Recurse into nested class_derivations in validate-spec

[amc-corey-cox]

Closes #211. Closes #213. Closes #219.

Summary

Walk nested class_derivations declared on slot_derivations and validate them like top-level CDs, threading the parent CD through so cross-table boundaries can be diagnosed at validate-spec time. Adds three checks plus the recursion shape that enables them.

Checks added

Cross-table mismatch — closes #211

When a nested CD's populated_from differs from its parent's:

• Explicit joins: covers the nested source → no message.
• No explicit join, pick_join_key would synthesize a join at runtime (Implicit cross-table join resolution for nested and regular class derivations #212) → INFO naming the column.
• No explicit join, no implicit join possible → WARNING with the same diagnostic the runtime would raise (no columns are shared or multiple candidate join columns).

The warning surfaces the silent-null hazard from the original issue at validate-spec time, instead of waiting until transform.

Joined-class attribute refs — closes #213

Expression refs like {alias.field} are validated against the joined class's slot list. Previously, the validator dropped {alias.*} patterns from expression-reference checking entirely (a workaround for false positives on join aliases); now those refs are checked against the actual joined class. Covers both:

• explicit joins: aliases (with class_named honored)
• implicit-join targets surfaced via nested CDs' populated_from

This is partial coverage of #213 — runtime-side eval-time validation is separately needed for the bare {scroe} typo case. With #230 + this PR landed, the silent-null mechanism the issue describes is closed at the validate-spec layer.

is_a / mixins resolution — closes #219 (Option C)

is_a (string) and each entry of mixins (list) resolve in this order:

1. Spec-internal class_derivation name pool — matches Transformer._find_class_derivation_by_name
2. Target schema classes via SchemaView — matches LinkML's plain-schema is_a convention
3. Else → error ('is_a: TotallyMadeUpName' does not resolve to ...)

Composes cleanly with #230's structural change: that PR makes the JSON Schema layer accept string references; this PR makes the semantic layer resolve them. They can land in any order; once both are in, the full chain works end-to-end.

Other changes

• ValidationMessage.severity widened to Literal["error", "warning", "info"] to carry the implicit-join INFO.
• New extract_expr_attribute_references() helper — returns {base: {attr, ...}} for {base.attr} patterns, mirroring extract_expr_slot_references().
• New _build_joined_class_map(), _check_cross_table_join(), _collect_class_derivation_pool(), _check_class_inheritance_refs() helpers.
• _validate_class_derivation now takes optional parent_class_deriv, parent_path, and derivation_pool kwargs; recurses into nested CDs.
• _validate_slot_derivation's joined_aliases: set[str] becomes joined_class_slots: dict[str, set[str] | None].

Test plan

• uv run pytest tests/test_validator.py — 86 passed (62 prior + 18 cross-table + 6 inheritance)
• uv run pytest — 828 passed, 4 skipped
• uv run ruff check . + uv run ruff format --check . — clean
• Regression test reproducing the exact Silent null values when nested object populated_from references a different table than outer block (no warning, no fallback join) #211 ARIC pht006467 / pht012814 shape — emits the expected INFO
• 3-level nesting test — cross-table check fires at each parent→child boundary
• Personinfo-to-agent fixture test — Agent.is_a: Entity resolves via pool-1 (the empty Entity: CD)

[Copilot]

Pull request overview

This PR enhances validate-spec semantic validation to traverse nested class_derivations under slot_derivations, enabling earlier detection of cross-table nesting issues, validating {alias.field} expression references against joined classes, and resolving is_a/mixins references against spec-internal derivations and/or the target schema.

Changes:

• Added recursive semantic validation of nested class_derivations, including a new cross-table joinability diagnostic for parent→child populated_from mismatches.
• Added extract_expr_attribute_references() and extended expression validation to check {alias.field} patterns against joined-class slot sets.
• Added is_a/mixins reference resolution via a top-level class-derivation name pool with fallback to target-schema classes; widened ValidationMessage.severity to include "info".

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/linkml_map/validator.py	Implements nested-CD recursion, cross-table join diagnostics, joined-class expression attribute validation, and inheritance reference resolution; adds "info" severity.
tests/test_validator.py	Adds unit tests covering attribute reference extraction, nested cross-table diagnostics, joined-class attribute validation, and is_a/mixins resolution behavior.

[amc-corey-cox]

Scope refinement after handover from a runtime-side session:

Re #213: The PR body says "partial coverage of #213 — runtime-side eval-time validation is separately needed for the bare {scroe} typo case." That undersells what's here. _validate_slot_derivation (validator.py:919-928) already validates bare-name expression refs against source_class_slots — so expr: '{scroe}' is caught at validate-spec time today. With this PR's joined-class attribute check on top, the static side of #213 is fully covered for both bare and dotted forms. The remaining #213 work is the runtime change in eval_utils._eval_name (stop swallowing NameNotDefined → None) plus defense-in-depth for cases where the validator has no schema — both separate from this PR.

Re #217 (new in d4a7ca2): Added a static warning for the column-overlap hazard described in #217 Bug 2. When _check_cross_table_join would synthesize an implicit join, it now also checks whether parent and child share non-join columns. If they do, runtime ambiguity behavior is data-dependent (rows with a join match merge and mark shared columns _AMBIGUOUS; rows without a match silently return the parent's value for the same column). The hazard is fully determined by the schemas, so the validator surfaces it instead of waiting for a sparse row to expose it. This addresses the static half of #217 Bug 2. Bug 1 (misleading {value: None} shape on join miss) and the runtime side of Bug 2 (consistent sentinel insertion regardless of join hit) remain follow-up work.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

[amc-corey-cox]

Correction to the previous comment re #217:

Reverted the column-overlap warning in 481d461. On a second look it was unreachable in practice: pick_join_key only returns a non-None key when there's exactly one non-identifier column in common, so any remaining overlap after subtracting the join key is identifier-only. Filtering identifiers (the only sensible refinement) would have made the check fire on zero realistic inputs.

The genuine #217 Bug 2 hazard — two non-identifier columns in common where one is the join key and the other is a real data column — is already caught: pick_join_key refuses to pick in that scenario, and we fall through to the existing "multiple candidate join columns" WARNING, not the implicit-join INFO branch.

So the practical position is:

• Static half of Sparse implicit-join semantics: nested object on no-match + data-dependent ambiguity enforcement #217 Bug 2: covered by the ambiguous-candidates WARNING that was already in this PR before d4a7ca2. No new code required.
• Bug 1 and the runtime side of Bug 2: still follow-up work.

Net: this PR no longer touches #217 with its own diagnostic; the issue's static hazard is incidentally covered by the cross-table check we already had.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

src/linkml_map/validator.py:768

• _check_cross_table_join() derives parent_source from parent_cd.get("populated_from") only. If the parent CD omits populated_from (identity mapping), the function returns early and skips cross-table mismatch diagnostics, but the runtime uses parent_class_deriv.populated_from or parent_class_deriv.name when resolving nested objects. To avoid missing #211-style warnings/infos, compute the effective parent source the same way (fallback to the parent CD’s name).

    nested_source = nested_cd.get("populated_from")
    parent_source = parent_cd.get("populated_from")
    if not nested_source or not parent_source or nested_source == parent_source:
        return