Fix 2601 #2602
Fix 2601 #2602
Conversation
Signed-off-by: Daniel Silverstone <[email protected]>
Sometimes someone might choose an additional component to add during an install/update which is wildcarded. We need to ensure that we detect this and request the install with the wildcard otherwise we can hit situations such as that identified in rust-lang#2601 Signed-off-by: Daniel Silverstone <[email protected]>
|
I can't use the artifact myself directly, but I can do what I did in NixOS/nixpkgs#106443, only just locally, and pointing to your branch, then I get a "distro I'll have to do that tomorrow though, but thanks for the fix! |
|
I can confirm the sequence from #2601 behaves correctly, but that's because it starts by removing But if I have So as I first suspected, this PR makes Interestingly enough, installing any other component triggers the self-repair, but for |
|
Oh crap yes, if you're already broken then it won't trigger the repair. Hmm, I'll need to decide how to handle this more cleanly. This was mostly to not break in the first place I hope. |
When self-healing from a bad rust-src install as a side effect of someone running `rustup component add rust-src` we would accidentally double-add the rust-src component resulting in a conflict. This fixes that. Signed-off-by: Daniel Silverstone <[email protected]>
|
I think this was being over-eager when self-healing rust-src misinstallation. We ended up double-installing rust-src which is what caused the actual conflict listed. Assuming CI goes green, would you give this branch a test for you. I think it's good now. |
|
My bad, I haven't had a chance to look at this yet, it slipped through the cracks. |
|
@eddyb That's quite alright, everyone's busy :D Just let me know when you can give it a go please? |
|
The issue I described in #2602 (comment) (with a a toolchain from an older Again apologies for the (extra one week) delay, I've been having a hard time doing software things lately. |
|
No need to apologise, this time of year is tough for most people to get time, let alone to get time to deal with small side issues. Thank you for checking it out, I'm going to merge this then. |
Previously we attempted to resolve a situation where a toolchain which had a wildcarded component installed as target-specific due to profiles would break things. Sadly it seems we missed one other mechanism by which such a component installation might occur - namely if someone requests a wildcarded component as an additional component during toolchain installation.
This PR resolves this by detecting such components during the building of the dist changeset. As such, it fixes #2601
@eddyb When this has built, assuming a green CI, if you're logged into Github then you should be able to download an artifact from the relevant CI run and try it out. Would you mind checking if this fixes your problem?