Rustup: [Feature Request] Warn users of component availability changes

[Frederick888]

Rustup should warn users of component availability changes before going on with an update.

For instance, the rls component was once renamed to rls-preview and I was quite confused then about the missing of rls after the update. And just yesterday, rls-preview was temporarily removed and guys are now struggling downgrading to the last working one (rust-lang-nursery/rls#535).

So whenever there is a new component available, it's better to inform users of it, e.g.

A new component `shiny-new-component` has been added to `nightly-toolchain`.
Run `rustup component add shiny-new-component` to install it.

And if there is an installed component that is removed, I reckon we should prompt users for a confirmation, e.g.

The component `dev-essential-component` has been removed from `nightly-toolchain 2017-12-12`.
Are you sure you want to continue? [Y/n]

[toothbrush7777777]

Yes, this is particularly important for RLS users. It is incredibly frustrating to find out that the latest nightly toolchain doesn't contain rls-preview _after_ you update. 😭

The least we need is a flag which causes rustup to only update the toolchain if all installed components exist in the latest version.

[nrc]

This should be designed to work well for both users and for automated upgrades (e.g., an editor doing rustup update should be able to not to do that if the update will cause problems.

[davidbarsky]

Hi! I was thinking about how to approach this problem and wanted to get feedback before writing code. Think of this as a mini-RFC/design doc:

1. Before running rustup update, check if static.rust-lang.org/dist/channel-rust-nightly.toml has all the user's installed components (Is there an authoritative source that describes all the user's installed components? I think it is defined in toolchains/<toolchain>/lib/rustlib/multirust-config.toml).

2. If a component availability regression is detected,rustup update should quit and communicate the issue to the user (Should there be a --force flag to update regardless of component availability?)

3. Proceed with the update normally.

I'm not entirely sure how to approach automated upgrades. What's the best way to communicate that to editors?

[imp]

I'd imagine having a special flag for IDE to use (like --auto-update) that will automatically set safe defaults. (No upgrades that break thing). It will be up to IDE plugin maintainer to use this flag for automatic update.

[nrc]

@davidbarsky this sounds basically right to me, and thanks for offering to do the work, it is much appreciated!

Before running rustup update, check if static.rust-lang.org/dist/channel-rust-nightly.toml has all the user's installed components

I believe rustup update downloads this manifest as the first step in doing an update, so you shouldn't have to check it separately. Be aware that the manifest can include renames and these should be taken into account (e.g., if foo is renamed to bar then foo should not be counted as missing).

Note also, that the component must be present for all targets and all channels that the user has installed, and when warning the user, we should list the tagets/channels where the component is missing.

If a component availability regression is detected,rustup update should quit and communicate the issue to the user

yes

Should there be a --force flag to update regardless of component availability?

Yes. And the message to the user should suggest this as a possible solution.

I'm not entirely sure how to approach automated upgrades. What's the best way to communicate that to editors?

I think this setup should work well for editors - the editor can attempt the update and if rustup quits, the editor can try to parse stdout to check for this issue and communicate with the user.

I'm very happy to help out with this - please feel free to ping me any time here or on irc.

[davidbarsky]

this sounds basically right to me, and thanks for offering to do the work, it is much appreciated!

@nrc Of course! I'll be able to take a look on later this week and get a draft out for review (or at least, something that other cans build upon).

I believe rustup update downloads this manifest as the first step in doing an update, so you shouldn't have to check it separately.

Gotcha. Several assorted questions:

• My understanding of rustup is that it tries to update a toolchain (which will include a compiler, etc.) and then install components, which are separate. Is my mental model correct? (The alternate model that I built up is that components are part of a toolchain, so it's one transaction).

• Do you have any package and tests that you recommend looking at that are relevant to this issue?

Be aware that the manifest can include renames and these should be taken into account (e.g., if foo is renamed to bar then foo should not be counted as missing).

Is the rename documented/handled in some existing code, or is rustup unaware of this renames?

I'm very happy to help out with this - please feel free to ping me any time here or on irc.

No, thank you for your help!

[nrc]

My understanding of rustup is that it tries to update a toolchain (which will include a compiler, etc.) and then install components, which are separate. Is my mental model correct? (The alternate model that I built up is that components are part of a toolchain, so it's one transaction

The compiler and libs etc., are just components like the tools. They are all part of a toolchain (rustup is aware of the target vs host difference). There are some differences because some components are optional (like tools) and some are not (like the compiler). Also, tools are installed differently to the compiler and libs.

I'm not sure how 'transactional' rustup actually is, but tools and the compiler are treated pretty much similarly.

Do you have any package and tests that you recommend looking at that are relevant to this issue?

I don't know, sorry

s the rename documented/handled in some existing code, or is rustup unaware of this renames?

Yes. The renames are an item in the manifest and rustup is aware of them.

[Frederick888]

Is it also possible to provide such a way so that new users can specify "essential components" when installing the Rust stack?

[davidbarsky]

Thanks very much, Nick! That’s a useful starting point!

(Sorry my questions weren’t better edited.)

[Xanewok]

Hey @davidbarsky! How is the work coming along?

[davidbarsky]

Hey Igor! Apologies in the delay in responding. Work and life happened, so
this fell to the way-side for a bit.

I did start work on issue, but if you'd like to pick this up, by all means!

On Thu, Dec 28, 2017 at 7:11 AM, Igor Matuszewski [e-mail]
wrote:

Hey @davidbarsky https://github.com/davidbarsky! How is the work coming
along?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/rust-lang-nursery/rustup.rs/issues/1277#issuecomment-354279117,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AB-NPuX2gPjs7wwUpX2ajefchEesjHB4ks5tE4V5gaJpZM4QKK6J
.