Steven Fackler

It does not use the rustup script: https://github.com/rust-lang/docker-rust/blob/master/1.50.0/buster/Dockerfile#L17

I don't have any ARM hardware to test with unfortunately, but it seems like this could be a problem with how libgit2 is compiled for Armv7 in Cargo's dists: https://www.gnu.org/software/coreutils/faq/coreutils-faq.html#Value-too-large-for-defined-data-type....

Definitely worth checking which specific rustup binary the script ended up downloading. If it was something other than the armv7-unknown-linux-gnueabihf version then that would explain the difference, though I don't...

This hasn't been done historically since it'll impose a pretty huge storage burden on Docker Hub/Github and I don't know if they're okay with that.

> theoretically, one should only build on released software, so using 0.x pre-releases for production software isn't looked on too well. Your theory does not appear to correspond to reality....

What is the point of doing this? If we're going to support non-parking-lot mutexes this should use [antidote](https://docs.rs/antidote/latest/antidote/) rather than reimplementig non-poisoning locks manually.

> What is the point of doing this?

Why are these two approaches better than parking lot with its own dependencies?

How much does this change affect your build times and docker image size?

This definitely seems plausible to me, but I think it'd be best to pull in someone to double check that this is the right move. Thoughts @Mark-Simulacrum?