[gcko]

PR checklist

• Ensure you have added or ran the appropriate tests for your PR.
• DCO signed

What type of PR is this?

fix

What this PR does / why we need it:

sqlglot 30.22.0 removed infer_schema from the optimizer (tobymao/sqlglot#8552). Before that, when no schema was given, qualify attributed an unqualified column to the one source in the scope with no known columns. Now it raises OptimizeError: Column 'x' could not be resolved. PR #1573 capped sqlglot<30.22 as a stopgap.

This PR adapts cll() and lifts the cap:

• recce/util/cll.py: new _qualify() helper. A non-empty schema goes through qualify unchanged. With an empty schema it qualifies with validate_qualify_columns=False, re-applies the removed attribution (_attribute_to_schemaless_source), then calls validate_qualify_columns. This is exact parity with sqlglot <30.22, which also inferred only when the schema was empty.
• Resolver moved from qualify_columns to its own module in late 2025, so the import falls back for the >=27.29.0 floor.
• _CACHE_SCHEMA_VERSION 3 → 4, so unknown results cached by users on 30.22–30.23 are dropped. Cache eviction is access-based, so hot entries would otherwise stay stale.
• pyproject.toml: sqlglot>=27.29.0,<31.0; uv.lock resolves 30.24.0.

Which issue(s) this PR fixes:

Closes DRC-4901

Special notes for your reviewer:

• Where this showed up in production: through DbtAdapter, cll() only gets an empty schema for a model that is in catalog.json but has no resolvable dbt parents, for example one that hard-codes from raw_events. The new adapter test test_model_reading_hardcoded_relation covers this; it reports unknown without the fix and renamed with it. Without a catalog, get_cll returns before calling cll(), so that path is unaffected.
• Not in this PR: when a parent is missing from the catalog, the adapter writes schema[parent] = {}, which raises SchemaError on every sqlglot version. That's a separate, older gap, tracked in DRC-4918.
• Why not a schema stub: sqlglot rejects {} (SchemaError), and {"*": ...} still fails to resolve.
• Version matrix: test_cll.py, test_dbt_cll.py and test_breaking.py pass on sqlglot 27.29.0, 30.21.0 and 30.24.0. The full suite on 30.24.0 gave 2040 passed, 5 skipped.
• recce/util/change_classifier.py also calls qualify, but only with a non-empty schema, so it never relied on the removed inference. Its tests pass unchanged on 30.24.
• The two guard tests (..._ambiguous_without_schema, ..._with_partial_schema) pass on main too. They pin parity so the attribution can't spread to ambiguous or partial-schema cases.

Does this PR introduce a user-facing change?:

Column-level lineage works again with sqlglot 30.22 and later. Models that select from hard-coded relations (no dbt parents) trace their columns again instead of showing them as unknown.

🤖 Generated with Claude Code

[kentwelcome]

LGTM

[codecov]

Codecov Report

❌ Patch coverage is 96.29630% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
recce/util/cll.py	93.54%	2 Missing ⚠️

Files with missing lines	Coverage Δ
tests/adapter/dbt_adapter/test_dbt_cll.py	99.36% <100.00%> (+<0.01%)	⬆️
tests/util/test_cll.py	100.00% <100.00%> (ø)
recce/util/cll.py	94.79% <93.54%> (+0.38%)	⬆️

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.