Hint that a new Rust release may be available

[FranciscoTGouveia]

Closes #4846.

After a rustup command completes (excluding proxy commands), rustup checks whether a new Rust release is available.

To avoid introducing any additional overhead, this feature does not communicate with the release server.
Instead, it checks whether the stable toolchain manifest date is more than six weeks old.
Since Rust follows a regular and predictable release cadence, this should provide a sufficiently accurate indication that a new release is available.

When an outdated stable toolchain is detected, rustup prints a hint suggesting that the user may update their stable toolchain; this hint is shown at most once per day.
Users who prefer not to receive these hints can opt out with rustup set release-hint false.

I have run some benchmarks (100 iterations) for the rustup show command and observed no significant slowdown (1.1x) when we do not show the hint.
However, due to manifest parsing, there is a 1.4x slowdown when the hint is shown (once per day).

Feedback is appreciated on:

1. Whether this approach and its associated overhead are acceptable, or if we should take a different direction;
2. Which rustup commands should display this hint;
3. Whether relying on Rust’s regular release cadence, instead of contacting the release server, is acceptable;

[djc]

LGTM modulo comments. Please squash the commits?

View changes since this review

[djc]

Do we have a strong reason to add this in addition to filesystem-based configuration? I am not a fan of the proliferation of environment variables that change rustup behavior.

View changes since the review

[rami3l]

@djc I don't think #4869 (comment) is considered "configuration" because it just notes the last timestamp the routine is run...

[djc]

I mean the overall disabling via the environment variable; that feels like it should be configuration rather than environment variable.

[FranciscoTGouveia]

Something like rustup set release-hint false?

[djc]

Yes.

[djc]

What duplication do you mean? Perhaps an inline comment would make this clearer?

[rami3l]

@FranciscoTGouveia Sorry, but may I ask what you meant by #4869 (comment)? Looking at this thread I am afraid that we haven't fully addressed all the concerns but please feel free to dismiss me if I missed something.

[FranciscoTGouveia]

Whilst changing the approach from an env var, I created a new ReleaseHintMode enum for the rustup set command. Such an enum is somewhat redundant, since we already have AutoInstallMode, which is basically the same, hence my "duplication" comment.

Apologies for not making this clearer from the start, @rami3l.
My idea was to, after this PR, send a simplification patch to address said duplication.

[djc]

What's the rationale for not just fixing it in this PR?

[FranciscoTGouveia]

My idea was to not elongate this PR further.

I have already fixed it and merged the ReleaseHintMode and AutoInstallMode.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[FranciscoTGouveia]

I believe all review comments are now addressed.
Should I update the initial PR message to reflect the changes done during the review process (i.e., using rustup set release-hint instead of an environment variable)?

[rami3l]

@FranciscoTGouveia I think we can merge this first and see how it goes. Many thanks for making this patch :)

PS: I won't backport this patch to 1.29 yet to give it more confidence WRT stability; if we hit #4942 in the process I think it's safe to just remove the state file. The more important thing here is that @Cloud0310 has a new file to migrate for XDG.