Skip to content

[BUG]: column answer mishandles some column names at the edges #1753

Open

Description

@ralphstodomingo

Description

impact_analysis's column answer and dbt_lineage mishandle some column names, which a review of #1710 found. No model is ever reported unaffected; these are wrong names, "has no column" errors, or a missing column.

  • A real column named _1 that a select * child inherits is dropped from the child: columnReach(child, "_1") throws "has no column".
  • A real alias named like the engine's own _N loses its edge when the engine also gives a _col_N for the same source column.
  • select *, val as id settles without the id_1 DuckDB makes, so id_1 throws "has no column".
  • A COLUMNS() call anywhere in the SQL makes a real "col_\0" alias disappear.
  • On Postgres and Redshift an unaliased cast of a column (amount::numeric) is named after the column, but the answer labels it _col_1, so the trace stops there.
  • DuckDB names a join's second val val_1; dbt_lineage for a select * child misses it.
  • The text line "K of them with no sign (not traced)" can read as "no effect".

Plugins

None needed; the CLI (altimate tool run impact_analysis and dbt_lineage).

OpenCode version

main after #1710.

Steps to reproduce

  1. DuckDB: f documents id, amount; r = select id, amount as _1 from f; child = select * from r.
  2. Run columnReach(child, "_1"): it throws "child has no column _1".
  3. The other shapes above reproduce the same way, each with its one-line model.

Screenshot and/or share link

No response

Operating System

Any

Terminal

Any

Activity

  1. ralphstodomingo commented on Nov 16, 2026

    @ralphstodomingo
    ContributorAuthor

    Deferred from #1754, for a follow-up. None is a regression against the current release; each under-traces or mislabels edge input, and nothing is reported unaffected:

    • A repeat beside a qualified or CTE-wrapped star (select f.*, val as id from f f, with q as (select *, val as id from f) select * from q) still settles, so r.id_1 gives "has no column" on DuckDB, as before. Only a bare outer * is checked.
    • A written _N/_col_N is looked for anywhere in the SQL, so a table alias or WHERE column with that name keeps the engine's label (a _1 table alias keeps a fake _1; a _col_1 relation stops a cast rename).
    • distinct on (...) with nested parentheses, and mixed cast chains (cast(amount as numeric)::int), keep _col_N.
    • Aliases after t.mod or a complete interval literal (t.mod id, interval '1 day' id) aren't read.
    • dbt_lineage shows a model's own cast edge under the engine's label while column_answer names it after the column.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions