Trino cannot read an Iceberg table that has dropped a partition field

[Kyo91]

After removing a partition field from an Iceberg table (using Iceberg/Spark's Table Evolution API), trino is no longer able to read from the table. Minimal example below:

SparkSQL to initialize the table:

-- mycatalog is configured to point to a Hive Metastore Service, using a common name across Spark & Trino
spark-sql> CREATE SCHEMA mycatalog.partition_evolution; 
spark-sql> CREATE TABLE mycatalog.partition_evolution.example (category STRING, n INT) USING ICEBERG PARTITIONED BY (category);
spark-sql> SELECT * FROM mycatalog.partition_evolution.example;

Trino query works as expected:

trino> SELECT * FROM mycatalog.partition_evolution.example;

Adding a new partition continues to work across both:

spark-sql> ALTER TABLE mycatalog.partition_evolution.example ADD PARTITION FIELD n;
spark-sql> SELECT * FROM mycatalog.partition_evolution.example;

trino> SELECT * FROM mycatalog.partition_evolution.example;

However, removing a partition field (either one) causes the query to fail for Trino

spark-sql> ALTER TABLE mycatalog.partition_evolution.example DROP PARTITION FIELD n;
-- spark still reads this fine
spark-sql> SELECT * FROM mycatalog.partition_evolution.example;

trino> SELECT * FROM mycatalog.partition_evolution.example;
Query 20210615_201107_00011_exxj8 failed: Unsupported partition transform: 1001: n: void(2)

At this point the table becomes unreadable within trino with seemingly no way to recover to a readable state.

[Kyo91]

From what I can tell, this is because when Iceberg deletes a partition field, it creates a VoidTransform, the justification for this is provided here. When Trino attempts to determine the correct transform in this switch statement it does not match on "void".