Summary

Closes #10307
Tested and verified in this PR: borodark/power_of_three#7
This PR fixes Postgrex/Ecto connectivity to Cube SQL API by addressing two issues:

1. LEFT JOIN with OR conditions - DataFusion converts these to Filter(CrossJoin), losing LEFT JOIN semantics and causing type bootstrap to fail
2. Missing pg_type columns - typsend, typoutput, typreceive columns were missing or NULL

Changes

New Files

• rust/cubesql/cubesql/src/compile/engine/df/optimizers/cross_join_to_left_join.rs

  New optimizer rule that detects Filter(CrossJoin) patterns for pg_catalog tables and converts them back to proper LeftJoin operations.

  Key features:

  • Identifies pg_catalog table scans by schema characteristics (not table names, which are aliases)
  • Extracts equi-join columns from OR/AND predicates
  • Handles nested JOIN structures
  • Only activates for pg_catalog queries to avoid affecting user queries

Modified Files

• rust/cubesql/cubesql/src/compile/engine/df/optimizers/mod.rs

  • Export new CrossJoinToLeftJoin optimizer

• rust/cubesql/cubesql/src/compile/query_engine.rs

  • Add CrossJoinToLeftJoin to optimizer pipeline (runs first)

• rust/cubesql/cubesql/src/compile/engine/information_schema/postgres/pg_type.rs

  • Add typsend, typoutput, typreceive columns with proper OID values

• rust/cubesql/pg-srv/src/pg_type.rs

  • Add type metadata for binary/text I/O functions

Test Updates

• Updated DBeaver and PowerBI introspection snapshots to reflect new pg_type columns

Testing

• All 655 unit tests pass
• Clippy clean
• Format check passes
• Manual testing with Elixir/Postgrex/Ecto confirms:
  • Type bootstrap completes successfully
  • Basic queries work (Repo.all, Repo.one)
  • Filtering works (where: field == ^value)
  • Aggregation works (group_by, sum, count)

Example Usage

After this fix, Elixir developers can query Cube using familiar Ecto patterns:

# Define Ecto schema for a Cube view
defmodule MyCubes.Orders do
  use Ecto.Schema
  @primary_key {:id, :float, autogenerate: false}

  schema "orders_no_preagg" do
    field :brand_code, :string
    field :total_amount_sum, :float
    field :count, :integer
  end
end

# Query using standard Ecto
import Ecto.Query

query = from o in MyCubes.Orders,
  where: o.brand_code == "Heineken",
  group_by: o.brand_code,
  select: {o.brand_code, sum(o.total_amount_sum)},
  limit: 10

Cubes.Repo.all(query)
# => [{"Heineken", 34968575.0}]

Closes

Closes #10307 (Postgrex/Ecto type bootstrap fails with pg_catalog LEFT JOIN queries)