Miri checks ptr.read() but not dereferencing a pointer (*ptr)

Hi,
miri complains on this:

fn main() {
    unsafe {
        let a = [1u8,2,3];
        let b = &a[0] as *const u8;
        println!("{}", b.add(1).read());
    }
}

https://play.rust-lang.org/?gist=5af194a447ac5e0d38796c809e3a0e6f
but not on this:

fn main() {
    unsafe {
        let a = [1u8,2,3];
        let b = &a[0] as *const u8;
        println!("{}", *b.add(1));
    }
}

https://play.rust-lang.org/?gist=6b7d13c8b77d8e14e06d9133a19e992f

I just tried on the playground, and I do get an error for the second example as well:

error: Undefined Behavior: trying to reborrow for SharedReadOnly, but parent tag <untagged> does not have an appropriate item in the borrow stack
 --> src/main.rs:5:24
  |
5 |         println!("{}", *b.add(1));
  |                        ^^^^^^^^^ trying to reborrow for SharedReadOnly, but parent tag <untagged> does not have an appropriate item in the borrow stack
  |
  = help: this indicates a potential bug in the program: it performed an invalid operation, but the rules it violated are still experimental
  = help: see https://github.com/rust-lang/unsafe-code-guidelines/blob/master/wip/stacked-borrows.md for further information

Ha, you're right, this is weird.
This on the other hand doesn't give an error:

fn main() {
    unsafe {
        let a = [1u8, 2, 3];
        let b = &a[0] as *const u8;
        if !a.is_empty() {
            println!("{}", *b.add(1));
        }
    }
}

But neither does this:

fn main() {
    unsafe {
        let a = [1u8, 2, 3];
        let b = &a[0] as *const u8;
        if !a.is_empty() {
            println!("{}", b.add(1).read());
        }
    }
}

This has nothing to do with deref vs read. This is caused by is_empty creating a raw pointer and Stacked Borrows not tracking raw pointers in detail -- so it "confuses" the raw pointer created there with b. That is currently expected behavior, though I would like to do more precise tracking of raw pointers eventually (but this might completely kill performance... and we'll have to see how compatible it is with real code).

@RalfJung

@RalfJung Yes. this is a completely different issue than what I was describing,
what happened is that I had a complex function, saw that it didn't error, I removed a lot of code, replaced with read() and it did, so I wrongly assumed it was the read() change. sorry.

So together with my confusion and the fact that this is expected behavior I'll close this issue.
if you're curious this was the code snippet that got me here: rust-lang/rust#75943 (trying to minimize that down in the playground got me to accidentally assume deref was the problem)

I opened rust-lang/unsafe-code-guidelines#248 to track the underlying problem.

Rollup merge of rust-lang#75943 - elichai:2020-align_offset-docs, r=R…

Navigation Menu

Repository navigation

Miri checks `ptr.read()` but not dereferencing a pointer (*ptr) #1526

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Miri checks ptr.read() but not dereferencing a pointer (*ptr) #1526

Description

Activity

RalfJung commented on Aug 26, 2020

elichai commented on Aug 26, 2020

RalfJung commented on Aug 26, 2020

RalfJung commented on Aug 26, 2020

elichai commented on Aug 26, 2020

elichai commented on Aug 26, 2020

RalfJung commented on Aug 26, 2020

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Miri checks `ptr.read()` but not dereferencing a pointer (*ptr) #1526