feat!: remove the no_map node mode so subjects and objects always resolve

[SkyeAv]

Removes no_map, the mode that let a statement subject or object skip fullmap entity resolution and emit its own CURIE verbatim. It broke emitted output: with no lookup there is no role to derive, so every verbatim node was pinned to biolink:NamedThing and each edge derived from it flattened to biolink:Association. The resolution path goes back to how it worked in 20.0.0, and configs that still set the key now fail with a message that says why.

Removal

• Config surface: NodeEncoding.no_map is gone from src/tablassert/models.py, along with the no_map_rejects_explicit_taxon, no_map_rejects_resolution_knobs, and no_map_literal_must_be_curie validators and the CURIE_PATTERN/is_curie helpers that only served them.
• Build path: lib.curie_audit and the verbatim branch of Tcode._node_ops are deleted, so subject and object go back through the shared resolve_batch (src/tablassert/lib.py:1417). The op's matching PHASE_OF and runcache.OP_COST entries go with it, which tests/test_run_cache.py:352 requires.
• Error codes: no-map-dead-config, no-map-bad-curie, and no-map-curie-violation are dropped from TablassertErrorCodes in src/tablassert/errors.py.
• Qualifiers: Qualifier.resolved is vocabulary is None again, so CURIE-ranged qualifiers always resolve while enum-ranged qualifiers still pass their vocabulary tokens through verbatim. The mode rode this one shared field, so its qualifier half is removed with it.

Migration notice

• DEPRECATED_KEYS: no_map is registered in src/tablassert/models.py so the before-hook names the removal and the migration before extra="forbid" rejects the key. Warn, then fail: popping the key instead would build the broken output the removal exists to prevent.
• Tables that already carry CURIEs: nothing else to edit, because the fullmap indexes CURIEs and their equivalent identifiers as terms, so a CURIE cell resolves back to its own CURIE (docs/cli.md:118, docs/api/fullmap.md:245). Resolution rules still apply: taxon, prioritize/avoid, exclude_prefixes/exclude_regex, and a blank cell still drops the row.

Docs

• docs/configuration/table.md loses the no_map row and the No-map mode section. tests/test_docs_source_of_truth.py::test_config_model_fields_match_live_models binds docs rows to live model fields in both directions, which is why code and docs land in one commit rather than two layers.
• CHANGELOG.md is untouched: the release commit writes it, so 21.0.0's record stays as shipped and the breaking note belongs in the next release entry. Its 21.0.0 entry still names no_map and curie_audit, which is shipped history, not a missed edit.

Testing

• uv run pytest -q --no-cov -> 1218 passed, 1 skipped (the skip is the opt-in corpus sweep at tests/test_realworld_configs.py:236, which needs TABLASSERT_CONFIG_CORPUS)
• uv run ruff check . -> All checks passed!; uv run ruff format --check . -> 70 files already formatted
• uv run pyright -> 0 errors, 0 warnings, 0 informations
• cargo test --manifest-path rust/Cargo.toml -> 127 passed, 11 passed; 1 ignored, 17 passed, 1 passed; cargo clippy --all-targets -- -D warnings -> exit 0
• uv run mkdocs build --strict -> exit 0
• uv run tablassert validate <config with no_map: true> --schema table -> prints the removal note, then statement.subject.no_map: Extra inputs are not permitted [extra_forbidden], exit 1; the same run with no_map: false fails identically, since the hook keys on presence
• Equivalence check: Tcode._node_ops, PHASE_OF, class NodeEncoding..class Qualifier, and Qualifier.resolved are byte-identical to git show 32f4478~1:src/tablassert/{lib,models}.py, so the restored path is the 20.0.0 path.

Risks

• Breaking: any table config that still sets no_map stops validating, at either value: the hook warns on key presence and extra="forbid" rejects the key regardless of what it is set to, so a no_map: false config breaks too. A graph built with the mode under 21.0.0 emits different node categories and edge types once rebuilt. Deferred: no replacement "verbatim identifier" surface; the supported path is fullmap resolution.

Questions for the reviewer

• Qualifier scope. no_map was a single field shared by subject, object, and qualifier, and the verbatim branch is the same code for all three, so this also drops the CURIE-ranged qualifier opt-out. If that knob has a real use case, it wants its own field and its own finalize op rather than the biolink:NamedThing category synthesis it inherited.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 87b00831-cb74-4d50-ac4c-3576d1b9d695

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}