Add a `into_parts()` method on `BufWriter`

[aryan9600]

Is your feature request related to a problem? Please describe.
It'd be helpful if there was a way to get the underlying buffer being used by a BufWriter instance. This could be helpful in scenarios where we want to do buffered writes while reusing the same buffer. The std library provides a way to do this using into_parts().

Describe the solution you'd like
Add a new method BufWriter:into_parts() which returns both the underlying stream and the buffer. This would return the buffer as is, i.e. any data thats sitting in the buffer is NOT flushed to the stream before the buffer is returned to the caller.

[aryan9600]

happy to create a PR if maintainers think it'd be okay to add this method

[Darksonn]

Hrm. Our current buffer does use Vec<u8>, but it also has a written counter for the number of bytes in the beginning that are already written. What should we do with that when returning a Vec<u8>? Move the data around to discard it? Return a different type from Vec<u8>?

[aryan9600]

is it okay to just discard it? from my understanding the written counter is just internal state to make sure that we dont write the same part of the buffer again (since poll_write can return Poll::Ready(n) with n being less than the length of the buffer passed into it).

[Darksonn]

Discarding the data is okay, but it may be an expensive operation if the unwritten buffered data is large.

[aryan9600]

yeah i understand that, but i dont think there's any other option. the thing is when using a BufWriter to process events in a loop can become expensive (especially when its in the hot path). it'd be nice if we could perform the following steps in a loop:

• construct a BufWriter using an existing buffer via a constructor like with_buf()
• perform writes
• flush the data into the socket
• call into_parts() to get the stream and buffer back

is such an API acceptable or does this make things too complicated?