Views on arrays should become views on a slice of the Memory that was accssible at view construction

[gbaraldi]

This has historically being a possible unchecked OOB access if a user made a vector and pushed/popped/sizehinted it while keeping a view into the array. Views do not do a double boundscheck (one on the view and one on the array), instead they boundscheck on their bounds and perform a check during view construction that all of their indices are valid. Before we had array this was an issue with no clean solution, we either do the double boundscheck or have what we have right now which is potential OOB. With Memory we can have the view ignore the array and turn that OOB/UB into defined code, because Memory bounds nevert change. This has a caveat that it now means if someone pushes/pops that vector while keeping a view, the view may no longer alias the array.

[sgaure]

Wouldn't it still be possible with OOB?

v = rand(4)
ii = [1,2,3]
vv = view(v, ii)
push!(ii, 42)
vv[4]