbed to the real css_parser.

Results: css.test.ts 3 → 574 pass, color.test.ts still 916/1 (no regression), doesnt_crash 8→12.

Remaining css.test.ts (458) is a diverse long tail — shorthand recombination (border → 4 longhands), declaration ordering, vendor-prefix pseudo-classes — each a per-property handler, not one systematic bug.

Session total: ~11 fixes, ~2200 tests flipped green (hostedGitInfo 649, css color 916 + css.test 571, arch-match 30, patch apply/parse, the shell cascade).

At-rule printers — @media/@viewport/@starting-style now emit

The css style-rule bug had a twin in the at-rule printers: rules.zig CssRule.toCss routed several variants to a no-op {}, the parser discarded its own parsed content (_ = switch(...); break :brk .{}), and the rule files imported the stale css_parser_stub.

• @media (0c5da23) — MediaRule.toCss hardcoded @media all; ported the pin's real toCss. css.test.ts 574→622, doesnt_crash 12→44 (+80; @media is everywhere).
• @viewport + @starting-style (f157706) — enable dispatch, use the parsed declarations/nested-rules, real import. Correct (rare in css.test.ts).

Still stubbed: @scope (no toCss ported), @layer / @supports (deeper — their prelude/condition grammar is unimplemented, not just the printer: @layer base and @supports (display:grid) throw parse errors).

Session total: 14 fixes, ~2280 tests flipped (hostedGitInfo 649, css color 916 + css.test 3→622, at-rules, arch-match 30, patch, the shell cascade). css.test.ts remaining (410) is a diverse long tail — vendor-prefix pseudo-classes, targets-based shorthand expansion, and the @supports/@layer grammar ports — each a focused per-feature fix.

css minify pipeline enabled — css.test.ts 622 → 868

Found the single lever behind most of the remaining css.test failures: StyleRule.minify + updatePrefix were no-op stubs, so the entire declaration-minify pipeline (targets-based shorthand collapse, logical-property expansion, selector downleveling) never ran for style rules — even though CssRuleList.minify already calls them and the whole chain (StyleSheet.minify → handlers → DeclarationBlock.minify) is real.

• StyleRule.minify (4f86449) — port the pin's (declarations.minify via handler context + recurse nested rules); also fixed PropertyHandlerContext.child initing ArrayList fields with .{} (Zig-0.17 needs .empty). css.test.ts 622 → 851 (+229).
• updatePrefix (4d74aff) — port selector.getPrefix + downlevelSelectors. 851 → 868 (+17).

color.test.ts still 916/1 (no regression).

Session total: 17 fixes, css.test.ts 3 → 868 (out of 1099). Remaining 164 dominated by per-property vendor prefixing (-webkit-/-moz- never emitted — most prefixable properties lack a handler, a large per-property feature), plus @supports/@layer grammar and color edge cases.

@supports parser ported — doesnt_crash 44 → 56

Ported the @supports condition parser (f7bbb65). SupportsCondition.parse/parseDeclaration were stubs that returned an empty condition without consuming the (...) tokens, so @supports (…) left a stray ( and threw 'Unexpected token: open_paren' — failing every real-world CSS file (bootstrap/uikit/pico) that uses @supports. Ported the pin's recursive parser (not/and/or/declaration/selector/parens) + un-stubbed SupportsRule.toCss. All deps already existed; the @supports rule construction already populated condition+rules (no discard bug there).

doesnt_crash.test.ts 44 → 56, @supports rules now round-trip. color.test.ts still 916/1.

Session total: 18 fixes. css.test.ts 3 → 868, doesnt_crash 8 → 56, color 916/1, plus hostedGitInfo (649) and the shell cascade. Remaining: per-property vendor-prefix handlers (css.test long tail), @layer prelude parser, and one elusive pico.css selector construct.

@layer + @supports conditions — 20 fixes this session

• @layer (21fac82) — ported LayerName.parse (was a stub that didn't consume → 'Unexpected token: base') + LayerBlock/Statement toCss + dispatch. css.test 868→871.
• @supports condition toCss (205ffa4) — was emitting @supports lab(0% 0 0) instead of @supports (color: lab(0% 0 0)) (declaration written as bare value). Now emits (name: value), selector(…), and separate and/or. (Remaining @supports css.test diffs are lab()-color body formatting, not the condition.)

Session total: 20 fixes. css.test.ts 3 → 871 (of 1099), doesnt_crash 8 → 56, color.test.ts 916/1, plus hostedGitInfo (649) and the shell-hang cascade. The at-rule printer pattern (dispatch {} + parser discards content + stale stub import) is now cleared for media/viewport/starting-style/scope/supports/layer, and the declaration-minify pipeline (prefixing/collapse/expansion) is live.

Remaining css.test (161): per-property vendor-prefix handlers (the biggest bucket — infrastructure proven, each property a bounded port), lab()/color body formatting, declaration ordering. Remaining doesnt_crash (4): one elusive pico.css selector construct.

CSS minify parity — session progress (css.test.ts 871 → 958, doesnt_crash 60/0)

Landed 7 fixes this session, all pushed to main:

1. color-fallback @supports condition — ColorFallbackKind.supportsCondition built .unknown (raw text) so lab()/color() fallbacks printed @supports lab(0% 0 0) instead of @supports (color: lab(0% 0 0)). Now builds a .declaration. (+11)
2. drop empty style rules — StyleRule.isEmpty() was stubbed return false, so logical-property expansion (border-inline → lang-scoped rules) left a stray .foo {} block. Ported the real isEmpty. (+23)
3. <time> dimension parse — Time.parse read a bare number, so 2s/500ms (dimension tokens) failed → .unparsed, which silently broke transition longhand→shorthand recombination. Ported the pin's dimension parse. (+6)
4. <time> shorter-unit serialize — 500ms → 0.5s. (+2)
5. <number> leading-zero trim — CSSNumberFns.toCss used bare {d}, emitting 0.5 where CSS wants .5. Ported dtoa_short + trimLeadingChar / to_css.float32. Broad win across every fractional value; doesnt_crash.test.ts → 60/0. (+40)
6. calc() in <time> — calc(1s - 50ms) → .95s. (+5)

css.test.ts: 871 → 958 / 1032. color.test.ts: 916/1 (no regression).

Remaining css.test failures (74) are now a diffuse long-tail: per-property flex/box vendor prefixing (-webkit-box-orient, -webkit-flex-*), a few lab()/oklch() color-format edge cases, and misc declaration-ordering.

CSS minify parity — css.test.ts 871 → 1027 / 1032 (color 916/1, doesnt_crash 60/0)

Continued the session with 8 more fixes (14 total), all pushed to main. The remaining 3 handlers/value-serializers that were stubbed are now ported:

7. flexbox handler — FlexHandler was a stub; ported the full handler (longhand↔shorthand recombination via prefix intersection, 2009 box-* / 2012 -ms-flex-* legacy fallbacks). (+19)
8. legacy fromStandard conversions — BoxPack/FlexPack/BoxAlign/FlexItemAlign/FlexLinePack.fromStandard returned null, so the (correct) AlignHandler emitted standard -webkit-justify-content but no box-pack/flex-pack legacy forms. (+7)
9. flex shorthand serialize — collapse 1 1 auto→auto, 0 0 auto→none. (+10)
10. box-shadow handler — stub → full port with rgb/p3/lab color fallbacks + prefixing. (+9)
11. box-shadow zero-blur — inset 12px 12px 0 8px no longer drops the zero blur.
12. easing functions — cubic-bezier(...)/steps(...) were unparseable (function tokens) → round-tripped raw; ported parse + keyword-normalizing toCss (cubic-bezier(.25,.1,.25,1)→ease, step-start/step-end). (+17)
13. flex parse/flex-flow — default basis 0% (not auto), none keyword; flex-flow: row wrap→wrap. (+7)

Earlier in the session (fixes 1–6): color-fallback @supports condition, drop-empty-style-rules, <time> dimension parse, <time> shorter-unit + <number> leading-zero trim, calc() in <time>.

Net: css.test.ts 871 → 1027 (+156 this session). doesnt_crash.test.ts → 60/0. color.test.ts steady at 916/1.

Remaining 5 css.test failures are genuinely distinct edge cases, no longer a systematic bucket:

• declaration ordering (height:infinity placement in a calc edge case)
• @media not all should minify to empty (never-matches drop)
• transform: scale(calc(.1 + .2)) should fold to scale(.3)
• 2× non-ASCII identifier/string serialization (emoji @keyframes name, astral-plane content) — a UTF-8 handling issue in the serializer.

css.test.ts 1031 / 1032 ✅ (color 916/1, doesnt_crash 60/0 — all green)

Two more fixes land the CSS minify suite at 1031/1032:

14. UTF-8 in String.fromBytes — used ZigString.init (Latin-1), so multi-byte UTF-8 in a CSS result (emoji @keyframes names, astral-plane content strings) reached JS as one char per byte → mojibake. Switched to ZigString.fromBytes (the pin's UTF-8-detecting variant). Runtime-wide correctness fix. (+2)
15. calc() in <number> — CSSNumberFns.parse did only expectNumber, so transform: scale(calc(.1 + .2)) round-tripped unfolded (while translate(calc(...)) folded via Calc(Length)). Ported the pin's Calc(f32) branch. (+1)

Session total: 17 fixes, css.test.ts 871 → 1031 (+160). color.test.ts steady 916/1, doesnt_crash.test.ts 60/0 throughout.

The single remaining css.test failure is a narrow declaration-ordering edge case (issue #18064's .rounded-full): height: infinity (unparsed, emitted in-place) should sort before the calc(infinity * …)-folded border-radius/width (which their handlers buffer and flush at finalize). Home emits border-radius in place rather than buffering it, so the order differs. It's a deep handler-buffering-timing nuance affecting only this fold-to-infinity case — left for a focused follow-up.

✅ 100% CSS parity — css.test 1032/1032, color 917/0, doesnt_crash 60/0

Two final items close the suite to a full pass:

16. border-radius handler — BorderRadiusHandler was the last stubbed handler (handleProperty→false, finalize→noop), so border-radius passed through in place instead of being buffered and flushed at finalize. That caused the sole remaining css.test failure (issue #18064's .rounded-full calc-edge, where border-radius mis-sorted ahead of an in-place height: infinity). Ported the pin's full handler (corner accumulation, shorthand recombination, logical border-*-radius, 2009/2012 prefixing) + isBorderRadiusProperty/isLogicalBorderRadiusProperty + Size2D.isCompatible. → css.test.ts 1032/1032.

17. fuzz ansi256 was a debug-build timeout, not a defect — the test runs 16.7M Bun.color(int, "ansi256") conversions and asserts none are null. In the debug build it produces 100% correct values but takes ~10.5s, exceeding the harness's 5s wall clock. Built -Doptimize=ReleaseFast (how Bun ships/runs its official suite): the loop finishes in 865ms and color.test.ts is 917/0. Confirmed css.test (1032/0) and doesnt_crash (60/0) also pass in release — no release-specific regression.

Final CSS-suite state (release build)

Full-session arc: css.test.ts 871 → 1032 (17 fixes). The whole style-rule / at-rule / property-handler / value-serialization layer that was stubbed at session start is now ported to the pinned Bun (fd0b6f1a27) behavior.

✅ Bun.markdown 100% — render() and react() ported (whole md suite 1022/1022)

Bun.markdown.render() and Bun.markdown.react() were the last throwTODO stubs. Ported the pin's two renderers:

• JsCallbackRenderer (render) — content-stack renderer that calls the JS callbacks (heading/paragraph/list/…) for each element with the accumulated children string + a metadata object, appending each callback's return value.
• ParseRenderer (react) — builds an object AST: real React elements in React mode (via the C++ fast path) or {type, props} objects in plain mode, with component overrides keyed by HTML tag.

Wired the pinned-obj C++ helpers as direct extern "C" fns (verified present in build/release/obj via nm) — BunMarkdownTagStrings__getTagString, BunMarkdownMeta__createList/createListItem/createCell/createLink, JSReactElement__create/createFragment — replacing the .zero soft-link stub. Also exposed md.helpers/md.types/md.BLOCK_FENCED_CODE, dropped the local BlockType/SpanType mirror enums (use md.\*), and handled Zig-0.17 (#-fields → plain, ArrayListUnmanaged .empty).

markdown test suite (all files now pass)

+116 tests. No regressions: css 1032/0, color 917/0. Combined with the earlier renderToAnsi work, the whole Bun.markdown API (html/ansi/render/react) now matches the pinned Bun.

Bun corpus porting update:

• Advanced copied Bun YAML corpus slot 739 from CXX2 through DK4H; current first failure is yaml-test-suite/DK95/04 (tab indentation handling).
• Pushed commits:
  • d1955c1 fix: port bun yaml document start anchor guard
  • 44e63a3 fix: port bun yaml trailing tab escapes
  • 79ce308 fix: port bun yaml flow null keys
  • 14961f8 fix: port bun yaml flow newline key guard
• Verification used the focused gate: HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures.

Next frontier: yaml-test-suite/DK95/04.

Bun corpus porting update:

• Advanced copied Bun YAML corpus slot 739 from DK95/04 through directive stream cases; current first failure is yaml-test-suite/E76Z (alias resolution in mapping values).
• Pushed commit:
  • d7094fc fix: port bun yaml tab directive streams
• Verification used the focused gate: HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures.

Next frontier: yaml-test-suite/E76Z.

Bun corpus porting update:

• Advanced copied Bun YAML corpus slot 739 from E76Z to EW3V. Current first failure is yaml-test-suite/EW3V (expected throw).
• Pushed commit:
  • d72ddae fix: port bun yaml key anchors
• Verification used the focused gate: HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures.

Next frontier: yaml-test-suite/EW3V.

Checkpoint pushed: 1a0d174 (fix: port bun yaml mapping indentation guard).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/EW3V to yaml-test-suite/FH7J.
• Change size: 2 harness lines flipped total, 2 insertions in packages/home_test/src/corpus_runner.zig.
• Behavior ported: completed scalar mapping values now reject a following indented mapping-looking continuation, matching Bun EW3V syntax error behavior.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/FH7J with null tag/key/value semantics mismatch.

Checkpoint pushed: 307222d (fix: port bun yaml empty node tags).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/FH7J through yaml-test-suite/FRK4 to yaml-test-suite/FTA2.
• Change size: 19 harness lines flipped total, 15 insertions and 4 deletions in packages/home_test/src/corpus_runner.zig.
• Behavior ported: empty tagged !!str nodes materialize as empty strings, empty tagged !!str/!!null keys normalize like Bun, and whitespace-only flow mapping keys resolve to YAML null keys.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/FTA2, where an anchored document-start sequence currently parses as null instead of ["a"].

Checkpoint pushed: c62ac95 (fix: port bun yaml document anchor blocks).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/FTA2 to yaml-test-suite/G4RS.
• Change size: 7 harness lines flipped total, 7 insertions in packages/home_test/src/corpus_runner.zig.
• Behavior ported: document-start anchors/tags that contain only YAML props now attach to the following root block, so a document like --- &sequence followed by a sequence parses like Bun.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/G4RS, where double-quoted escape handling currently collapses escaped CR/LF sequences incorrectly.

Checkpoint pushed: f8067bc (fix: port bun yaml quoted escapes).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/G4RS to yaml-test-suite/G5U8.
• Change size: 2 harness lines flipped total, 1 insertion and 1 deletion in packages/home_test/src/corpus_runner.zig.
• Behavior ported: single-line quoted scalars now preserve escaped LF and escaped CRLF output instead of folding decoded newlines as if they came from multiline YAML source.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/G5U8, an expected-throw YAML case.

Checkpoint pushed: a1ee70a (fix: port bun yaml flow dash rejection).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/G5U8 to yaml-test-suite/H7TQ.
• Change size: 1 harness line flipped total, 1 insertion in packages/home_test/src/corpus_runner.zig.
• Behavior ported: plain dash entries inside flow sequences now throw like Bun instead of being accepted as string scalars.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/H7TQ, another expected-throw YAML case.

Checkpoint pushed: 1851ac6 (fix: port bun yaml directive arity).

Frontier update:

• Copied Bun corpus gate js/bun/yaml/yaml-test-suite.test.ts now advances from yaml-test-suite/H7TQ to yaml-test-suite/HS5T.
• Change size: 6 harness lines flipped total, 6 insertions in packages/home_test/src/corpus_runner.zig.
• Behavior ported: %YAML and %TAG directives now reject extra or missing arguments like Bun, including %YAML 1.2 foo.

Verification:

• env HOME_BUN_CORPUS_FULL=1 HOME_BUN_CORPUS_PROGRESS=1 HOME_BUN_CORPUS_START=739 HOME_BUN_CORPUS_LIMIT=1 timeout 1200s ./pantry/.bin/zig build test -Dfilter=home_test --global-cache-dir .zig-cache/global --summary failures
• Result: expected focused gate failure moved to yaml-test-suite/HS5T, where our tab-indentation guard rejects a Bun-accepted YAML case.