Repository navigation
New compiler error in beta: this constant cannot be used, attempted to read undefined bytes #54387
Description
Since latest beta (and nightly), some of my crates have errors like the following
error: this constant cannot be used
--> src/udpsrc.rs:47:1
|
47 | const DEFAULT_CAPS: Option<gst::Caps> = None;
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ attempted to read undefined bytes
A standalone testcase replicating all the involved types can be found below (depends only on the libc crate).
I don't know if the error is valid, but in any case I don't understand what it means so at least the error message could be improved a bit :)
use std::marker;
use std::ptr;
extern crate libc;
use libc::{c_int, c_uint, c_void, size_t};
#[repr(C)]
pub struct CapsRef(ffi::GstCaps);
pub type Caps = GstRc<CapsRef>;
pub trait MiniObject {}
impl MiniObject for CapsRef {}
pub struct GstRc<T: MiniObject> {
obj: ptr::NonNull<T>,
borrowed: bool,
phantom: marker::PhantomData<T>,
}
mod ffi {
use super::*;
#[repr(C)]
#[derive(Copy, Clone)]
pub struct GstCaps {
pub mini_object: GstMiniObject,
}
#[repr(C)]
#[derive(Copy, Clone)]
pub struct GstMiniObject {
pub type_: GType,
pub refcount: c_int,
pub lockstate: c_int,
pub flags: c_uint,
pub copy: GstMiniObjectCopyFunction,
pub dispose: GstMiniObjectDisposeFunction,
pub free: GstMiniObjectFreeFunction,
pub n_qdata: c_uint,
pub qdata: gpointer,
}
pub type GstMiniObjectCopyFunction =
Option<unsafe extern "C" fn(*const GstMiniObject) -> *mut GstMiniObject>;
pub type GstMiniObjectDisposeFunction =
Option<unsafe extern "C" fn(*mut GstMiniObject) -> gboolean>;
pub type GstMiniObjectFreeFunction = Option<unsafe extern "C" fn(*mut GstMiniObject)>;
pub type gboolean = c_int;
pub type GType = size_t;
pub type gpointer = *mut c_void;
}
const FOO: Option<Caps> = None;
fn main() {
let _meh = FOO;
}
Activity
csmoe commentedon Sep 20, 2018
the testcase compiles with
stable-1.29.0.hanna-kruppe commentedon Sep 21, 2018
Minimized:
pnkfelix commentedon Sep 27, 2018
Visited during T-compiler meeting. Tentatively assigning to @eddyb for investigation.
eddyb commentedon Sep 27, 2018
cc @RalfJung @oli-obk I can't find anything in the validation code that's reading the wrong thing - it should be able to tell this is a
None, which has0fields.AstralSorceress commentedon Sep 27, 2018
I've been looking into this. It seems that the issue stems from here. Specifically, because the niche-filling enum representation is used for
Option<GstRc>and similar types, and the niche that gets filled is the second field (thebool), it is actually the second part of theScalarPairthat is defined, not the first.eddyb commentedon Sep 28, 2018
Ahh, you're right! I don't know why only the second component was special-cased.
RalfJung commentedon Sep 29, 2018
Seems like @oli-obk's quest to keep
Undefcontained within the CTFE engine failed at last. :/However, in this case nothing should even be trying to convert to a
ConstValue. So I am not sure if that's really the problem.RalfJung commentedon Sep 29, 2018
Actually it looks like this is not a
ScalarPair, but aScalarbeing undef. At least that's what the line numbers are saying. It does not make much sense though.EDIT: It is definitely a
ScalarPair, just the line numbers seem to be wrong.RalfJung commentedon Sep 30, 2018
I am working on a fix.
EDIT: Submitted as #54693
Auto merge of #54693 - RalfJung:ctfe-scalar-pair-undef, r=oli-obk
eddyb commentedon Oct 2, 2018
Should this be closed before we get a beta backport?
emilyalbini commentedon Oct 9, 2018
Backported to beta. Closing this.