Erik Welch

That's an interesting case with "assign" with duplicate indices. I see you discuss this (or something very similar) in section 9.9. If I understand correctly, I think what you want...

It's good to bring select into the conversation, but I really would like a better reduction with binaryop. For example, performing a reduction on one field of a user-defined type...

I like that your select variant has a clean API that is easy to understand. Instead of operating on the absolute index I or J of an element, it operates...

Thanks for taking the time to reply. I think I can do a better job explaining what I consider most important in this issue, so please indulge me as I...

Understood, thanks. There are a few reasons why being able to set the name of the UDT when serializing would be nice for us: - To be able to save...

I see. My bad again for not reading the docs thoroughly. I guess a more general goal would be to allow users to include extra metadata (perhaps as a string)...

Requiring the name to be a valid identifier really is kind of a pain. In Python, making new UDTs is as easy as: ```python # Simple struct >>> gb.dtypes.lookup_dtype({'x': int,...

Thanks for the consideration. I'm looking forward to a JIT, although I'm a little paranoid about struct alignment. imho, it may be reasonable to distinguish between a _name_, which is...

Thanks again for your time. Yeah, both of those options _probably_ work for me. Does or will your serialization include the `type_defn` string? If so, then, yes, both solutions work...

> I think it would be better to start simple, and see how far we can take the existing monoid/semiring idea I agree with this sentiment. I think there is...