Baunsgaard commented on pull request #1234: URL: https://github.com/apache/systemds/pull/1234#issuecomment-826721786
Looking through your comments, it seems like you insist on the "universal " Operation node approach. This is something that i would like not to happen, because this means for every operation called we would need to differentiate between "am i single return?", "am i multi return?", "am i a matrix?", "am i a frame?". Currently we have the luxury that there is only Matrices and scalars, but we already have a PR for Frames that become a subclass to Operation Node, And in the future we might add more datatype, or abstractions needed. Therefore the perfect design in my book for this multi return support is: - Make a class that is can contain multiple different Operation Nodes(that each correspond to a single output). - This new class should implement the type DAGNode. One of the reasons why this is important to separate, is because many operations supported directly on a OperationNode is not valid on a "MultiNode", like for instance min. Rather than having to introduce an if else inside the min function (if we chose the universal approach) trowing an invalid exception, i would prefer not having that if else. Again comments and discussions are welcome if you disagree -- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
