miri doesn't catch a violation of `&mut [u8]`'s validity invariants

[anp]

I think miri should fail on https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=a79f547a85d6d09e52756c2da3c121f4 because it produces an invalid &mut [u8] on line 11 (see context for reference interpretation).

Should it be an error to produce a &mut [u8] that references uninit bytes?

[anp]

I missed this bit from the README: "In particular, Miri does currently not check that integers/floats are initialized or that references point to valid data."

[RalfJung]

For the int/float part, #1340 tracks adding an option to check that.

But for references, I don't currently have plans to implement that -- it would be prohibitively expensive. Moreover I am anyway lobbying for relaxing that UB rule. ;)

[anp]

Makes sense. I'm assuming it would still be prohibitive to treat slices differently than "thin" references.

I'd be interested to see a path for relaxing that rule, the ergonomics of single-write-only buffer usage is pretty rough. I have no idea what that'll look like, but you've got an eager audience :D.

[RalfJung]

I'd be interested to see a path for relaxing that rule,

The discussion is at rust-lang/unsafe-code-guidelines#77. ;)
To summarize, I think we should change the description of invalid references from

A reference or Box that is dangling, unaligned, or points to an invalid value.

to

A reference or Box that is dangling, unaligned, or points to an uninhabited type (i.e., a type with no valid value).