Replace all uses of the never type (!) with the stable Infallible
#966
Conversation
|
Looking at the final diff, it looks like VS Code auto-formatted when saving. I can manually undo this if wanted. |
There was a problem hiding this comment.
Given the history of the never type, I think it's reasonable to switch to Infallible, which will later become an alias for !. Aside from this This should be fine to include in a major release.bail!() thing,
This is a breaking API change, albeit unlikely to break much if any code in the wild.
|
NB: this needs a minimum rustc version bump, probably the same version as #967. |
|
@jebrosen This could technically be done without the version bump if necessary, as It would actually be pretty simple to switch over, since we'd just need to add |
|
The version bump should be covered by #967 (merged). I personally think a custom empty enum is the worst option - not in |
|
@SergioBenitez I can perform a rebase if a merge is desired; it shouldn't be too difficult. |
|
LGTM pending rebase. |
For the most part, this is simple substitution and nothing else. In one specific location (core/lib/src/config/mod.rs), the compiler threw an error, as the `Infallible` type is not a special case in the compiler. As such, we have to explicitly tell it that code after that point cannot be reached using `unreachable_unchecked()`.
|
@jebrosen Rebase is complete — should be able to merge after CI is complete. |
There was a problem hiding this comment.
LGTM. I had one point about consistency - some places write out type Error = std::convert::Infallible and some use std::convert::Infallible, even if there's only one use. Similarly, some places use Result<T, Infallible> and some use Result<Self, Self::Error>. That should be easy to do during merge, though.
|
wrt consistency, I actually intended on doing another PR to switch to Edit: A quick |
|
Wonderful! This was merged in 34a741a. |
|
I didn't see your early comment; I switched to using std::convert::Infallible everywhere to make switching back to |
Infalliblewill be stabilized on Thursday.Most of the diff is simple substitution of types (along with the relevant
use).There is one situation that required a compiler hint regarding the function not actually returning;Infallibleisn't a special case in the compiler, so it regards it the same as returning any otherenum. I worked around this by creating a trivial macro (for readability) that addsunsafe { std::hint::unreachable_unchecked() }after a call; this will allow the compiler to avoid related errors and perform the relevant optimizations.Related: Compile on stable Rust (#19).
Pinging @jebrosen for review.