Remove BodySerializer

[adamw]

Instead, bodies need to be serialized explicitly. E.g. instead of

import sttp.client4.jsoniter._

basicRequest.body(someInstance)

you now need to call the serialization method explicitly:

import sttp.client4.jsoniter._

basicRequest.body(asJson(someInstance))

[Ravenow]

With this change all code with body serialization should be touched. It would be interesting to know, whats the reason for removing BodySerializers? I mean - one import on top vs explicit functions calls everywhere

[adamw]

Yes, that's true. The reason is that implicit BodySerializers are ... well, implicit ;). They're non-obvious when you look at the code - to which form the code is serialized (if it's json or xml etc.). It's also hard to "discover", using auto-import for example.

So yes, the new approach is more explicit, requires more code, but is also predictable (when working with JSON, you always need asJson, both for request & response) and discoverable

[Ravenow]

Is it possible to keep old way with implicits? In our code base we supply 'case classes' with circe encoder, and thats enough. Without importing of required serializer code will be 'red'.

Its not obvious why sttp force users to use single way as a general purpose library. Someone like explicit approach, someone not

just my 2 cents :)

[adamw]

Yes, some code will be read upon migration, I'm aware of that - but that's why this change happens across major versions (v3 -> v4).

I'd like to keep a single way to do things, if possible. You can always create your own sttp stack (by extending the SttpApi, SttpCirceApi traits), and adding an implicit conversion from T: Encoder to StringBody using asJson

Then you could have a single-sttp import with all the features that you use: import my.package.mysttp.*