Reduce the number of required nightly features

[kawogi]

First off: this is an awesome and impressive project!

I just skimmed over the code and noticed the unusual high number of nightly features in use. Most project I've been working with in the past were trying to minimize the use of those features or even mandate being buildable on a stable release. Is there a policy for this project?

Tinkering with the code I found:

1. lots of features can be removed without breaking any code
2. some features are only being used to make the code a little more convenient here and there
3. some features seem to have made it into current stable Rust so bumping the version should be sufficient to get rid of them
4. some features are being used to (seemingly) improve performance (notable portable_simd)

Is there any interest in getting the number of nightly features down? If so I'd try to tackle them in individual PRs.

[Magnie]

@andrewgazelka has talked about reducing nightly feature usage, but not eliminating:

Regarding nightly features - you're right that we use several. We don't have immediate plans to move entirely to stable, as some features like specialization significantly improve our architecture. However:

• We're open to incrementally removing nightly dependencies where practical (please make GitHub issue about it first if it is a major change)

valence-rs/valence#685 (comment)

And here is a PR he is working on for switching to stable it looks like: #944

[kawogi]

Just found this PR: #944

IMO this implies eliminating all nightly features, which answers my request.

[TestingPlant]

I'm not sure if Andrew is still working on #944.

Stable Rust support would be a nice feature, but it's not a priority right now since there are other reasons making Hyperion not ready to be used as a library for third party projects (although people are still welcome to use Hyperion, we just don't have much documentation or any API stability guarantees). There are many other issues to work on that would achieve greater benefits compared to the amount of time required to refactor Hyperion to not use nightly features in my opinion. If you're still interested in stable support, feel free to continue #944 and make sure Andrew doesn't have any unpublished changes to the PR first.

[kawogi]

Taking over Andrew's work would likely be too ambitious for me as it touches lots of code locations.

By reducing the number of nightly features I would've guessed the risk of unnecessarily running into breaking changes and/or ICEs would go down.

Closing for now. Maybe I can find something more important.