Do you have a preferred message protocol?

JSON with a good schema if you're doing it across languages?

XML, JSON, and YAML belong to a totally diffent class of use than protobuf, flatbuffers, capnproto, etc

It's like comparing apples and screwdrivers.

Exactly because the consumers of those API's exhibit a strong preference for one or the other.

Library/module/API developers often seek to deliver maximimum adaptibility in their interfaces so that consumers of them can satisfy their own more specific/pressing requirements.

imo it's mainly because protos are unusable on the web without something like grpc-web (which requires a proxy) and in general web support in protobufs always seemed like an afterthought, just barely working (if at all) which is kind of ironic considering it came out of Google

I've never used protobufs with web, but I don't see why not. gRPC, on the other hand, requires HTTP/2 which can be a luxury. I remember making a side project that uses gRPC a while ago and trying to deploy on GCP AppEngine, only to find out it doesn't support gRPC.

> I've never used protobufs with web, but I don't see why not

We have a bunch of apis like this where I work. It's a huge pain, tbh. You lose the ability to just curl endpoints, or to see what data came over the wire for debugging, not to mention having to use esoteric client libs because every off-the-shelf one just assumes json over http.

well if you can't use gRPC which, as you correctly notes, is protocol-incompatible with browser APIs, you kinda need to roll your own RPC framework. Various attempts had been made at this - see connectrpc, for example, by the same company that authored this post. imo, big advantage of protobuf/gRPC ecosystem is that Google is doing all the heavy lifting for you, so unless something else catches on with wider community it's going to be pretty risky beyond hobby projects.

Yeah, that's why I wouldn't do it. Not totally unusable, but painful.

I'm sure there's some great schema implementation that you can bolt onto JSON, but personally I have never seen anyone actually using one in the wild.

With protobuf the schema is essential, it is the foundation. I feel this is a huge advantage especially for communication protocols.

We've used it. It's useful to define the contract between teams.

So you have 2 teams do a meeting, we work out a contract we like, dump that into a jsonschema and you can enforce by unit tests that schema is respected on each side (producer/consumer)

JSON is the serialization format. What's the IDL?

I've seen OpenAPI/Swagger in the wild a few times. It's a good way to specify the API for an HTTP server, whether it's JSON or something else. Biggest downside is the confusing name change.