Feature proposal: structured concurrency

[Graeme22]

After reading a fascinating article on structured concurrency, I'm convinced it's a better way to write async code. It allows for better guarantees, more predictable behavior, easier bug fixes, and simplified code.

Expected Behaviour

The functionality of the library wouldn't change at all. Rather, we'd replace use of asyncio with anyio across the whole code base. This then allows users to enjoy the guarantees of structured concurrency when using coredis. Additionally, I expect that reliability will improve across the project and future bugs will be averted. As a bonus, any users who prefer trio to asyncio could use coredis--currently, there is no maintained Redis client library for trio users.

Let me know if this sounds interesting, I would like to contribute.

[alisaifee]

This is great! I've wanted to try anyio out but kept putting it off! Excited to see what this ends up looking like!

[Graeme22]

Note that this will be a pretty large rewrite. anyio disallows a lot of things this library does (eg command_request returns a Future which then has to be awaited). But it should make things more reliable and simpler.

[alisaifee]

command_request returns a Future which then has to be awaited

So, this was only added in the 5.0 release and essentially allowed all the redis api methods to become regular functions that returned awaitables. This is also what unlocked the .transform functionality. I'm curious - is this a deal breaker or is there a way forward?

[Graeme22]

command_request returns a Future which then has to be awaited

So, this was only added in the 5.0 release and essentially allowed all the redis api methods to become regular functions that returned awaitables. This is also what unlocked the .transform functionality. I'm curious - is this a deal breaker or is there a way forward?

I think it can still be done with a little creativity

[agronholm]

Perhaps this will help with the planning? https://github.com/agronholm/redis-anyio
I never finished the project because deploying it to production at my work project back then did not go as planned, and I didn't quite figure out where the problem was. It's also missing important features like watchers and sentinel/cluster support. But it's built for structured concurrency from the ground up, so it might be useful to you as a reference.

[Graeme22]

Hi @agronholm, I've actually already been looking at that repo as well as your PR in order to inform some of the changes going on in #294. Thanks for the great work, should make getting this going a lot easier!