MIR borrowck doesn't accept the example of iterating and updating a mutable reference #56649
Conversation
This commit adds the test for writing into a projection of a local to confirm there are no remaining borrows.
rust-highfive
added
the
S-waiting-on-review
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
davidtwco
added
S-waiting-on-author
davidtwco
force-pushed
the
issue-46589
branch
from
b43129f to
17f2976
Compare
davidtwco
changed the title
davidtwco
added
S-waiting-on-review
|
This PR now takes a more reasonable approach, I expect it'll require some changes but it should be good to review now. |
davidtwco
force-pushed
the
issue-46589
branch
from
17f2976 to
a3021a7
Compare
davidtwco
force-pushed
the
issue-46589
branch
from
a3021a7 to
fb6c7fc
Compare
| // a projection (`foo.bar`). | ||
| self.kill_borrows_on_local(sets, local); | ||
| } | ||
| self.kill_borrows_on_place(sets, location, lhs); |
There was a problem hiding this comment.
There is a NOTE in a comment below this line, and I always wonder whether the directive in this comment has been (or is being) followed...
The local assigned_place has presumably been alpha-renamed, probably to lhs here)... but in any case, do we need to make modifications to propagate_call_return to make the logic here match up with the logic there? Are the comments still relevant in the first place?
There was a problem hiding this comment.
Okay, after thinking the matter through, I think the PR as it stands is sound; it is just more conservative than it could be, because I think we could apply the same place-overwritten borrow-killing logic to propagate_call_return, since that represents an overwrite of the dest_place that is passed to that method.
It would be best to first come up with a separate example illustrating this, rather than blindly adding the code to the PR.
There was a problem hiding this comment.
Experimented with adding similar logic to the propagate_call_return function for a test case that might have been right. Not sure it is possible to get the MIR building to generate a call terminator that assigns into a destination like (_1.0) which would be required for this.
This commit extends previous work to kill borrows from a local after assignment into that local to kill borrows from a projection after assignment into a prefix of that place.
davidtwco
force-pushed
the
issue-46589
branch
from
fb6c7fc to
db635fc
Compare
This avoids all sorts of confusing issues with using both `dest_place` and `self` in the `propagate_call_return` function in the `BitDenotation` implementation for `Borrows`.
|
@bors r+ |
|
📌 Commit 7b628e1 has been approved by |
bors
added
S-waiting-on-bors
MIR borrowck doesn't accept the example of iterating and updating a mutable reference Fixes #46589. r? @pnkfelix or @nikomatsakis
|
☀️ Test successful - status-appveyor, status-travis |
Fixes #46589.
r? @pnkfelix or @nikomatsakis