Github user gatorsmile commented on the pull request: https://github.com/apache/spark/pull/9548#issuecomment-155835390 @cloud-fan Before discussing the solution details, let us first talk about the design issues. IMO, the `DataFrame` is a query language, kind of a dialect of SQL. Or, maybe, SQL is a dialect of `DataFrame`. We need to formalize it and clearly define the concepts of each major classes like `DataFrame` and `Column`. If `Column` represents a concept independent of `DataFrame`, could you define what it is? If one `Column` with the same ID can appear in different `DataFrame`, how to enforce such a "referential integrity" between different `DataFrame`? If two `Column` with different ID could represent the same entity, should we keep such a relation for generating a better physical plan? In the current implementation, each `Column` actually corresponds to an expression in logical plans, but we are unable to apply an expression above `Column` instances to generate a new expression. So far, `Column` is kind of a wrapper, but it is not a subclass of `TreeNode`. When more components are built on `DataFrame` to access and operate, we have to carefully think about this problem. If possible, I think we need to resolve it in the release of Spark 2.0. Will answer your design suggestion in a separate post.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org