rdblue opened a new pull request #924: Add void transform that always produces null URL: https://github.com/apache/incubator-iceberg/pull/924 This adds a new transform function, `void`, that always produces a null value. Because `void` and `null` are Java keywords, the `PartitionSpecBuilder` is configured using `alwaysNull`. The purpose of this transform is to be a stand-in for partition transforms that are removed from a spec. In the v1 table format, IDs for partition fields are not tracked by `PartitionSpec`. Instead, they are assigned starting at 1000 for each spec. Because tables may have more than one spec, manifest files could have incompatible partition field structs. This is not a problem for job planning because each manifest is read independently, but it can break metadata tables that show a union of all manifest data files or entries. The `void` transform can be used to avoid a problem with ID assignment. If a table has two partition fields, `1000: categorical string, 1001: ts_day int`, then removing the `categorical` partition will create a new partition spec with `1000: ts_day int`. That would create a problem in the metadata tables. Instead of deleting the categorical partition, it should be replaced with a `void` partition to keep the IDs aligned: `1000: always_null string, 1001: ts_day int`.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
