Build cloud backends separately

[djmitche]

TaskChampion itself doesn't have a lot of dependencies, and builds pretty quickly. However, the GCP and AWS cloud SDKs each have large dependency trees, with surprisingly little overlap.

What could we do to de-couple those? Ideally saving on one or more of

• Binary size
• Build time
• Dependency churn (i.e., number of dependabot updates)

I don't have any good ideas for how to do this. Runtime dynamic linking in Rust basically requires FFI, so that's not a great option. And invoking a separate process is difficult for a library - how would that work from Python? from C++?

[djmitche]

I suppose one option would be to find or build simpler interfaces to these cloud providers -- the HTTP APIs they are using are not complicated.

[djmitche]

Another option might be a kind of IPC protocol, between processes. In TaskChampion the config would look like a Server that gets a pair of fd's passed to its constructor, and performs the IPC over those fd's. The caller would run a task-sync-aws process and pass its stdin/stdout to this constructor. That task-sync-aws would just wait for requests on its stdin, call a Server implementation, and pass the results back on its stdout.