Styxis
Company
Knowledge

Check reserved-name readiness in Trino metadata queries

Exercise each consumer-facing metadata query for a table with a reserved-word column. Success on another metadata path, or an abbreviated SQL description, cannot stand in for the required query parsing successfully.

Identifier validity and generated SQL

A valid base-table identifier does not prove every generated metadata query quotes it correctly. Truthound Core can compare a prepared interface-readiness state with the requirement that the metadata path needed by operations actually parses. [1, 2, 3]

A reserved-column metadata report

Trino issue #31366 reports $partitions failing on an Iceberg table with a column named values while $files and $snapshots work. The abbreviated INSERT has a schema/value-count mismatch, so it is not a complete reproduction to execute unchanged. A follow-up links a PR without establishing local installation. [1]

Build a valid minimal control

Build a valid minimal schema with quoted reserved and ordinary column-name controls. Record results and errors for $partitions, $files, and $snapshots separately. A working interface must not stand in for a different metadata path required by the consumer. [1, 3]

The narrow parse contract

Declare quoted values to be a valid actual column name and require $partitions to parse. parse_success=true passes this narrow readiness check. false fails even when $files succeeds. It does not establish the correctness of the resulting statistics. [1, 3]

The SDK does not generate SQL

ColumnPairValidator in Truthound Core Python SDK 3.3.2 can compare an operator-recorded non-null parse-success flag with true. The operator executes each interface query and retains the error. The SDK does not generate Trino quoting or repair its SQL. [1, 2, 3]

Review renaming consequences

Use a separately validated minimal reproduction and the ordinary-name control. Before renaming business columns, review downstream dependencies and whether the deployed version has the relevant supported correction. [1, 3]

Acceptance precedes data validity

Keep parser acceptance and returned-data validity as separate approval stages. Treat the report’s incomplete shorthand as source context rather than manufacturing evidence that its exact SQL ran successfully. [1, 3]

Application conditions

Prepare per-interface execution status using a valid schema.

Alternatives

Check an ordinary-name control and a supported quoting path with consumer impact reviewed.

Limitations

  • Do not treat the abbreviated INSERT’s column/value mismatch as execution evidence.

Evidence

Event checked
Subject checked
Modified