feat: dbt_sql_graph tool — a model's CTEs, joins and filters as a graph - #1735

reviews:
  auto_review:
    drafts: true

• 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.

👋 This PR was automatically closed by our quality checks.

Common reasons:

• New GitHub account with limited contribution history
• PR description doesn't meet our guidelines
• Contribution appears to be AI-generated without meaningful review

If you believe this was a mistake, please open an issue explaining your intended contribution and a maintainer will help you.

👋 This PR was automatically closed by our quality checks.

Common reasons:

• New GitHub account with limited contribution history
• PR description doesn't meet our guidelines
• Contribution appears to be AI-generated without meaningful review

If you believe this was a mistake, please open an issue explaining your intended contribution and a maintainer will help you.

Review of this change (two independent reviewers, then a separate check of the fixes). Each finding and what happened to it, fixed in f6aed61:

Fixed

• Column case — mergeColumns lower-cased everything unless it carried literal quote characters, so Postgres Foo/foo and Snowflake Foo/FOO merged into one column. It now follows dbt Power User's rule: a name the adapter would only keep when quoted keeps its case (adapter-aware, same regexes as isQuotedIdentifier).
• Session id — session_id was only the SQL hash, so two sessions asking for the same SQL sent the same id. It is now <session id>-<md5 of the compiled SQL>, as Power User does.
• Cancel before compile — an abort during parseManifest still compiled the model. The abort signal is now checked after the manifest and after the compile, before the warehouse read.
• constructor / __proto__ columns — a warehouse column with an Object member's name hit the inherited property and wrote to Object.prototype. The merge now uses a Map.
• Model name vs. CTE — a CTE or {{ this }} table named like the model (e.g. with customers as (...) in customers.sql) was overwritten by the model's own select and could form a cycle. The model's select is now <model> (final select) when the name is taken. Power User's remap has the same overwrite; this is a deliberate difference.
• Seeds, snapshots, Python models — a name that resolved to one of those had its file compiled and sent. Only resource_type: model with a .sql path is accepted now.
• Comment — formatSqlGraph no longer claims dependency order.
• Tests — full model_info request with toEqual, both case families, the Object-member names, the name collision (CTE and source), refused node types, cancel during parse and compile, session id per session. dbt-adapter-access.test.ts now skips this tool in its placeholder loop.

Not changed, on purpose

• Credentials are checked inside apiRequest, after the adapter has started and the model has compiled. Checking earlier is a cost nicety; left for a follow-up.
• The backend fetch ignores the abort signal and has a fixed 15 s timeout. That is in the shared client (api/client.ts), not this change.
• Node name vs. edge key, and two parents with the same alias: unconfirmed against the backend's real replies; left as is.
• Column description/meta go to the backend with the compiled SQL: same as Power User.

Cost side effects are unchanged: one backend call and one warehouse metadata read per call.

👋 This PR was automatically closed by our quality checks.

Common reasons:

• New GitHub account with limited contribution history
• PR description doesn't meet our guidelines
• Contribution appears to be AI-generated without meaningful review

If you believe this was a mistake, please open an issue explaining your intended contribution and a maintainer will help you.