libraries/tock-register-interface: split out into separate repository

[lschuermann]

Pull Request Overview

This commit removes tock-register-interface from the kernel repository source tree. From this point onward, this subtree instead lives at https://github.com/tock/tock-register-interface. Revision b512f97e5eb5 of the tock-register-interface repository corresponds to this remove tree's contents, modulo added license headers, a CHANGELOG entry, and a logo for the repository.

This change is motivated (and based on) PR #4587, and implements a long-standing goal of maintaining tock-registers as a first-class crate with its own stability guarantees, separate from the kernel. It also follows in line with other kernel dependencies which we depend on for the Tock kernel, but maintain as part of the Tock GitHub organization, and thus retain full control over.

Testing Strategy

make prepush

TODO or Help Wanted

N/A

Documentation Updated

• Updated the relevant files in /docs, or no updates are required.

Formatting

• Ran make prepush.

[brghena]

crates.io will need to be updated to point at the new repo. https://crates.io/crates/tock-registers

[lschuermann]

crates.io will need to be updated to point at the new repo. https://crates.io/crates/tock-registers

Good point, we should do that after merging, presumably.

Edit: actually, it might be that we cannot update this information standalone, as it is pulled out of the Cargo.toml. I will update that file in the tock-register-interface repo. But the metadata might be incorrect until we tag the next release.

[lschuermann]

While we're at it, should we resolve the inconsistency in calling it tock-register-interface in some, and tock-registers in other places? The crate is called tock-registers, but the repo is, and subdirectory was called tock-register-interface, which is unnecessarily inconsistent and confusing.

I propose we call everything tock-registers, as that's already used for the crate and changing that would be breaking. This seems like the exact right time to rename the repository, before anyone relies on it. @tock/core-wg (and in particular @ppannuto @brghena) thoughts?

[bradjc]

We can't do this without updating our external dependencies policy as it does not permit external dependencies in crates like arch and chips.

[bradjc]

I support standardizing on the name tock-registers.

[lschuermann]

We can't do this without updating our external dependencies policy as it does not permit external dependencies in crates like arch and chips.

I thought that the (verbal) argument around https://github.com/tock/firmware/ was that we're more comfortable with "external" dependencies that are under our control, i.e., ones under the tock GitHub organization. But I acknowledge that was for dependencies of boards (and perhaps individual capsule-crates), and we'd need to codify this exception in the ExternalDependencies.md document. I can draft up a PR.

[lschuermann]

@bradjc Another semantics question we haven't addressed in this document is what constitutes an "external dependency"? External to the kernel git repo? External to the Tock git org? Is depending on a crate maintained in the kernel repo, but pulled in using crates.io external? What about a crate in the Tock GitHub org, and depended upon with a git hash?

[lschuermann]

I support standardizing on the name tock-registers.

I have done the rename: https://github.com/tock/tock-registers
Nothing should have used or reference the old name, given that the old repository path contained very outdated code up until an ~hour ago.

[ppannuto]

(edit) Discussion moved to #4589

[bradjc]

We should add a link in the libraries readme to the new location.

[lschuermann]

We should add a link in the libraries readme to the new location.

@bradjc Done. I removed all other references to the old crate path in the repository as well, so with the Cargo.lock in place this should be ready to go now.

Right now, I expect merging this to fail. This is because this tree was branched off before we checked in the Cargo.lock file, and hence it's still .gitignored here, so the PR doesn't update it. I'd like to try merging this nonetheless, as a sanity check to make sure our CI checks for Cargo.lock files work. Once the merge fails, I'll update the PR to include the required lockfile changes as well.

Edit: it never came to this. Because CI runs on the merged result, which includes a new crate (pci-x86 that references the old tock-register-interface crate path and fails. So I'll simply rebase this PR which will fix the issue.