std: add pointer stability locks to array list

[emar-kar]

Closes #19326

[vortexofdoom]

clone shouldn't require pointers to be unlocked, right? The only allocations are for the newly cloned ArrayList, and the original allocation is unchanged.

[emar-kar]

@vortexofdoom Do we consider async here? if we do, then if I try to clone it while it's locked by the update operation, it can corrupt the result. Or am I missing something?

[emar-kar]

And I was thinking if appendSliceAssumeCapacity should lock the array as well?

[vortexofdoom]

Interesting point. I guess my gut instinct would be to avoid making things that should be trivially possible in non-async settings impossible for the sake of async. In an async setting you can manually lock it before a clone.

Appending, even assuming capacity, could "invalidate" a pointer to the last element or similar, so that seems maybe less objectionable, but I could see the argument for removing it.

Although I guess the argument exists that you won't be locking much in a non-async context, so it may be a non-issue.

[emar-kar]

I see the point for removing assertion from clone. And actually I do support it. It's easy to bring it back in case

Anyway, the question about appendSliceAssumeCapacity is still open for discussion.

[vortexofdoom]

Looking at it, I think appendSliceAssumeCapacity is also better without. There's already an assertion that the items can fit, it's supposed to be the utility function for actually doing the thing under the hood, I think if it requires being unlocked suddenly there's a case for making an even deeper function that doesn't require it being unlocked.

The buck has to stop somewhere, and I think it's fine to stop short of functions with assume in the name.