Open bigfootjon opened 2 months ago
Hey @bigfootjon — thanks for this.
So, yes, we should definitely leverage the new APIs where we can.
The standard approach would be to try and import the new code (or branch on Django version number), use it if it's available, otherwise fallback to our existing code.
We won't drop 4.2 support until 5.2 is released (and maybe a teeny bit after that) but that's no reason we can't get folks on the latest versions updated.
We test against Django main branch, which is already 5.1 (although pre-alpha), so no-reason not to already begin with the code for the upcoming version.
Perhaps separate pull requests for sessions
and auth
to simplify testing and code review and also to provide more flexibility in deploying the PRs.
Sounds good, I'll put up 2 different PRs and import testing to see if the new features exist.
While figuring out what would be needed to do that I noticed some bits in the docs that need to be updated for async django ORM support: https://github.com/django/channels/pull/2090
Here we go:
I also noticed an out-of-date fallback import in https://github.com/django/channels/pull/2091
As discussed on the Django forumns Carlton pointed out that one area we'd like to invest in in is Channels support for the new async functionality that has been added over the last several releases (refer to the above thread for details).
I've scanned the channels repo by searching for
database_sync_to_async
and observed a few areas that could benefit from these new APIs:(this list is the same as Carlton pointed out on that thread, something tells me he might be some kind of expert!)
The other calls to
database_sync_to_async
are from wrappers for intentionally sync code, or tests (neither of which are pertinent).I'd like to propose using these native APIs. The one hitch I see is that these features are too new to be used presently:
Considering that Channels currently supports 4.2 (and definitely doesn't yet support unreleased 5.1) we might have to do some feature-checking or wait until Channels bumps its minimum supported version. Should I start now with writing some polyfills/feature-checks or set this aside until minimum versions are bumped?