Phase1

[Lokathor]

TODO:

• ?

[MaulingMonkey]

What's the advantage of (a ...).as_ref().unwrap_or_else(...) over &*(a ...)?

[Lokathor]

Ah, right, none. I just literally forget that you can do that

[MaulingMonkey]

...and I guess Box casts are maybe probably sound due to A and B having compatible Layouts?

While all the unsafe frightens me and threatens to make me spontaniously combust, I'm not seeing any more soundness concerns 👍

[Lokathor]

@MaulingMonkey there was a request for zeroed boxes without using the stack, so i put that in.

[MaulingMonkey]

Reviewed new commits... approved pending fix for ZSTs in try_zeroed_box

[danielhenrymantilla]

What about returning Ok(Box::new(T::zeroed())) ?

[Lokathor]

That puts T::zeroed on the stack, which is exactly what the goal was to avoid.

You're the second person to ask that one. Do yall read the doc comments?

[danielhenrymantilla]

Sorry, I posted a global comment instead of answering to the thread about ZST in try_zeroed_box. In that case instead of erroring an Ok(Box::new(T::zeroed())) is acceptable since T is zero-sized so there is nothing in the stack.

[Lokathor]

ahh, hmm, that sounds like a good fix

[MaulingMonkey]

#[inline(never)] fn try_zeroed_box_on_stack<T>() -> Box<T> {
    Ok(Box::new(T::zeroed()))
}

Would help ensure it never gets inlined onto the stack for size_of::<T>() != 0... although I imagine the branch would get optimized out in release builds, so maybe just a regular fn would be fine?

[Lokathor]

If we only use that for zero sized types it doesn't matter if it is "on the stack" since the size is zero anyway

[MaulingMonkey]

Right, I'm just being paranoid about the case where non-executing code isn't fully eliminated (debug builds?), which can still contribute to stack size as I've found out before (last time I overflowed the stack was thanks to conditional logging macros, that had fixed length char buffers for formatting strings... apparently that's enough to blow the stack even when they're not executing....)

Or to put it a little more concretely, I'd worry:

if size_of::<T>() != 0 {
    Ok(Box::new(T::zeroed()))
} else {
    ...
}

Could still put size_of::<T> bytes on the stack when it's nonzero, despite the branch never executing.