Possibilities

[Dennis4b]

Hi,

The ergonomics (... derives ReadWriter) of upickle have become very good, as I've migrated from Scala 2.13 to Scala 3.5.

For my browser <-> server API I had been using boopickle, which unfortunately does not really seem to be supported anymore and has an annoying runtime-only bug with nested sealed traits on Scala 3, which as upickle shows should not be an issue in modern Scala versions.

There are a couple of things where boopickle, for this particular purpose, and of course only IMHO, has some nice features, and I was wondering if these would be possible to implement in upickle. I'd be willing to take a stab at it but don't know where to start.

Specifically:

• If my API uses upickle the resulting Javascript bundle size is about 1,5 megabytes larger than when using boopickle (for about ~1500 types). I don't know if code size is at all a consideration, but I wonder what is possible to improve on this. Despite the new module optimizations in scala-js this increase still hurts as picklers tend to clump together due methods operating on [T : ReadWriter].

• boopickle uses the implicit index/order of the fields to populate an object. This is a great feature as it cuts down on the network traffic (if I send 10,000 books in a List, I don't need to repeat "title" 10,000 times, "date" 10,000 times, and so on, if I simply know that for Book the first field is the title and the second is the date), and also obfuscates somewhat (instead of myapp.usertype.admin it would just have 1 as that is its index in the type definition).

Since upickle supports both json and messagepack, I wonder how hard it would be to create a new serialization format that uses the implicit order of fields (thus requiring client and server to be based on the same definitions) ? I imagine this gets challenging with trait inheritence etc, but I can't really judge this myself.

Boopickle also deduplicates and sizes-optimizes encodings, but this is, for me, less important. I'd like to mostly avoid using and repeating the key names, and am always interested in reducing generated code size.

[lihaoyi]

Both things you mentioned should be possible. Someone would need to work on it, but there arent any technical blockers