jianguotian opened a new pull request, #9086:
URL: https://github.com/apache/paimon/pull/9086
### Purpose
`RawFileSplitRead` can be reused with different projected `RowType` values,
but its cached `FormatReaderMapping` entries are keyed only by file schema and
format. Reconfiguring the reader can therefore reuse a stale projection. In
addition, lazily-created file readers need to retain the output type selected
when each reader is created.
This PR:
- invalidates cached format mappings only when `withReadType` receives a
different `RowType`;
- captures the current output type when a reader is created;
- passes that same type through the mapping builder, lazy file-reader
suppliers, and `DataFileRecordReader`;
- preserves existing Manifest partition-column restoration when partition
fields are requested.
The core change is format-neutral. It does not add Direct Bundle or writer
behavior.
### Tests
Added coverage for:
- changed projections not sharing cached format mappings;
- equal `RowType` instances reusing the existing mapping;
- physical Mosaic files containing only business columns;
- full, business-only, and partition-only projections;
- row tracking and existing-versus-future lazy reader stability.
Focused validation:
```bash
mvn -B -ntp \
-pl paimon-common,paimon-core,paimon-mosaic -am \
-Dtest=ColumnarRowIteratorTest,RawFileSplitReadTest,PartitionColumnsOutsidePhysicalMosaicTest
\
-Dsurefire.failIfNoSpecifiedTests=false test
```
- `ColumnarRowIteratorTest`: 1 passed
- `RawFileSplitReadTest`: 2 passed
- `PartitionColumnsOutsidePhysicalMosaicTest`: 2 passed
- Checkstyle, Spotless, Maven Enforcer, and `git diff --check`: passed
--
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]
