alamb commented on PR #18552: URL: https://github.com/apache/datafusion/pull/18552#issuecomment-3765532088
> Use DFField as you've mentioned. This was actually the first path I tried but more plumbing was necessary and for this PoC I used FieldMetadata instead. What I had in mind is that when creating the execution plan, all "LogicalTypes" "disapper" and the information necessary should be extracted by the planner. I think this is the key observation / design point -- specifically, what is the shape of "information necessary" 🤔 It seems to me that we already have broad consensus that `Field` and `Metadata` is the way and these changes are slowly making their way across DataFusion. Thus, from where I sit the only way we are going to get LogicalTypes through DataFusion without a bunch more re-plumbing will be to use the `Field` metadata. However, as you have noted, the `Field` metadata has only a Hashmap of strings, (not a pointer) so to find the relevant code we'll either need have a registry or attach some sort of pointer to the Field (aka some sort of `dyn Any`) At least that is where it looks to me from here -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
