Introduce liballoc::actual_usable_size

[nox]

In Servo we use heapsize to be able to report about heap memory usage. It would be way more robust if heapsize's function heap_size_of were in liballoc itself.

Probably that function should be parameterised by T instead of taking a *mut c_void, to handle alignment without relying on HeapValidate on Windows.

[SimonSapin]

CC @pnkfelix

[pnkfelix]

... I don't know if we can reliably provide this for all potential allocators ...

Would you be willing to make the API more flexible, e.g. have it return Option<usize>, with a best-effort expectation?

[SimonSapin]

je_malloc_usable_size only takes a pointer and somehow that’s enough for jemalloc to return a meaningful value. But we don’t need all allocators to be that magical, we can provide them with more information:

// In the alloc::heap module
pub unsafe fn usable_size(ptr: *mut u8, size: usize, align: usize) -> usize { ... }

Namely, the same information we need to keep track of anyway for alloc::heap::deallocate. Simply returning the size argument would be a valid implementation of usable_size.

[retep998]

Since the size returned by HeapSize on Windows is always exactly the size you asked it to allocate, without any rounding up or extra space, simply returning the size argument would be just as accurate on Windows.

[pnkfelix]

Okay I'm fine with this as long as it meets the same input requirements as deallocate

[nox]

Since the size returned by HeapSize on Windows is always exactly the size you asked it to allocate, without any rounding up or extra space, simply returning the size argument would be just as accurate on Windows.

@retep998 You mean return size + align, right?

[nox]

The point of this function isn't to return how much space can be used in a give pointer, but about how much space did I need to actually allocate.

[retep998]

@nox Well in the case of overaligned types it is size + align. if align <= MIN_ALIGN { size } else { size + align }.

[nox]

Yeah ok, just checking that we were talking about the same thing. :)

[SimonSapin]

The point of this function isn't to return how much space can be used in a give pointer

FWIW early versions of https://github.com/servo/tendril did use je_malloc_usable_size that way. (I removed that when porting it to stable Rust.)

[nox]

What can we do to get some progress about this?