Brought up in #154004 (comment): Should the method on Alignment that gives a NonNull<usize> be spelled as_nonzero, as_nonzero_usize, as_non_zero_usize, get, or what?
Nominating for libs-api thoughts so someone can make a PR for whatever the answer is.
Existing APIs on the type: https://doc.rust-lang.org/1.94.0/std/ptr/struct.Alignment.html
Other things like https://doc.rust-lang.org/1.94.0/std/ptr/struct.NonNull.html#method.as_non_null_ptr use non_null for NonNull, so non_zero would arguably be consistent with that. copy_nonoverlapping was mentioned as well, though there's no NonOverlapping type so the parallel isn't as close. Apparently I'm consistent in what I write, because I made https://doc.rust-lang.org/nightly/std/intrinsics/fn.ctlz_nonzero.html too.
Alternatively, this could be more like NonZero itself and just have new/get, since .get().get() would exist to go Alignment -> NonZero<usize> -> usize. Going safely & infallibly that way is important. Going the other way of course is fallible or unsafe so the types involved are in some sense less important. new_unchecked's precondition is is_power_of_two whether it's on usize or NonZero<usize>, though on the latter that's slightly more efficient of a check (for new), though I suspect the combination optimizes together nicely either way.
Brought up in #154004 (comment): Should the method on
Alignmentthat gives aNonNull<usize>be spelledas_nonzero,as_nonzero_usize,as_non_zero_usize,get, or what?Nominating for libs-api thoughts so someone can make a PR for whatever the answer is.
Existing APIs on the type: https://doc.rust-lang.org/1.94.0/std/ptr/struct.Alignment.html
Other things like https://doc.rust-lang.org/1.94.0/std/ptr/struct.NonNull.html#method.as_non_null_ptr use
non_nullforNonNull, sonon_zerowould arguably be consistent with that.copy_nonoverlappingwas mentioned as well, though there's noNonOverlappingtype so the parallel isn't as close. Apparently I'm consistent in what I write, because I made https://doc.rust-lang.org/nightly/std/intrinsics/fn.ctlz_nonzero.html too.Alternatively, this could be more like
NonZeroitself and just havenew/get, since.get().get()would exist to goAlignment -> NonZero<usize> -> usize. Going safely & infallibly that way is important. Going the other way of course is fallible or unsafe so the types involved are in some sense less important.new_unchecked's precondition isis_power_of_twowhether it's onusizeorNonZero<usize>, though on the latter that's slightly more efficient of a check (fornew), though I suspect the combination optimizes together nicely either way.