chrysn
commented
2 years ago Right now I'm leaning toward just removing nightly; I don't see any upcoming components that'd need nightly again, and it's the easier path. There's nothing that keeps us from re-introducing it on demand.
In terms of keeping upcoming versions tested, we might consider adding a second container (possibly built on the regular one) that takes nightly and makes it the default toolchain. That container would be useful for those who build Rust code on RIOT and add crates that depend on nightly; it'd get the regular testing any container gets (without, obviously, any guarantee that it builds any nightly-dependent code that's not even in the RIOT tree).
After the Rust 1.65 release (mid-November), we won't have an urgent need for nightly Rust any more. Options then are:
Remove nightly Rust from our images.
Users who build applications that do add nightly-only crates couldn't use our images, just like users who build applications that need anything else fancy in the build system are on their own.
Unpin nightly, and run a test or two additionally with nightlies in the image build process.
This would help users who use such crates, and would also help us spot regressions in stable Rust that hit us (I've never seen such a thing, but we'd contribute to keeping it that way) early on, whenever we update riotdocker.
The nightlies would generally grow outdated at some point, but never be older than the stable Rust we ship, and would be easy to refresh by re-running the Action if we go without any other riotbuild updates for a long time.
I haven't made up my mind, don't have preferences yet, and appreciate input.