"It is a more conservative approach to introduce that in a new method rather than changing the existing one under the hood and potentially break existing pipelines silently”
I like the idea actually, but if ChangelogMode.INSERT is the default, existing pipelines should be compatible. We can see the other kinds of ChangelogMode as an extension. “for `toDataStream` users need to be able to express whether they would prefer Row, POJO or atomic” I think most of the cases people do not need to convert the stream to a Row or POJO, because the table projection always returns a flatternned internal row, if people did want a POJO there, how about we bind the DataType to the existing schema, like this toDataStream(table, schema.bindTo(DataType)) Best, Danny Chan 在 2020年9月3日 +0800 PM3:18,dev@flink.apache.org,写道: > > It is a more conservative approach to introduce that in a > new method rather than changing the existing one under the hood and > potentially break existing pipelines silently