Repository navigation
Poor diagnostic combining #[derive(Debug)] with #[repr(C, packed)] struct #110777
Description
Code
#[derive(Debug)]
pub(crate) struct UInt16LE {
bytes: [u8; 2],
}
#[derive(Debug)]
#[repr(C, packed)]
pub(crate) struct Header {
version: u8,
kind: UInt16LE,
}Current output
error[E0507]: cannot move out of `self.kind` which is behind a shared reference
--> src/lib.rs:11:5
|
7 | #[derive(Debug)]
| ----- in this derive macro expansion
...
11 | kind: UInt16LE,
| ^^^^^^^^^^^^^^ move occurs because `self.kind` has type `UInt16LE`, which does not implement the `Copy` trait
|
= note: this error originates in the derive macro `Debug` (in Nightly builds, run with -Z macro-backtrace for more info)Desired output
"cannot form reference to `self.kind` because struct `Header` is `repr(packed)`"
"note: `self.kind` has type `UInt16LE`, which does not implement the `Copy` trait"Rationale and extra context
I thought this was a regression in 1.69, and it sort of was, but presumably the old code was incorrect, since you can't form a reference to a field in a repr(packed) struct. Nothing in the diagnostic pointed me to that, though, so I had to figure out why this struct was different.
Other cases
No response
Anything else?
No response
Reacted by toymil and saruman9
Activity
Lokathor commentedon Apr 24, 2023
This diagnostic is poor, but I'd like to point out that
packedis pointless in theHeaderstruct and you'd have a better time without it.jrose-signal commentedon Apr 24, 2023
This is stripped down from a full example, but you're probably right…it would be enough to use
repr(C)and static-assert that the alignment is 1.RalfJung commentedon Apr 24, 2023
Yeah our diagnostics regressed when we fixed #27060, good point. Not sure how we usually handle special diagnostics for
derive-- do we really want code littered all over the compiler that special-casesderive? I don't think we can emit this diagnostic in the derive code itself, but it would still be valuable for compiler maintenance to keep this clean, somehow.RalfJung commentedon Apr 24, 2023
Also FWIW, the intended fix is to declare
UInt16LEasCopy.The derive macro has to decide whether to take a reference or a copy without having any type information. And for a packed struct, taking a copy seems like the more promising strategy in general, so that's what we do.
Noratrieb commentedon Apr 25, 2023
Maybe we could add a
PackedDeriveFieldhelper trait (with aT: Copyblanket impl) and addField: PackedDeriveFieldtrait bounds on the impls. Then, throw arustc_on_unimplementedon that trait and hope the diagnostics machinery picks it all up.Auto merge of rust-lang#117511 - gurry:117406-err-packed-structs, r=c…