Tracking issue for comparing raw pointers in constants

[oli-obk]

Comparing raw pointers in constants is forbidden (hard error) in const contexts.
The const_compare_raw_pointers feature gate enables the guaranteed_eq and guaranteed_ne methods on raw pointers. The root problem with pointer comparisons at compile time is that in many cases we can't know for sure whether two pointers are equal or inequal. While it's fairly obvious that a pointer to a static and a local variable won't be equal, we can't ever tell whether two local variables are equal (their memory may alias at runtime) or whether any pointer is equal to a specific integer. In order to make it obvious that there is a discrepancy between runtime and compile-time, the guaranteed_* methods only return true when we can be sure that two things are (in)equal. At runtime these methods just return the result of the actual pointer comparison.

This permits const fn to do performance optimization tricks like the one done in slice comparison (if length is equal and pointer is equal, don't compare the slice contents, just return that it's equal).

Since we aren't sure yet what the semantics of functions that return different values at runtime and at compile-time are, we'll keep this function unstable until we've had that discussion.

Unresolved questions

• should guaranteed_eq return an Option<bool> to allow the caller to know whether they got unclear results? What would the use-case for this be?

[RalfJung]

I either forgot this exists or was never aware.^^ But it is certainly scary... pointer equality is a really complicated subject.

But also, so far the feature flag doesn't actually let you do anything, does it? binary_ptr_op still always errors in the CTFE machine, and even unleashed Miri cannot do pointer comparison.

[oli-obk]

Yea it's a fairly useless feature gate as of now