Repository navigation
chore(OMN-15604): bump omnibase-infra to 0.38.9 (cascade leg recovery) - #831
Conversation
The release Dependency Cascade for omnibase-infra 0.38.9 never opened a PR
here: [tool.uv.sources] pinned omnibase-infra to git rev ead076af (v0.38.4-era,
2026-07-25, `# raw-override-ok: OMN-15145`), which forces uv resolution to
that exact commit regardless of the declared >=0.38.4,<0.39.0 range. Any
automated `uv lock --upgrade-package omnibase-infra` against this override is
a structural no-op — this is the concrete instance of the failure mode
OMN-15604 AC4 flags for git-pinned first-party deps.
Verified before removing the override (not inferred from ancestry alone,
since squash-promoted history can make ancestry checks vacuous — see
OMN-15604's own correction on that point):
- `git merge-base --is-ancestor ead076af5762be7777a42d52e3ed8668f2d0fa0d v0.38.9`
= true — the pinned rev IS a real ancestor of the released tag here.
- `git diff --stat ead076af v0.38.9 -- src/` = 346 files changed — the pin is
materially behind, not equivalent-content-different-label.
- No other [tool.uv.sources] entry (omnimarket, onex-change-control) declares
an exact-pin on omnibase-infra that would override the range; omnimarket's
own internal override of omnibase-infra is not inherited (it is not a uv
workspace member of this project).
What changed:
- Removed the omnibase-infra [tool.uv.sources] git-rev override.
- Regenerated uv.lock (`uv lock`) — omnibase-infra now resolves from PyPI:
`version = "0.38.9"`, `source = { registry = "https://pypi.org/simple" }`.
Verification:
- scripts/check_dep_provenance.py: OK, no forbidden override / stale pin.
- ruff format --check / ruff check: clean.
- mypy src/ --strict: 0 issues, 1076 source files.
- pytest -m unit: 5069 passed, 4 skipped, 6 xfailed, 3 errors — the 3 errors
are the same pre-existing test_multi_repo_crawl_counts.py filesystem-state
failure documented in the OMN-16287 precedent (omnibase_infra#828), a
gitignored local build-staging artifact unrelated to this dependency change.
- pytest -m audit: 25 passed, 3 skipped.
- pre-commit run --all-files: clean (incl. "Forbid first-party git-source
overrides (OMN-13873)").
No integration-marked tests run — Postgres/Kafka unreachable from this host;
dependency-source-only change, no runtime-touching behavior change.
Ticket: OMN-15604
Evidence-Ticket: OMN-15604
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (1)
Included review availability: 0 reviews are currently available. Your included PR review attempts over the past 7 days set your current allowance at 1 review per hour. 📝 WalkthroughWalkthroughThe dependency configuration removes the ChangesDependency source resolution
Estimated code review effort: 1 (Trivial) | ~5 minutes Merge Risk: ⚪ Minimal · up to This is a localized dependency-source update that moves omnibase-infra to version 0.38.9 without runtime behavior changes. No actionable merge-blocking risk remains beyond normal checks and review. 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
OCC autobind did not mint a companion for this PR: no changed-file candidate could be proven RED against the merge base, and emitting a PR-existence probe instead would be non-falsifiable evidence (OMN-15247). Hand-authored evidence is required. |
…enance cascade recovery) Two PASS receipts binding OmniNode-ai/omniintelligence#831 to the OMN-15604 contract, computed with the canonical per-entry hash (omnibase_core.validation.validator_receipt_gate.compute_contract_entry_sha256) against the OMN-15604.yaml committed and pushed in the prior commit (post-yamlfmt, stable bytes, remote-reachable at 32cac9b), and verified to bind via check_receipt_contract_binding before commit: - dod-OmniNode-ai-omniintelligence-pr-831: live `gh pr view 831` readback — PR exists, OPEN, correct branch. - dod-OmniNode-ai-omniintelligence-pr-831-ci: live file-scope assertion (OMN-14425) — PR #831 touches exactly pyproject.toml + uv.lock. Local admissibility validator: `uv run pytest tests/test_evidence_admissibility.py -q` — 76 passed (matches the OMN-16287/#6770 precedent baseline). Self-bind entry (occ-self-bind-pr-<this PR number>) follows in a third commit once the PR number is known, same pattern as #6770/#6862. Ticket: OMN-15604 Evidence-Ticket: OMN-15604
… (#6866) * evidence(OMN-15604): mint contract entries for OmniNode-ai/omniintelligence#831 Local CLI mint fallback (occ-autobind declined on omniintelligence#831 — "no changed-file candidate could be proven RED against the merge base"; pure lockfile/dependency-source change, no RED-then-GREEN diff to key off). Two new dod_evidence entries appended to the existing OMN-15604 contract: - dod-OmniNode-ai-omniintelligence-pr-831 (PR identity) - dod-OmniNode-ai-omniintelligence-pr-831-ci (exact file-scope: pyproject.toml,uv.lock) Committing the contract change alone first so yamlfmt reflows/stabilizes the file before contract_sha256 / contract_entry_sha256 are computed for the receipts (per-entry hash, OMN-13888 — appending an entry does not disturb prior entries' hashes). Ticket: OMN-15604 Evidence-Ticket: OMN-15604 * evidence(OMN-15604): PASS receipts for omniintelligence#831 (dep-provenance cascade recovery) Two PASS receipts binding OmniNode-ai/omniintelligence#831 to the OMN-15604 contract, computed with the canonical per-entry hash (omnibase_core.validation.validator_receipt_gate.compute_contract_entry_sha256) against the OMN-15604.yaml committed and pushed in the prior commit (post-yamlfmt, stable bytes, remote-reachable at 32cac9b), and verified to bind via check_receipt_contract_binding before commit: - dod-OmniNode-ai-omniintelligence-pr-831: live `gh pr view 831` readback — PR exists, OPEN, correct branch. - dod-OmniNode-ai-omniintelligence-pr-831-ci: live file-scope assertion (OMN-14425) — PR #831 touches exactly pyproject.toml + uv.lock. Local admissibility validator: `uv run pytest tests/test_evidence_admissibility.py -q` — 76 passed (matches the OMN-16287/#6770 precedent baseline). Self-bind entry (occ-self-bind-pr-<this PR number>) follows in a third commit once the PR number is known, same pattern as #6770/#6862. Ticket: OMN-15604 Evidence-Ticket: OMN-15604 * evidence(OMN-15604): self-bind contract entry for OCC companion #6866 Ticket: OMN-15604 Evidence-Ticket: OMN-15604 * evidence(OMN-15604): PASS receipt for OCC self-bind #6866 Ticket: OMN-15604 Evidence-Ticket: OMN-15604
…eceipt hash for omniclaude#2012 The pass-1 autobind commit (55c755a) minted a DoD receipt at drift/dod_receipts/OMN-15604/dod-OmniNode-ai-omniclaude-pr-2012/command.yaml but never added the corresponding dod_evidence entry to contracts/OMN-15604.yaml (unlike the omnibase_infra and omniintelligence consumers, which each got both a direct-PR entry and a self-bind entry). That left the receipt's evidence_item_id dangling (no matching contract entry), and its legacy contract_sha256 whole-file hash went stale the moment the concurrent OmniNode-ai/omniintelligence#831 cascade-leg companion (OCC#6866) merged its own entries into dev underneath this branch. - Adds the missing "dod-OmniNode-ai-omniclaude-pr-2012" dod_evidence entry, matching the shape of the omniintelligence direct-consumer entry. - Switches the receipt from the legacy whole-file contract_sha256 to the preferred per-entry contract_entry_sha256 (OMN-13888), computed via omnibase_core.validation.validator_receipt_gate.compute_contract_entry_sha256 against the entry above, so future unrelated appends to this ticket's contract cannot invalidate this receipt again. Re-verified the hash is stable after the yamlfmt pre-commit reflow (folded-scalar rewrap only). - Verified (uv run) that this receipt and all 6 pass-1 supersede receipts pass scripts/validation/check_receipt_hardening.py locally (exit 0), and that none of the pre-existing entries' per-entry hashes were disturbed by the addition.
…6863) * evidence: OCC companion pass 1 for OmniNode-ai/omniclaude#2012 * evidence: OCC companion self-bind for #6863 * fix(OMN-15604): add missing direct-consumer contract entry + repair receipt hash for omniclaude#2012 The pass-1 autobind commit (55c755a) minted a DoD receipt at drift/dod_receipts/OMN-15604/dod-OmniNode-ai-omniclaude-pr-2012/command.yaml but never added the corresponding dod_evidence entry to contracts/OMN-15604.yaml (unlike the omnibase_infra and omniintelligence consumers, which each got both a direct-PR entry and a self-bind entry). That left the receipt's evidence_item_id dangling (no matching contract entry), and its legacy contract_sha256 whole-file hash went stale the moment the concurrent OmniNode-ai/omniintelligence#831 cascade-leg companion (OCC#6866) merged its own entries into dev underneath this branch. - Adds the missing "dod-OmniNode-ai-omniclaude-pr-2012" dod_evidence entry, matching the shape of the omniintelligence direct-consumer entry. - Switches the receipt from the legacy whole-file contract_sha256 to the preferred per-entry contract_entry_sha256 (OMN-13888), computed via omnibase_core.validation.validator_receipt_gate.compute_contract_entry_sha256 against the entry above, so future unrelated appends to this ticket's contract cannot invalidate this receipt again. Re-verified the hash is stable after the yamlfmt pre-commit reflow (folded-scalar rewrap only). - Verified (uv run) that this receipt and all 6 pass-1 supersede receipts pass scripts/validation/check_receipt_hardening.py locally (exit 0), and that none of the pre-existing entries' per-entry hashes were disturbed by the addition. --------- Co-authored-by: node-occ-companion-effect <occ-companion-effect@omninode.ai> Co-authored-by: jonahgabriel <jonah@omninode.ai>
… (#6865) * evidence: OCC companion pass 1 for OmniNode-ai/omniintelligence#831 * evidence: OCC companion self-bind for #6865 --------- Co-authored-by: node-occ-companion-effect <occ-companion-effect@omninode.ai> Co-authored-by: jonahgabriel <jonah@omninode.ai>
OMN-15604: recover the missed omnibase-infra 0.38.9 Dependency Cascade leg
The post-release Dependency Cascade for
omnibase-infra0.38.9 never opened a PR against this repo. Root cause:[tool.uv.sources]pinnedomnibase-infrato git revead076af5762be7777a42d52e3ed8668f2d0fa0d(# raw-override-ok: OMN-15145), which forcesuvresolution to that exact commit regardless of the declared>=0.38.4,<0.39.0range in[project.dependencies]. Any automateduv lock --upgrade-package omnibase-infraagainst this override is a structural no-op — this is the concrete instance of the failure mode OMN-15604 AC4 flags ("cascade no-op is surfaced, not silent") for git-pinned first-party deps, and mirrors the same mechanismomniintelligence#828(OMN-16287) already fixed foromnibase-corein this same file.Verified before removing the override
Per the "content is the correct oracle, ancestry alone is not dispositive under squash-promotion" finding in OMN-15604 itself:
git merge-base --is-ancestor ead076af5762be7777a42d52e3ed8668f2d0fa0d v0.38.9→ true — the pinned rev IS a real ancestor of the released tag here (unlike the omnibase-core case inomniintelligence#828, where ancestry across the dev/main squash boundary wasfalseand content-diff was the only valid oracle).git diff --stat ead076af v0.38.9 -- src/→ 346 files changed — the pin is materially behind, not an equivalent tree under a different label.[tool.uv.sources]entry in this file (omnimarket,onex-change-control) declares an exact-pin onomnibase-infrathat would force resolution elsewhere;omnimarket's own internalomnibase-infraoverride is not inherited here —omnimarketis pulled as a git dependency, not a uv workspace member, so its[tool.uv.sources]is local to its own project resolution only.omnibase-core==exact-pin from any pinned transitive dep constrainsomnibase-infraitself.What changed
omnibase-infra[tool.uv.sources]git-rev override.uv.lock(uv lock) —omnibase-infranow resolves from PyPI:version = "0.38.9",source = { registry = "https://pypi.org/simple" }.Verification
scripts/check_dep_provenance.py --pyproject pyproject.toml:OK: no forbidden first-party git-source override and no stale cross-repo git pin.ruff format --check/ruff check: clean.mypy src/ --strict: 0 issues, 1076 source files.pytest -m unit: 5069 passed, 4 skipped, 6 xfailed (matchesomniintelligence#828's baseline exactly), 3 errors — the 3 errors are the same pre-existingtests/unit/nodes/node_ast_extraction_compute/test_multi_repo_crawl_counts.pyfailure documented inomniintelligence#828's body: it crawls live sibling-repo directories on disk and trips on a gitignored local build-staging artifact, unrelated to any tracked source tree or this dependency change.pytest -m audit: 25 passed, 3 skipped.pre-commit run --all-files: clean, including "Forbid first-party git-source overrides (OMN-13873)".pre-commit run --all-files --hook-stage pre-push: clean (mypy pre-push + pytest-tools full).No integration-marked tests run — Postgres/Kafka are unreachable from this host and this is a dependency-source-only change with no runtime-touching behavior change (same scope precedent as
omniintelligence#828).Precedent
omniintelligence#828(OMN-16287, merged 2026-08-20) is the identical mechanism applied toomnibase-corein this same file, one PR prior. That PR's own sweep table flagged theomnibase-infraoverride as the next one to resolve, tracked under OMN-15145; this PR closes that gap under the cascade-recovery umbrella of OMN-15604.Ticket: OMN-15604
proof_class: deployed (pending merge)
Evidence-Ticket: OMN-15604
Evidence-Source: OCC#6865
Summary by CodeRabbit