feat(validation): accept a checkout as a meta-schema source

[simontaurus]

The blocker for replacing validate.mjs in oold-schema's CI.

The problem

Every route this package has to a meta-schema is a released tag or refs/heads/main:

selector	resolves to
latest, 0.7.0, ...	a vendored release tag
remote	refs/heads/main, and its file list carries no oold-rules.json at all

So a rule added on a branch is invisible to both. The rule.* checks bound to it skip, each reporting that the version never stated it, and the run passes. The pull request that introduces a rule is the one run that cannot enforce it. Pointing oold-schema's CI at this validator without fixing that would be a straight regression against validate.mjs, which reads the working tree.

The change

--meta <path> loads the meta-schemas and the catalogue from a checkout. Both the repository root and its meta/ directory work.

A selector is treated as a path by being a directory, not by a prefix, so --meta ../oold-schema reads the way a path should and no tracked version name can collide - none of them is a directory here.

Nothing loaded this way is checksummed, deliberately. These are working files and expected to change, which is the point. load_tracked stays the only route to a released version, so its checksums keep meaning what they say.

Verified against the real repository

Planting an unreleased rule in an oold-schema checkout and reading both sources:

1.0.0-rc.3  rules= 73  CNF rules=[OOLD-CNF-1120, OOLD-CNF-22d3, OOLD-CNF-d71d]
local       rules= 74  CNF rules=[OOLD-CNF-1120, OOLD-CNF-22d3, OOLD-CNF-d71d, OOLD-CNF-dabd]

The vendored release cannot see OOLD-CNF-dabd; the checkout can. Tests cover both selector forms, that the tracked copy is untouched, and that a directory holding no meta-schemas is rejected by name - the likely mistake being a wrong path.

558 tests pass, make check exits 0.

Next

With this in place, oold-schema's CI can run oold validate --meta <working tree> and see every rule in the branch under test. Parity gating (#125 item 1) should land before the swap, so the equivalence claim is enforced at the moment it becomes load-bearing.

[github-actions]

Release preview

Merging this PR would release v0.18.0 (current: v0.17.1).

Changelog preview (truncated)

## v0.18.0 (2026-08-24)

### Features

- **validation**: Accept a checkout as a meta-schema source
  ([`d2b632f`](https://github.com/OO-LD/oold-python/commit/d2b632f820a933ab242a3f2cfd00bbf2cd0901d1))

_{Preview via python-semantic-release and conventional commits.}

[github-actions]

📊 Benchmark Results

Click to see benchmark comparison

📊 Benchmark Comparison (threshold: 1.3x)
============================================================

✅ Performance Improvements:
  ✅ test_core[v1]: 0.0335s → 0.0241s (-28.2%, ratio: 0.72x)

➖ Unchanged (within threshold):
  ➖ test_simple_dict_document_store: 0.0010s → 0.0011s (+7.1%)
  ➖ test_sqlite_document_store: 0.0011s → 0.0011s (-1.7%)
  ➖ test_local_sparql_store: 0.0237s → 0.0258s (+8.6%)
  ➖ test_oneof_subschema: 0.0385s → 0.0365s (-5.1%)
  ➖ test_enum_docstrings: 0.0503s → 0.0417s (-17.1%)
  ➖ test_subclass_inheritance: 0.0362s → 0.0387s (+7.1%)
  ➖ test_class_hierarchy: 0.0413s → 0.0485s (+17.6%)
  ➖ test_core[v2]: 0.0426s → 0.0329s (-22.7%)
  ➖ test_schema_generation[v1]: 0.0010s → 0.0012s (+18.3%)
  ➖ test_schema_generation[v2]: 0.0017s → 0.0020s (+17.2%)
  ➖ test_simple_json: 0.0004s → 0.0005s (+16.0%)
  ➖ test_complex_graph: 0.0009s → 0.0010s (+12.6%)

============================================================
Summary: 0 regressions, 1 improvements, 12 unchanged
============================================================

✅ No significant performance regressions

Threshold: 1.3x (30% slower triggers a regression warning)

Note: Benchmarks are informational only and won't fail the build.

_{💡 Tip: Download the benchmark-results artifact for detailed JSON data}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

📊 Benchmark Results

Click to see benchmark comparison

📊 Benchmark Comparison (threshold: 1.3x)
============================================================

➖ Unchanged (within threshold):
  ➖ test_simple_dict_document_store: 0.0018s → 0.0018s (+1.1%)
  ➖ test_sqlite_document_store: 0.0020s → 0.0019s (-0.8%)
  ➖ test_local_sparql_store: 0.0442s → 0.0447s (+1.2%)
  ➖ test_oneof_subschema: 0.0661s → 0.0656s (-0.8%)
  ➖ test_enum_docstrings: 0.0553s → 0.0554s (+0.1%)
  ➖ test_subclass_inheritance: 0.0602s → 0.0595s (-1.2%)
  ➖ test_class_hierarchy: 0.0575s → 0.0570s (-0.9%)
  ➖ test_core[v1]: 0.0422s → 0.0546s (+29.3%)
  ➖ test_core[v2]: 0.0486s → 0.0484s (-0.4%)
  ➖ test_schema_generation[v1]: 0.0017s → 0.0018s (+2.1%)
  ➖ test_schema_generation[v2]: 0.0030s → 0.0030s (+1.0%)
  ➖ test_simple_json: 0.0007s → 0.0007s (-0.1%)
  ➖ test_complex_graph: 0.0016s → 0.0016s (-0.5%)

============================================================
Summary: 0 regressions, 0 improvements, 13 unchanged
============================================================

✅ No significant performance regressions

Threshold: 1.3x (30% slower triggers a regression warning)

Note: Benchmarks are informational only and won't fail the build.

_{💡 Tip: Download the benchmark-results artifact for detailed JSON data}