fix(ci): allow packaging>=26 so twine check accepts Metadata-Version 2.5 - #411
Conversation
![Mikhail [azalio] Petrov](https://github.githubassets.com/images/gravatars/gravatar-user-420.png){kind=avatar}
The packaging>=24.2,<26 cap (#195) downgrades packaging to 25.0 in the build job, and 25.0 does not recognize Metadata-Version 2.5 emitted by current setuptools. twine check then fails with InvalidDistribution: '2.5' is not a valid metadata version. Verified locally: same wheel fails twine check with packaging<26 and passes with packaging>=26.
📝 WalkthroughWalkthroughThe workflows now install ChangesPackaging validation
Specula applicability assessment
Estimated code review effort: 3 (Moderate) | ~20 minutes Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/spikes/specula-applicability.md`:
- Around line 368-380: Extend the process-interruption scenario in the
experiment plan through restart and reconciliation: restart the interrupted
runtime, reconcile the persisted partial state, and verify resumption or
terminal handling as required. Repeat reconciliation to demonstrate idempotence,
asserting that group visibility, main HEAD, worktrees, and rendered runtime
artifacts remain correct after both passes.
- Around line 392-404: Update the bounded exploration and resource-bound
requirements to define finite constants for every state dimension, including
event counts, retries, failure choices, Git states, and coordinator ownership.
Require TLC to exhaust that explicitly finite state space and report complete
exploration separately from successful execution or resource-limit termination.
- Around line 165-190: Define the versioned Invariant_Lineage_v1 schema before
implementing validation, including source-specific reference shapes for code,
test, issue, commit, document, user_decision, and external references. Add an
explicit verification-method vocabulary and a rationale field required when
fault_model is none, while preserving the stated lineage and validation
requirements.
- Around line 7-8: Update the spike’s MAP metadata to record the exact commit
SHA instead of main, and revise all source and test lineage references to
include both a revision and a stable symbol or content hash rather than relying
on file/line ranges. Apply this consistently to the captured revision and the
referenced sections.
- Around line 162-183: Update the validator described in the Requirements Index
contract to require non-empty lineage for every entry whose kind is invariant,
regardless of its identifier prefix. Keep INV-* naming validation as a separate
rule only if that convention remains mandatory, so identifier format does not
control lineage enforcement.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 709b3013-b544-42d5-8e2e-7c463224954d
📒 Files selected for processing (2)
CHANGELOG.mddocs/spikes/specula-applicability.md
| **Captured:** 2026-07-29 against MAP Framework `main` and Specula revision | ||
| [`d45b873`](https://github.com/specula-org/Specula/tree/d45b873bc19cf51bc26fd845fce072d9b99b5cd0). |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift
Pin the MAP revision and make lineage references stable.
The spike records MAP as main, and the validator checks current file/line or test references. A line range can remain valid while pointing to different code. The required stale-lineage cases can then pass incorrectly.
Record the exact MAP commit SHA. Require each source and test reference to include a revision plus a stable symbol or content hash.
Also applies to: 182-190, 306-316
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/spikes/specula-applicability.md` around lines 7 - 8, Update the spike’s
MAP metadata to record the exact commit SHA instead of main, and revise all
source and test lineage references to include both a revision and a stable
symbol or content hash rather than relying on file/line ranges. Apply this
consistently to the captured revision and the referenced sections.
| Adopt a low-cost `Invariant_Lineage_v1` contract inside each Requirements Index | ||
| `kind: invariant` entry: | ||
|
|
||
| ```yaml | ||
| - id: INV-1 | ||
| kind: invariant | ||
| lineage: | ||
| sources: | ||
| - kind: code | ||
| ref: src/module.py:120-138 | ||
| scenario: concurrent wave abort after a partial squash merge | ||
| fault_model: merge conflict, process interruption, or failed verification | ||
| verification: | ||
| method: test | ||
| ref: tests/test_wave.py::test_abort_restores_base | ||
| ``` | ||
|
|
||
| The source-kind vocabulary should cover `code`, `test`, `issue`, `commit`, | ||
| `document`, and `user_decision`. A deterministic validator should: | ||
|
|
||
| 1. require non-empty lineage for every indexed `INV-*`; | ||
| 2. validate repo `file:line` ranges, test references, commit existence, and |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
Use one authoritative invariant predicate.
The contract targets Requirements Index entries with kind: invariant, but the validator rule targets only INV-* identifiers. If an invariant uses another identifier prefix, it can bypass lineage validation.
Require lineage when kind == invariant. Validate the INV-* naming convention separately if that prefix is mandatory.
Proposed clarification
- require non-empty lineage for every indexed `INV-*`;
+ require non-empty lineage for every indexed entry where `kind: invariant`;
+ validate the `INV-*` identifier format separately, if required;📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| Adopt a low-cost `Invariant_Lineage_v1` contract inside each Requirements Index | |
| `kind: invariant` entry: | |
| ```yaml | |
| - id: INV-1 | |
| kind: invariant | |
| lineage: | |
| sources: | |
| - kind: code | |
| ref: src/module.py:120-138 | |
| scenario: concurrent wave abort after a partial squash merge | |
| fault_model: merge conflict, process interruption, or failed verification | |
| verification: | |
| method: test | |
| ref: tests/test_wave.py::test_abort_restores_base | |
| ``` | |
| The source-kind vocabulary should cover `code`, `test`, `issue`, `commit`, | |
| `document`, and `user_decision`. A deterministic validator should: | |
| 1. require non-empty lineage for every indexed `INV-*`; | |
| 2. validate repo `file:line` ranges, test references, commit existence, and | |
| Adopt a low-cost `Invariant_Lineage_v1` contract inside each Requirements Index | |
| `kind: invariant` entry: | |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/spikes/specula-applicability.md` around lines 162 - 183, Update the
validator described in the Requirements Index contract to require non-empty
lineage for every entry whose kind is invariant, regardless of its identifier
prefix. Keep INV-* naming validation as a separate rule only if that convention
remains mandatory, so identifier format does not control lineage enforcement.
| ```yaml | ||
| - id: INV-1 | ||
| kind: invariant | ||
| lineage: | ||
| sources: | ||
| - kind: code | ||
| ref: src/module.py:120-138 | ||
| scenario: concurrent wave abort after a partial squash merge | ||
| fault_model: merge conflict, process interruption, or failed verification | ||
| verification: | ||
| method: test | ||
| ref: tests/test_wave.py::test_abort_restores_base | ||
| ``` | ||
|
|
||
| The source-kind vocabulary should cover `code`, `test`, `issue`, `commit`, | ||
| `document`, and `user_decision`. A deterministic validator should: | ||
|
|
||
| 1. require non-empty lineage for every indexed `INV-*`; | ||
| 2. validate repo `file:line` ranges, test references, commit existence, and | ||
| typed external/user-decision references without dereferencing secrets; | ||
| 3. require a non-empty scenario, an explicit fault model (`none` allowed only | ||
| with rationale), and a named verification method; | ||
| 4. preserve the lineage through parsing, `coverage_map`, verification summary, | ||
| and review bundle; and | ||
| 5. report `missing_lineage` separately from `missing_evidence`—declaring an | ||
| owner or repeating `[INV-1]` must not self-certify provenance. |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift
Define the Invariant_Lineage_v1 schema before implementation.
The example defines only a code source and verification.method: test. The rules also require issue, commit, document, user-decision, and external references. Their shapes and allowed values are unspecified.
The rule for fault_model: none requires a rationale, but the example has no rationale field. Add a versioned schema, source-specific reference types, a verification-method vocabulary, and an explicit rationale field.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/spikes/specula-applicability.md` around lines 165 - 190, Define the
versioned Invariant_Lineage_v1 schema before implementing validation, including
source-specific reference shapes for code, test, issue, commit, document,
user_decision, and external references. Add an explicit verification-method
vocabulary and a rationale field required when fault_model is none, while
preserving the stated lineage and validation requirements.
| Bound the first model to two subtasks and one wave, then cover: | ||
|
|
||
| - full success; | ||
| - conflict on the second squash merge; | ||
| - post-wave verification failure after both squash merges; | ||
| - external HEAD movement before merge; | ||
| - process interruption after `begin_wave_group` and before all terminal events; | ||
| - rollback that returns without moving HEAD. | ||
|
|
||
| Use actual temporary Git repositories and the rendered runtime for trace | ||
| collection. Mock-only lifecycle traces are insufficient because the experiment | ||
| is specifically testing conformance between the model, Git effects, and | ||
| sidecar state. |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major | 🏗️ Heavy lift
Extend the interruption scenario through reconciliation.
The scenario stops after process interruption. This tests persisted partial state, but it does not test restart, reconciliation, resume, or idempotence. Those behaviors are required by the crash-visible lifecycle and Phase 2 plan.
Add a restart and reconciliation step. Repeat reconciliation and assert that group visibility, main HEAD, worktrees, and runtime artifacts remain correct.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/spikes/specula-applicability.md` around lines 368 - 380, Extend the
process-interruption scenario in the experiment plan through restart and
reconciliation: restart the interrupted runtime, reconcile the persisted partial
state, and verify resumption or terminal handling as required. Repeat
reconciliation to demonstrate idempotence, asserting that group visibility, main
HEAD, worktrees, and rendered runtime artifacts remain correct after both
passes.
| 4. **Bounded exploration:** TLC exhausts the two-subtask bounded state space; | ||
| simulation-only results do not satisfy the experiment. | ||
| 5. **Real confirmation:** every reported counterexample is reproduced against | ||
| the rendered Python runtime in an isolated real Git repository. An | ||
| unreproduced model violation is recorded as a model/harness gap, not a MAP | ||
| bug. | ||
| 6. **Incremental value:** compare findings against the existing wave tests. A | ||
| promotion decision requires at least one previously uncovered reachable | ||
| defect or a concrete reachable transition gap that becomes a regression | ||
| test; restating existing tests is a negative result. | ||
| 7. **Resource bound:** stop at four wall-clock hours, 32 GB TLC memory, and | ||
| $75 equivalent model cost. Record actual time, token cost, peak memory, | ||
| model/effort, Specula SHA, MAP SHA, and explored-state count. |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major | 🏗️ Heavy lift
Define a finite TLC state space.
Two subtasks and one wave do not bound event counts, retries, failure choices, Git states, or coordinator ownership. TLC cannot exhaust the model unless every such domain is finite.
Define finite constants for each state dimension. Require the run to report complete state-space exploration, not only successful execution within the resource limit.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/spikes/specula-applicability.md` around lines 392 - 404, Update the
bounded exploration and resource-bound requirements to define finite constants
for every state dimension, including event counts, retries, failure choices, Git
states, and coordinator ownership. Require TLC to exhaust that explicitly finite
state space and report complete exploration separately from successful execution
or resource-limit termination.
…notes excerpt (#412) Two release-pipeline defects found while releasing 3.25.0: 1. pypa/gh-action-pypi-publish@v1.13.0 bundles an older twine that rejects Metadata-Version 2.5 wheels (InvalidDistribution). The repo's own twine-check step was already fixed in #411, but the action runs its own check internally. v1.14.2 ships twine v7 with core-metadata 2.5 support (upstream #416). Applied in release.yml and test-pypi.yml. 2. The GitHub Release changelog excerpt was always empty: the two-address awk range /^## \[VERSION\]/,/^## \[/ collapses to a single line because the version heading matches both patterns, so every release body fell back to 'See CHANGELOG.md for details.' (verified on v3.24.1). Replaced with a flag state machine in release.yml, the release-checklist issue template, and the map-release skill (.jinja source + rendered trees). Co-authored-by: Mikhail [azalio] Petrov <azalio@yandex-team.ru>
1 participant
Summary
Latest CI run on main (31465691220) fails in the
buildjob:Root cause
The
packaging>=24.2,<26cap introduced in #195 forces pip to downgrade the runner's preinstalledpackaging26.3 to 25.0. Current setuptools emits Metadata-Version 2.5 wheels, whichpackaging25.0 does not recognize, sotwine checkaborts.Fix
Bump the pin to
packaging>=26inci.yml,release.yml,test-pypi.yml+ CHANGELOG entry.Verification (local, same wheel)
uv run --with 'twine>=6.2.0' --with 'packaging>=24.2,<26' python -m twine check dist/*→ FAILS (repro)uv run --with 'twine>=6.2.0' --with 'packaging>=26' python -m twine check dist/*→ PASSEDUnblocks the release:
release.ymlhas the identical step and would fail at publish time.🤖 Generated with Claude Code
Summary by CodeRabbit
Bug Fixes
Documentation