Writing to uninitialized buffer about byteorder HOT 10 CLOSED

Exactly. I'd submit a PR, but I'm currently low on time. I expect to have more time after ~2 weeks.

from byteorder.

This is weird, I completely forgot about this. I do remember trying to add unsafe but something stopped me and then I didn't continue for some reason. Can't remember anything more. I guess it's a good thing because MaybeUninit was introduced since then and it doesn't make sense to guarantee anything about uninitialized &mut [u8] as that's UB already.

No change is actually required because performant writing can be done safely already: https://play.rust-lang.org/?version=stable&mode=release&edition=2018&gist=26babbb49afda7644f61f4e7ffe1ad2c

TL;DR is that writing to a safe, 0-init array first and then copying it into uninit slice gets optimized just fine.
It could be nice to guarantee that unsafe code can rely on values being written as expected but I personally wouldn't consider unsafe code being broken if it relies on it implicitly as that's a reasonable assumption.

Maybe a more appropriate label would be magically-fixed-itself instead of wontfix. 😆

from byteorder.

This seems reasonable to me. I can't imagine any circumstances under which the write methods would actually want to read from the buffer. Can you?

from byteorder.

I can't. Would it be breaking change to make the trait unsafe? If the only implementors are LittleEndian and BigEndian, then it should be fine...

from byteorder.

Ping?

from byteorder.

I don't think it would be a breaking change.

Why does the standard io::Read trait not make this guarantee?

from byteorder.

There's a PR addressing this.

from byteorder.

I see. So it seems like the preferred solution is to make the trait unsafe, but we couldn't do that for std::io::Read. However, I believe we can here since the trait is sealed.

from byteorder.

Sorry, I forgot about this completely. I'll look at it today (not sooner than 9 hours from now).

from byteorder.

I think that if you're already dealing with unsafe code and uninitialized buffers, then it probably doesn't make sense to use byteorder's safe APIs, since they impose a cost.

from byteorder.