Skip to content

Consolidate unmerged branches: F-34 Qiskit worker, IDE API 1.5.0; fix CI - #25

Merged
quantum-osiris merged 8 commits into
mainfrom
chore/consolidate-branches-20261117
Nov 17, 2026
Merged

quantum-osiris merged 8 commits into
mainfrom
chore/consolidate-branches-20261117

Conversation

@quantum-osiris

@quantum-osiris quantum-osiris commented Nov 17, 2026 •

Copy link
Copy Markdown
Owner

Summary

This brings the last work that existed only on local branches onto main, so those branches can be deleted without losing anything. It also fixes the two CI failures on main.

Change Source branch Commits
F-34: Qiskit/Aer runs on one long-lived worker thread with a bounded backlog. A full backlog returns 503 SIMULATOR_BUSY instead of waiting without limit. This replaces the unbounded _QISKIT_THREAD executor. feat/agent-council 36f7d838, merged in f9c36e38
IDE API 1.5.0: adds POST /api/v1/files/rename and the GHZ correctness fixtures claude/notebook 66e0643f
CI: installs cryptography. osiris_cli/edge_ipc_client.py (a74cf58f) imports it, so tests/test_edge_bridge_ipc.py could not be collected. Every test job on main has failed since the 04:28Z push. this branch 38c6f6bf
CI: skips test_ide_api_accepts_x_osiris_token_x_session_token_and_bearer_auth when osiris_governance is not installed. This failure was already on main, hidden behind the collection error, and appeared once the suite could collect. The guard is inside this one test, so the file's other two tests still run on CI. It matches the importorskip pattern used in tests/test_ide_api.py. this branch 0a2186f4

The net change against main is 12 files. Every line this removes is replaced by new code: the API version bump, and the executor swapped for the worker. No feature on main is dropped.

Branches already on main (no merge needed)

  • sprint1.1/operator-key-authority: git cherry shows its two commits, 92ba55ba and 61144e1e, are already on main as 92bf5945 and 6164b231. The branch's three files are identical to main's.

Verification

  • Tests: pytest tests/ phase2/ on this branch gave 2,259 passed, 26 skipped, 0 failed. This ran locally on Python 3.14, with no IBM credentials in the environment.
  • Lint: the CI step passes locally.
  • Running server: a smoke test ran the merged IDE API from this branch on a separate local port, with receipts and token written to a scratch directory and no IBM credentials:
    • /meta reports 1.5.0.
    • dnalang/compile and slice/run work on the GHZ fixture, on local Aer.
    • With 8 concurrent compile/run requests from separate threads, the server stayed up. That was the F-34 crash case. 5 requests returned 200 and 3 returned 503 SIMULATOR_BUSY from the bounded backlog.
    • files/rename answers with a structured error when no user session is configured.
    • The server shuts down cleanly on SIGINT.
  • test_contracts_and_s26_cartographer.py: 3 passed locally. With osiris_governance blocked, as on CI, 2 passed and 1 skipped.
  • Scan: the diff was checked for credentials and personal data, and none was found.

Not changed

  • The claims register.
  • The locked Phase 3 artifacts.

Follow-up

  • cryptography is still not declared in pyproject.toml. A clean install of the package fails when anything imports osiris_cli.edge_ipc_client.

🤖 Generated with Claude Code

quantum-osiris and others added 8 commits November 16, 2026 23:18
qiskit 2.6.2 / qiskit-aer 0.18.2 / Python 3.14: a second simulation started from a new
thread segfaults the process in qiskit._accelerate (ThreadingHTTPServer opens one thread
per request). compile, slice/run, organism/run, files/run and dnalang/info, plus the
compile step of assist, now run on osiris_cli.qiskit_worker: one thread, at most
OSIRIS_QISKIT_MAX_PENDING (default 4) admitted calls, 503 SIMULATOR_BUSY beyond that,
admitted runs never cancelled (their receipts are written), shutdown waits for them.

tests/test_qiskit_worker.py: unit tests plus live runs through the real server and Aer:
sequential, concurrent with per-request result/receipt association, overload, failure
then success, client timeout then next request, subprocess restart with one valid chain.
The restart test fails on fe35c91 (service drops the connection) and passes here.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- POST /api/v1/files/rename: revision-guarded path change (same owner policy and trigger as
  save; id, project and content unchanged, so receipts still resolve). Stale base revision is
  CONFLICT, a taken path is EDITOR_PATH_EXISTS, another account gets 404. No migration change.
- tests/fixtures/aura_ghz_proposal_2026-11-16.json: the GHZ change AURA (gemma3:4b) proposed in
  the notebook acceptance run, recovered verbatim from the stored revision 3, with its receipt.
  It compiles and runs and is not GHZ (H on all three qubits -> uniform). tests/test_ghz_fixtures.py
  checks an ideal GHZ fixture against the {000,111} support with an explicit 5-sigma binomial
  bound and checks that the recorded proposal fails it. Fixture-specific; not a product check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replaces the unbounded _QISKIT_THREAD executor that came in with 618aa0a. notebook_run stays on
the worker; test_qiskit_operations_share_one_worker_thread now uses ide_api.qiskit_worker().

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brings the last two commits on feat/agent-council that were not on main: the long-lived Qiskit/Aer
worker thread with a bounded backlog, and its merge into the agent-council line, which replaced the
unbounded _QISKIT_THREAD executor from 618aa0a.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… fixtures (66e0643)

The last commit on claude/notebook that was not on main. Its F-34 commit (36f7d83) arrived with
feat/agent-council in the previous merge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
osiris_cli/edge_ipc_client.py (a74cf58) imports cryptography at module level, so
tests/test_edge_bridge_ipc.py fails to collect and every test job on main has failed since the
04:28Z push. cryptography is not declared in pyproject.toml; this only makes CI install it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ide_api.default_backend imports github_gateway, which needs osiris_governance. CI does not
install it, so this test failed on every Python version once the cryptography fix let the suite
collect. The guard is inside this one test, so the file's other two tests still run on CI.
This matches tests/test_ide_api.py and the other modules that importorskip the governance package.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@quantum-osiris
quantum-osiris merged commit 3cf2cb4 into main Nov 17, 2026
10 checks passed
@quantum-osiris
quantum-osiris deleted the chore/consolidate-branches-20261117 branch November 17, 2026 08:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant