perf(cll): skip Jinja compilation when compiled_code exists

[even-wei]

Summary

• Skip expensive Jinja re-rendering in CLL when manifest nodes already have compiled_code (from dbt compile)
• Build table_id_map from parent node aliases instead of custom ref/source functions
• Fall back to Jinja rendering if two parents share the same alias (prevents incorrect column mappings)

Changes

• recce/adapter/dbt_adapter/__init__.py: In get_cll_cached(), check for compiled_code on the manifest node before calling generate_sql(). If present, use it directly and build table_id_map from parent aliases/identifiers. Detect alias collisions and fall back to Jinja path.
• tests/adapter/dbt_adapter/test_dbt_cll.py: 5 new tests — basic compiled_code, custom alias, source parent, multiple parents (join), alias collision fallback.

Context

generate_sql() calls into dbt's Jinja engine (macro resolution, context setup, rendering) for every node during CLL. When a user has already run dbt compile, the manifest contains fully-rendered SQL — re-compiling is wasted work.

sqlglot's qualify() strips database/schema qualifiers, so d.node is always just the table name (= model alias). This means table_id_map can map alias → unique_id directly from the manifest.

Test plan

• All 44 tests/util/test_cll.py tests pass (pure SQL parsing, unchanged)
• All 20 existing tests/adapter/dbt_adapter/test_dbt_cll.py tests pass (Jinja path unchanged)
• 5 new compiled_code tests pass (basic, alias, source, multi-parent, collision)
• Manual test with a real dbt project that has compiled_code in manifest

Size

+269/-50 across 2 files

🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

[even-wei]

Self-Review

Correctness ✅

• sqlglot qualifier stripping verified: cll() always returns just the table name in d.node regardless of qualification level ("db"."schema"."table" → table). Confirmed via direct testing.
• table_id_map mapping is correct: Pre-compiled path maps alias.lower() → unique_id, which matches what sqlglot returns in d.node. Schema dict keys also use the alias, matching sqlglot's qualify() expectations.
• Alias collision fallback: When two parents share an alias, table_id_map is cleared and pre_compiled is set to None, correctly falling through to the Jinja path with globally unique table names.
• lru_cache compatibility: Cache key is (node_id, base) — compiled_code comes from the manifest which is stable during a session. No stale cache risk.

Edge Cases Covered ✅

Scenario	Handling
No compiled_code (default)	Falls back to existing Jinja path — zero behavioral change
Custom alias on parent	Uses alias attr, tested in test_cll_with_compiled_code_custom_alias
Source parent	Uses identifier (or name), tested in test_cll_with_compiled_code_source
Parent not in nodes/sources	continue — silently skipped (matches existing behavior)
Alias collision (cross-project)	Detects and falls back to Jinja, tested in test_cll_with_compiled_code_alias_collision_falls_back
compiled_code present but SQL parsing fails	Caught by existing except RecceException/Exception → _apply_all_columns(node, "unknown")

Potential Concerns

1. Test helper _set_compiled_code patches compiled=True alongside compiled_code because dbt's CompiledResource.__mashumaro_post_serialize__ strips compiled_code from to_dict() when compiled=False. This is only needed in tests — production manifests loaded from JSON already have compiled=True.

2. Repeated alias resolution (3 occurrences of getattr(parent_node, 'alias', None) or parent_node.name): Considered extracting a helper but it would be a premature abstraction for the current scope.

Verdict

No critical or important findings. Implementation is correct and well-tested. Ready for human review.

[codecov]

Codecov Report

❌ Patch coverage is 97.00599% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
recce/adapter/dbt_adapter/__init__.py	94.38%	5 Missing ⚠️

Files with missing lines	Coverage Δ
tests/adapter/dbt_adapter/test_dbt_cll.py	99.02% <100.00%> (+0.32%)	⬆️
recce/adapter/dbt_adapter/__init__.py	77.69% <94.38%> (+0.87%)	⬆️

... and 5 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.

[wcchang1115]

Tested with a Snowflake dbt project, and the pre-compiled path worked correctly.
Thanks!