Disable some nice region errors in NLL mode. #53115
Conversation
rust-highfive
added
the
S-waiting-on-review
|
This disables all but the |
| @@ -65,10 +65,16 @@ impl<'cx, 'gcx, 'tcx> NiceRegionError<'cx, 'gcx, 'tcx> { | |||
| } | |||
|
|
|||
| pub fn try_report(&self) -> Option<ErrorReported> { | |||
There was a problem hiding this comment.
I think we should add a try_report_from_nll method instead, and invoke that from the NLL borrow checker, vs checking the flag. Just seems a bit cleaner, but it may also improve the error messages in some cases below
| | -- -- ^ function was supposed to return data with lifetime `'b` but it is returning data with lifetime `'a` | ||
| | | | | ||
| | | lifetime `'b` defined here | ||
| | lifetime `'a` defined here |
| LL | | ||
| LL | if x > y { x } else { y } //~ ERROR lifetime mismatch | ||
| | ^^^^^ ...but data from `x` is returned here | ||
| | ^^^^^ requires that `'1` must outlive `'a` |
There was a problem hiding this comment.
interesting that we highlight the comparison here -- seems like a problem, though obviously not from this change...maybe we should file a bug? Presumably the actual span we should highlight is the use of x...
There was a problem hiding this comment.
This is mentioned on the .nll.stderr/.stderr comparison here.
| LL | | ||
| LL | x //~ ERROR lifetime mismatch | ||
| | ^ ...but data from `x` is returned here | ||
| | ^ function was supposed to return data with lifetime `'a` but it is returning data with lifetime `'1` |
There was a problem hiding this comment.
maybe we should file a bug here -- I think it'd be nice to add another note here and use this as an opportunity to teach the elision rules. e.g.,
error: unsatisfied lifetime constraints
--> $DIR/ex1-return-one-existing-name-return-type-is-anon.rs:18:5
|
LL | fn foo<'a>(&self, x: &'a i32) -> &i32 {
| -- - - defaults to `&'1 i32` per the elision rules
| | |
| | let's call the lifetime of this reference `'1`
| lifetime `'a` defined here
LL |
LL | x //~ ERROR lifetime mismatch
| ^ function was supposed to return data with lifetime `'a` but it is returning data with lifetime `'1`
| | ^ ...but data from `value` flows into `data` here | ||
| | -- -- lifetime `'b` defined here | ||
| | | | ||
| | lifetime `'a` defined here |
There was a problem hiding this comment.
I think we should file a bug here too -- in general, it's often dissatisfying to highlight the declaration. To do better, though, we'd presumably have to look at the MIR and see if we can identify the parameter. (You could imagine checking if the lifetime only appears in one parameter's type and using that to guess that we should highlight the parameter type, also)
There was a problem hiding this comment.
issue-52663-lifetimes-not-included-in-span in Zulip and this comment discussed this briefly as a potential part of #52973.
|
@bors r+ p=1 Giving p=1 because NLL |
|
📌 Commit 785b06c36804529092d304426f785b5e31cdcee7 has been approved by |
bors
added
S-waiting-on-bors
|
The job Click to expand the log.I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact |
|
Oops! Fixed those |
|
@bors r+ |
|
📌 Commit 37ba9ca has been approved by |
Disable some nice region errors in NLL mode. Fixes #52739. cc #52742. r? @nikomatsakis
|
☀️ Test successful - status-appveyor, status-travis |
Fixes #52739. cc #52742.
r? @nikomatsakis