Implement BufMut for std::collections::VecDeque? about bytes HOT 4 CLOSED

I'm not entirely sure what an implementation of BufMut would look like for VecDeque. That being said, both Cursor<T: AsRef<[u8]>> and Bytes / BytesMut support "draining" without a memmove.

from bytes.

Thanks for the answer.

Using a Cursor<T: AsRef<[u8]>> would not have the benefit of a ring buffer, which is that once part of the buffer is drained, it can be reused. For example, if you have a sequence like "append 2000 bytes, drain 2000 bytes, append 2000 bytes, drain 2000 bytes...", then a ring buffer can do that without allocating or memmoving, because it just wraps around.

Regarding how an implementation for VecDeque will look, I am not sure. But generally for a ring buffer, say in it's in a state like this:

# = free
x = used
        (region A)     (tail)    (region B)    (head)   (region C)
##########################xxxxxxxxxxxxxxxxxxxxxxxx####################

then:

• remaining_mut would return the size of region C + size of region A.
• bytes_mut would return region C, or region A if there is no region C.
• bytes_vec_mut would return region C, region A.
• advance_mut would advance head.

I assume it is not possible to implement this without relying on internal implementation details of VecDeque, which will never be guaranteed, unlike Vec. So this issue can probably be closed, and anyone who wants something like that would have to implement their own ring buffer. But thought I'd ask anyway.

from bytes.

Yes, I am familiar w/ the argument for a ring buf. There used to be an implementation of a ring buffer in crate, but it was removed a while ago.

I didn't realize that VecDeque provided as_slices which would make it plausible.

If you want to take a shot at implementing Buf / BufMut for VecDeque<u8>. I think there still may be some missing fns on VecDeque to impl it correctly though.

from bytes.

Closing due to inactivity. Please reopen or create a new issue. if the status changes.

from bytes.