Repository navigation
feat: add no_map mode so subjects and objects use their own CURIEs verbatim - #219
Merged
Merged
Conversation
…lidation Add no_map: bool = False to NodeEncoding so statement subjects and objects whose data already carries CURIEs can skip fullmap entity resolution and use the encoded value verbatim as the node id. - reject no_map combined with resolution-only knobs (prioritize, avoid, exclude_prefixes, exclude_regex, explicit non-default taxon) as dead config - validate no_map literals are CURIE-shaped (after prefix/suffix), so a bad literal fails at config time instead of silently emptying the build - Qualifier.resolved becomes the single routing predicate: enum-ranged qualifiers stay verbatim, CURIE-ranged ones resolve unless no_map opts out; no_map on an enum-ranged qualifier is rejected as redundant
…ction pipeline _node_ops now splits node columns into resolved (resolve_batch specs), verbatim (no_map subject/object and no_map CURIE-ranged qualifiers), and literal (enum-ranged qualifiers) groups. - new lib.curie_audit op: drops blank/null-like rows (a missing CURIE cannot form a node), raises no-map-curie-violation on non-CURIE cells, and synthesizes <col>_category = biolink:NamedThing so edge_category degrades to biolink:Association like the documented no-role fallback - resolve_batch receives zero specs for an all-verbatim section, so the fullmap redb is never opened; mixed mapped/verbatim statements resolve the mapped side as before - fullmap_audit naturally skips verbatim columns (it iterates the resolved node_columns list) - register curie_audit in PHASE_OF (encode) and the runcache op-cost table
…IE nodes Add offline build_pipeline smokes for the no_map path: - verbatim CUSTOM: CURIEs absent from the redb appear in nodes and edges, proving no lookup happened; blank-subject row dropped, never emitted - emitted nodes carry the synthesized biolink:NamedThing category and edges reference the emitted node ids - negative smoke: a non-CURIE cell fails the build with no-map-curie-violation Also emit the <col>_pre_resolution snapshot for no_map columns: graph compile discovers node columns through that snapshot, so without it a verbatim subject/object never split out of the edges into nodes and the RIG audit rejected the empty category lists.
Add the no_map row to the NodeEncoding field table (keeps the docs source-of-truth binding green) and a No-map mode subsection covering the verbatim pass-through, blank-drop/non-CURIE-fail behavior, prefix/suffix CURIE formation, and the rejected resolution-filter combinations.
…tinct values only The field description and docs table claimed non-CURIE cells are dropped; the shipped curie_audit raises no-map-curie-violation on them (mirroring qualifier_vocabulary_audit), so both now state: blanks drop, non-CURIE cells fail the build. Also judge CURIE shape on distinct non-blank values so the per-element check does not scale with row count.
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
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 |
SkyeAv
changed the title
SkyeAv
added a commit
that referenced
this pull request
Oct 15, 2026
Cut 21.0.0 and bump the package version in pyproject.toml, CITATION.cff, and uv.lock. Major: two breaking changes ship since 20.0.0. Derived-path builds emit section source record URLs on the upstream repository supporting sources entry (infores:pubmed-central for PMC sections) instead of the primary graph-infores entry, and the inlined supporting Study is named after the table config file plus its 1-based section index (#220); validate-infores is folded into validate-kgx as the knowledge-source identity half, and the standalone validate-infores command is gone (#218). Also adds no_map on NodeEncoding to pass subject/object CURIEs through verbatim (#219) and normalizes documentation prose to plain ASCII (#217). Changelog: - Versioned the Unreleased section as 21.0.0 - 2026-10-15. - Added the missing #218 (breaking, with migration note), #219, and #217 entries so no merged PR ships unmentioned. Docs: none needed here; the release commit only versions existing prose. Testing: - make check -> ruff clean, pyright 0 errors, pytest 1239 passed + 1 skipped, cargo test ok, cargo clippy --all-targets -- -D warnings clean - uv run mkdocs build --strict -> pass
SkyeAv
added a commit
that referenced
this pull request
Oct 16, 2026
…olve (#221) * feat!: remove the no_map node mode so subjects and objects always resolve `no_map: true` on a statement subject or object (#219, released in 21.0.0) skipped fullmap entity resolution and emitted the cell value as the node id verbatim. It breaks the output: with no lookup there is no role to derive, so the build pinned every verbatim node's category to `biolink:NamedThing` and each edge derived from it degraded to `biolink:Association`, losing the typed predicate categories KGX consumers expect. The mode is removed rather than patched, restoring the 20.0.0 resolution path. - `NodeEncoding.no_map` is gone, along with the `no-map-dead-config` and `no-map-bad-curie` config validators, the `no-map-curie-violation` code, and the `CURIE_PATTERN`/`is_curie` helpers that only served them. - `lib.curie_audit` and its `Tcode._node_ops` verbatim branch are gone, so subject and object columns go back through the shared `resolve_batch`, plus the matching `PHASE_OF` and `runcache.OP_COST` entries. - CURIE-ranged qualifiers always resolve again: `Qualifier.resolved` is `vocabulary is None`, so enum-ranged qualifiers still pass their vocabulary tokens through verbatim and nothing else does. - The `no_map` row and the "No-map mode" section are removed from `docs/configuration/table.md`, and the tests written for the mode are gone. Migration: drop `no_map: true` from subject and object encodings. Terms resolve as they did before 21.0.0; tables that already carry CURIEs also resolve, since the fullmap indexes CURIEs and their equivalent identifiers as terms. * feat: explain the removed no_map key to configs that still set it With `no_map` deleted from `NodeEncoding`, a table config written against 21.0.0 failed with pydantic's bare "Extra inputs are not permitted", which names no history and offers no migration step for an unattended pipeline run. `models.DEPRECATED_KEYS` is the repo's scaffold for exactly this: the before-hook warns, and `extra="forbid"` still rejects the key, so the config never builds the output the removal exists to prevent. - `no_map` is registered with a note that says the mode is gone, why (verbatim nodes pinned to `biolink:NamedThing` flattened edge typing), and what old configs should expect now: a CURIE cell resolves back to its own CURIE because the fullmap indexes CURIEs as terms, with the usual `taxon`, `prioritize`/`avoid`, and `exclude_*` rules still applying. - The key is rejected on presence, not truthiness, so `tests/test_models.py` pins the warn-then-reject path for both `no_map: false` and `no_map: true`. - The two existing deprecation tests' "registry is EMPTY" wording is corrected, since the registry now names one removed key.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a
no_map: trueopt-in onNodeEncodingthat skips fullmap entity resolution forstatement subjects and objects, so users whose data already carries its own CURIEs pass
those values through as node ids as-is. Enum-ranged qualifiers already bypass the fullmap
through the
resolvedproperty; this rolls the same pass-through up to the node level asan explicit config choice.
Config surface
no_map: bool = FalseonNodeEncoding(src/tablassert/models.py): default keepsevery existing config resolving exactly as before.
no_mapwith any resolution-only knob (prioritize,avoid,exclude_prefixes,exclude_regex, explicit non-defaulttaxon) raisesno-map-dead-configat config time, matching thenullable-on-literal posture.method: valueliteral must be CURIE-shaped afterprefix/suffix(no-map-bad-curie), so a bad literal fails at config time instead ofsilently emptying the build.
method: columncells are audited per row at build time.Qualifier.resolvedstays the single routing predicate(
vocabulary is None and not no_map);no_mapon an enum-ranged qualifier is rejectedas redundant.
Pipeline
lib._node_opssplits node columns into resolved (resolve_batchspecs), verbatim(no_map), and literal (enum-ranged) groups. An all-verbatim section sends zero specs, so
the fullmap redb is never opened; mixed statements resolve the mapped side as before.
lib.curie_auditop (registered inPHASE_OFand the runcache cost table):drops blank/null-like rows (a missing CURIE cannot form an edge), raises
no-map-curie-violationon non-blank non-CURIE cells (mirroringqualifier_vocabulary_audit), and synthesizes<col>_category = biolink:NamedThingsoedge_categorydegrades tobiolink:Associationlike the documented no-role fallback.<col>_pre_resolutionsnapshot, which_collect_subframesuses to discover node columns; without it verbatim nodes neversplit out of the edges and the RIG audit rejects the empty category lists.
Design
non-CURIE cells fail the build. Free text must not ship as a KGX identifier.
biolink:NamedThing, so their edges derivebiolink:Associationunlesscategory_overridepins a class. No name/taxon enrichmentexists without a lookup.
disable mapping for a whole build.
Docs
docs/configuration/table.md:no_maprow in the NodeEncoding table (keeps the docssource-of-truth binding green) plus a No-map mode subsection with a YAML example.
Testing
make check-> exit 0: ruff clean,pyright0 errors,pytest1235 passed + 1 skipped,cargo testok,cargo clippy -D warningsok.uv run pytest tests/test_e2e_smoke.py -q -n 0 --no-cov-> 10 passed, including abuild_pipelinerun whoseCUSTOM:CURIEs are absent from the redb (proving verbatimpass-through) and a negative smoke asserting
no-map-curie-violationon free text.build_pipelineagainst a nonexistent redb path: build OK, nodes emitted withverbatim ids and
biolink:NamedThingcategories, blank row dropped.Questions for the reviewer
biolink:NamedThingkeeps nodes valid KGX but loses roleprecision in the RIG summaries; is the generic category the right default, or should
no_map accept an optional explicit category?