plusplusjiajia opened a new pull request, #9503:
URL: https://github.com/apache/paimon/pull/9503
### Purpose
Follow-up to #9455, which was right that the Iceberg mirror must not
publish nanosecond timestamps — Paimon writes them as Parquet INT96, which
Iceberg reads as microseconds. It added two defences, and I think
only one belongs.
`SchemaValidation#validateIcebergTimestampPrecisions` is the right one: it
refuses only when the table has Iceberg metadata enabled, and is registered by
the same condition as the Iceberg publisher itself, so
no table that can publish is missing it.
The other narrowed `IcebergDataField#toTypeString` from 3..9 to 3..6. That
method only names the Iceberg type for a Paimon type — it has no table to ask,
so it also refuses callers that never let Paimon write
the data, for whom the INT96 hazard does not arise. It also deleted the
mapping of 7..9 to the v3 types `timestamp_ns` and `timestamptz_ns`, which
`toDatatype` still reads back as precision 9, so the class has
been claiming to read two names it stopped writing.
Whether nanoseconds may be published is a property of the table, not of
the type. This restores the mapping and leaves the refusal to the validator.
`TIME` has the same shape but is left alone: Iceberg has no
precision-carrying time type and the reverse mapping reads `time` back as
precision 3, so widening it there would trade a loud error for a silent
precision loss.
### Tests
The three classes #9455 touched all pass — `IcebergCompatibilityTest` 48,
`SchemaManagerTest` 57, `IcebergDataFieldTest` 31.
`IcebergDataFieldTest` again requires 7..9 to map to the v3 names.
`IcebergCompatibilityTest`'s commit-path refusal test now runs from a
`MethodSource` and additionally covers
`TIMESTAMP_WITH_LOCAL_TIME_ZONE(9)` — the family this restores a mapping
for, and previously uncovered there. It still passes, i.e. the mirror still
refuses it.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
