Repository navigation
Conversation
de366cc to
d35e72e
Compare
4344c8d to
1f043f3
Compare
8c2f9d8 to
471310d
Compare
decf029 to
c68f402
Compare
The packager now maps Darwin x86_64 to the darwin-amd64 name already used in release-platforms.json. The Intel smoke job prints that name and still withholds the tarball from GitHub Releases. Co-authored-by: Sash <SashaMIT@users.noreply.github.com>
darwin-amd64 is the package name. release-platforms.json maps it to x86_64-darwin. The signer allowlist still refuses that release id, so the name is not a signed download. Co-authored-by: Sash <SashaMIT@users.noreply.github.com>
471310d to
18a2a66
Compare
c68f402 to
3e52e38
Compare
Full packaging stops on darwin-amd64 before it writes a tarball, including a Rosetta shell. --print-platform still prints the name and writes nothing. The signer allowlist is one frozenset, and an optimized interpreter run refuses x86_64-darwin without writing a signature. Co-authored-by: Sash <SashaMIT@users.noreply.github.com>
The base now requires HTTP 200 from the process that owns the host lock. docs/MAC.md keeps that text and this branch's package key paragraph. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
The old check ran prepare() in an optimized child and only looked for "ok". The new child runs main() with python -I -S -O on an x86_64-darwin input that passes every other gate. It must stop with "release platforms refused", build no key backend and write no output. A second case turns that refusal into an assert statement and proves the first check fails under -O. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
A Rosetta shell on Apple silicon reports uname -m x86_64. The packager now names a host darwin-amd64 only when sysctl shows Intel hardware. It stops when sysctl is missing, fails, or gives an unknown answer. Before it writes a tarball, each staged binary must pass the native header rule in scripts/release-platform-input.py for the package key. A universal binary, or one built for another processor, stops packaging. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
docs/MAC.md#intel-mac now names the file that holds each Intel rule: signing, package name, packaging, and install. README, INSTALL, GETTING_STARTED and the launcher contract link there instead of restating Intel support. The signer constant name is no longer copied into the doc. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 No conflicts: docs/MAC.md keeps the one Intel Mac section and takes the base's localhost:8090 readiness text. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 zip 9 for RUSTSEC-2026-0367/0368/0369 so cargo-audit passes. The base already merges fix/zip-9-rustsec; merging the base keeps this PR's diff against it unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 Brings in the ported held-lock journal test fix (develop 5051505) so test-capsules stops flaking. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 Brings in the macOS fix for the gateway smoke test stub, the step that runs those tests on the Intel Mac runner, and the longer cleanup-test deadline. No conflicts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 Brings in the MAC.md correction: the gateway listens on every localhost address. No conflicts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…ame-1297 Brings in #394's proof of gateway lock ownership from the held flock (3672007), through #395. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
|
Round 3 review: PASS (B). Body matches the head. Waiting on the #398 decision. |
…ame-1297 Brings in #394's one-step flock ownership proof, the subordinate lane on the same proof, and a held lock with an unreadable record counting as held (4bb19dc), through #395. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
Brings in #394's one-step flock ownership proof, the subordinate lane on the same proof, and a held lock with an unreadable record counting as held (4bb19dc), through #395 and #397. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
|
CI on 56a8084: only The same run, which carries #394's Generated by Claude Code |
…ame-1297 Brings in, through #395, #394's statement that the host lock proves possession, not that the holder is the gateway (13cdcad), and #395's hermetic default-command smoke test (630efc6) and "holds" wording (ac97a4a). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
Brings in, through #397 and #395, #394's statement that the host lock proves possession, not that the holder is the gateway (13cdcad), and #395's hermetic default-command smoke test (630efc6). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
Brings in #398's portable swapped-Runtime test (da63c23: the Linux case mocks a GNU-like mktemp instead of using the host's), and, through #397 and #395, #394's statement that the host lock proves possession, not that the holder is the gateway (13cdcad). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
…-root Brings in #398's portable swapped-Runtime test (da63c23: the Linux case mocks a GNU-like mktemp instead of using the host's), and, through #397 and #395, #394's statement that the host lock proves possession, not that the holder is the gateway (13cdcad). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K
|
Round 5 at aa7bd1c: PASS (A). Genuine merge, own diff unchanged; packaging suite 9/9 on macOS. |
Summary
Stack: #394 ← #395 (
cursor/intel-mac-gateway-smoke-1297) ← #397 (this) ← #398 ← #399. #399 replaces this branch's signer allowlist with its ownRELEASE_PLATFORMS.scripts/package-release-binaries.sh --print-platformprintsdarwin-amd64only whensysctlshows Intel hardware (check_native_intel_mac, lines 14-37). A Rosetta shell on Apple silicon (sysctl.proc_translated1 orhw.optional.arm641) stops with "This shell runs under Rosetta on an Apple silicon Mac. Open a native arm64 shell." A missingsysctl, a failingsysctl, or an unknown answer also stops. Full packaging still refusesdarwin-amd64before it creates a tarball (lines 63-65).check_native_headerinscripts/release-platform-input.py(the same rule the release input verifier uses), keyed throughscripts/release-platforms.json(lines 109-123). A universal Mach-O, a binary for another processor, or a non-executable file stops packaging.x86_64-darwinwithrequire()(aValueError, notassert). The allowlist is atscripts/release-signer.py:104and:112-114. It runs before provenance at:1054and again at:1095.docs/MAC.md#intel-macis the one Intel Mac record. README,docs/INSTALL.md,docs/GETTING_STARTED.mdanddocs/LAUNCHER_CONTRACT.mdlink to it. It names the file that holds each rule. It no longer copies the signer constant name.scripts/package-release-binaries.sh --print-platformprintsdarwin-amd64on the runner (.github/workflows/ci.yml, "require an Intel Mac with room to compile").What this branch carries
Head:
aa7bd1ce. Only merges of the base, never rebases. None of them changes this PR's diff against #395 (12 files, +463/-10).09092a41,7b8bffc1: earlier Smoke the gateway health probe on Intel Mac #395 heads (the/healthzowner check andlocalhost:8090readiness text).7b8bffc1had no conflicts:docs/MAC.md#intel-mackeeps this branch's one Intel section.b4316341: Smoke the gateway health probe on Intel Mac #395 at028c6137, which carries the zip 9 fix (RUSTSEC-2026-0367/0368/0369) from Define the launcher contract for gateway supervision #394.37101760: Smoke the gateway health probe on Intel Mac #395 at3ab6cb82, which carries Define the launcher contract for gateway supervision #394's cherry-pickf1fcb219of develop50515055(held-lock journal test takes its lock with the bounded retry) and Define the launcher contract for gateway supervision #394'sce5d187b(the gateway listens on every localhost address).3cffb7aa: Smoke the gateway health probe on Intel Mac #395 atc3812551: the gateway smoke test stub keepshost-process.lockwhere the real binary does on macOS, the Intel job runs those tests on the Mac, and two cleanup tests get a 3 second deadline. No conflicts.03d22f0f: Smoke the gateway health probe on Intel Mac #395 at3cffaad4:docs/MAC.mdno longer says--addr localhost:8090binds one address. No conflicts.8dd9e851: Smoke the gateway health probe on Intel Mac #395 ated21329f, which carries Define the launcher contract for gateway supervision #394's36720070:/healthzand the gateway start prove that the gateway holds the lock from the held flock, not the JSON record inhost-process.lock(docs/LAUNCHER_CONTRACT.md, "What proves that this gateway holds the lock").docs/LAUNCHER_CONTRACT.mdauto-merged. No conflicts.56a8084f: Smoke the gateway health probe on Intel Mac #395 at12fc78b2, which carries Define the launcher contract for gateway supervision #394's4bb19dcf: the ownership proof is one step (each registry entry keeps a second descriptor of the locked open file, and the registry mutex is held from the lookup to the end of the probe;85f3142e); the managed-runtime subordinate lane uses the same proof as/healthz, and a Home terminal gets an inherited lock descriptor,ELASTOS_GATEWAY_HOST_LOCK_FD(eed83a72); a held lock whose record cannot be read still counts as held (4bb19dcf). No conflicts.aa7bd1ce: Smoke the gateway health probe on Intel Mac #395 atac97a4aa, which carries Define the launcher contract for gateway supervision #394's13cdcad0(the contract states that the host lock proves possession, not that the holder is the gateway, also forELASTOS_GATEWAY_HOST_LOCK_FD; docs and tests only) and Smoke the gateway health probe on Intel Mac #395's630efc6d(the default-command smoke test no longer depends on port 8090) andac97a4aa("holds", not "owns", in the smoke header and CI comment).docs/LAUNCHER_CONTRACT.mdand.github/workflows/ci.ymlauto-merged. No conflicts. The contract sentences that Define the launcher contract for gateway supervision #394's Rust tests check are all present in the merged contract (checked by text).Candidate:
aa7bd1ce. Draft. Do not merge. The exit-code policy lives only on pull request #407.Tests
All on
3cffb7aa, Linux x86_64 (03d22f0fchanges onlydocs/MAC.md; on itpackage-release-binaries-test.py,ci-release-policy-test.pyandgateway-healthz-smoke-test.py -k contractpass again; on8dd9e851package-release-binaries-test.pyandci-release-policy-test.pypass again; on56a8084fpackage-release-binaries-test.pypasses again; onaa7bd1cepackage-release-binaries-test.py9 OK,release-signer-test.py87 OK underenv -i ... python3 -I -S,ci-release-policy-test.py103 OK andgateway-healthz-smoke-test.py26 OK (3 skipped) pass again):Against the packaging script at
44b5b20d, the same file givesFAILED (failures=15): every Rosetta, missing or unclearsysctl, and wrong-processor case was accepted.The child runs
main()underpython -I -S -Oon anx86_64-darwininput that passes every other gate (Intel component keys are admitted in the fixture). The person confirms with the right DID, so a missing refusal would reach the key backend. Outcomes:With the refusal changed to an
assertin the real file,test_optimized_cli_refuses_x86_64_darwin_and_signs_nothingfails:'ValueError: package artifact prefix refused' != 'ValueError: release platforms refused'. The later artifact-prefix gate still stops that input, so nothing was signed. The test names the refusal it expects.Also on
3cffb7aa:release-signer-test.py87 OK underpython3 -I -Swith no TLS override variables set (the signer refusesSSL_CERT_FILEand similar, so a shell with a TLS proxy set gives 11 failures and 2 errors for that reason alone),ci-release-policy-test.py103 OK,gateway-healthz-smoke-test.py26 OK (3 IPv6 cases skipped: this host has no IPv6 loopback), the same 26 OK with bash 3.2.57 and aunameshim that printsDarwin,check-product-data.mjsok,git diff --checkclean.Disputed finding
The signer gate was never an
assert. It isrequire(), which raisesValueError. The test above proves that under-O, and proves that the test would catch anassert.Branch names stay on the
cursor/<name>-1297line this agent uses.CI
Principles check
sysctlstops with a message. A mislabelled binary stops packaging.check_native_header) serves the packager and the release verifier. One Intel record indocs/MAC.md.x86_64-darwinand does not write its tarball.Gaps: no real Mac has run the
sysctlpath yet. In run 39018375981 (b4316341) thegateway-smoke-macos-inteljob never started: the run was cancelled when37101760was pushed. The next run of that job is the first native run of--print-platformwith the new check.Intel stays C. No grade here is a merge approval.
Team issues to open
Signed Intel artifact and journey. Title:
Signed x86_64-darwin artifact, darwin-amd64 payloads, and Home journey.🤖 Generated with Claude Code
https://claude.ai/code/session_01W6Qn9RX7LDqrwiAfNC2F2K