Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/6075
The dependencies of the `flink-connector-filesystem` are not well set up
already, having an Avro dependency and a Hadoop dependency. I agree that it
would be good to not introduce yet more dependencies, or at the very least,
make them optional dependencies.
FYI: In the re-work of the BucketingSink under
[FLINK-9749](https://issues.apache.org/jira/browse/FLINK-9749), we want to
introduce a `BulkEncoder` interface in `flink-core` that is used by the
BucketingSink, and can be implemented by classes in `flink-orc` (and later a
new `flink-parquet` project). That way we cleanly separate dependencies of the
projects.
---
