Provide and use a constructor function for feature creation

[dcj]

We've been discussing/debating how best to represent/implement a feature (in Zulip/Clojurians/data-science).

Currently features are implemented as hashmaps, with two keys, :geometry and :properties, and I/we would like to experiment with other possible implementations/representations (e.g. via deftype, and maybe as a "flat" hashmap.

AFAICT, ovid currently provides functions/methods for operating on a feature, e.g. feature/geometry, feature/properties, etc. which is great.

But it seems like the construction of a feature is currently done by explictly creating a hashmap, which will make testing alternate implementations more difficult.

I propose that we change the code to always create a feature using a feature-constructor function, and then we'd be able to change just that function to experiment with alternate implementations/representations, also, arguably it this would just be a better coding implementation, even if the current hashmap representation is deemed best....

I'm happy to help try and make these changes and submit as a PR, but want to discuss here beforehand to gauge your interest and receptivity to such a change.

[willcohen]

Hi, sorry (again) for the delay! To be clear, you're really just talking about internally, what (-to-feature) does, right? I know one option is hash-map (current version), another is record (which would have a very obvious constructor, with some of the caveats we discussed elsewhere), and another might be the custom deftype. Should ovid be picking any of these anyway, if it's a protocol? If someone made a hash-map that they're happy with (say, by reading in a geojson or something using geo), should we be forcing them into whatever type we think is best? Alternatively, should Featurelike just roll with whatever the user presents to us as long as possible, to avoid roundtrip coercions? The only reason -to-feature is picking anything is because there's no set type. If we were to pick, say, a record AND a deftype, wouldn't the protocol want to deal with the record, deftype, and pure hashmap?