Lifetimes bug on beta 1.31: the associated type may not live long enough #55756
Description
A reduced test case (play):
#![crate_type="lib"]
pub trait Database<'a> {
type Guard: 'a;
}
pub struct Stateful<'a, D: 'a>(&'a D);
impl<'b, D: for <'a> Database<'a>> Stateful<'b, D> {
pub fn callee<'a>(&'a self) -> <D as Database<'a>>::Guard {
unimplemented!()
}
pub fn caller<'a>(&'a self) -> <D as Database<'a>>::Guard {
let state = self.callee();
unimplemented!()
}
}Error diagnostic output in details block below
error[E0311]: the associated type `<D as Database<'_>>::Guard` may not live long enough
--> src/lib.rs:14:13
|
14 | let state = self.callee();
| ^^^^^
|
= help: consider adding an explicit lifetime bound for `<D as Database<'_>>::Guard`
note: the associated type `<D as Database<'_>>::Guard` must be valid for the block suffix following statement 0 at 14:9...
--> src/lib.rs:14:9
|
14 | / let state = self.callee();
15 | | unimplemented!()
16 | | }
| |_____^
note: ...so that variable is valid at time of its declaration
--> src/lib.rs:14:13
|
14 | let state = self.callee();
| ^^^^^
error[E0311]: the associated type `<D as Database<'_>>::Guard` may not live long enough
--> src/lib.rs:14:13
|
14 | let state = self.callee();
| ^^^^^
|
= help: consider adding an explicit lifetime bound for `<D as Database<'_>>::Guard`
note: the associated type `<D as Database<'_>>::Guard` must be valid for the block at 13:63...
--> src/lib.rs:13:63
|
13 | pub fn caller<'a>(&'a self) -> <D as Database<'a>>::Guard {
| _______________________________________________________________^
14 | | let state = self.callee();
15 | | unimplemented!()
16 | | }
| |_____^
note: ...so that references are valid when the destructor runs
--> src/lib.rs:14:13
|
14 | let state = self.callee();
| ^^^^^
error[E0311]: the associated type `<D as Database<'_>>::Guard` may not live long enough
--> src/lib.rs:14:21
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^
|
= help: consider adding an explicit lifetime bound for `<D as Database<'_>>::Guard`
note: the associated type `<D as Database<'_>>::Guard` must be valid for the method call at 14:21...
--> src/lib.rs:14:21
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^
note: ...so type `<D as Database<'_>>::Guard` of expression is valid during the expression
--> src/lib.rs:14:21
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^
error[E0311]: the associated type `<D as Database<'_>>::Guard` may not live long enough
--> src/lib.rs:14:21
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^
|
= help: consider adding an explicit lifetime bound for `<D as Database<'_>>::Guard`
note: the associated type `<D as Database<'_>>::Guard` must be valid for the destruction scope surrounding statement at 14:9...
--> src/lib.rs:14:9
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^^^^^^^^^^^^^^
note: ...so that references are valid when the destructor runs
--> src/lib.rs:14:21
|
14 | let state = self.callee();
| ^^^^^^^^^^^^^
error: aborting due to 4 previous errors
Original bug report
Versions
rustc 1.31.0-beta.4 (04da282 2018-11-01)
cargo 1.31.0-beta (efb7972a0 2018-10-24)
Expected behavior
Everything compiling correctly like it does on stable 1.30
Actual behavior
A lot of lifetime-related errors:
error[E0311]: the associated type `<D as trie::Database<'_>>::Guard` may not live long enough
--> /home/mike/dev/etcdev/sputnikvm/stateful/src/lib.rs:536:23
|
536 | let account = state.get(&address);
| ^^^^^^^^^^^^^^^^^^^
|
= help: consider adding an explicit lifetime bound for `<D as trie::Database<'_>>::Guard`
note: the associated type `<D as trie::Database<'_>>::Guard` must be valid for the method call at 536:23...
--> /home/mike/dev/etcdev/sputnikvm/stateful/src/lib.rs:536:23
|
536 | let account = state.get(&address);
| ^^^^^^^^^^^^^^^^^^^
note: ...so that a type/lifetime parameter is in scope here
--> /home/mike/dev/etcdev/sputnikvm/stateful/src/lib.rs:536:23
|
536 | let account = state.get(&address);
| ^^^^^^^^^^^^^^^^^^^
It has been an issue in nightly 1.31 for a long time now, but I haven't reported the issue while it was just nightly. Now the regression is in beta channel and may land in stable, breaking existing code.
My own findings on the subject: regression is related to an old lifetime system, it's not reproducible with nll feature on, so on nightly the code can be compiled with nll, but not without it. Same holds for beta now except nll isn't quite there yet.
The subjected codebase: https://github.com/ETCDEVTeam/sputnikvm/tree/master/stateful