Add an into vec implementation?

[Sushisource]

Hi there. Like Prost a lot. I find that by far the most common case for me when encoding is

1. Make a new empty vector
2. Immediately encode into it.

I absolutely understand why the encode function works the way it does, but given how common the case I just mentioned is, seems worth making this even easier on the caller. Namely:

let encoded: Vec<u8> = message.into()

This would be super. If you like the idea, I'm happy to contribute a PR (admittedly, not a whole lot to do here 😄 )

Thanks!

[rubdos]

Hey! Just jumping in here. Into is typically used for cheap, infallible conversions. Allocating does not (in my opinion) fall in that category. However, I found it tiresome too to write Vec::with_capacity(encoded_len()) the whole day, so I made #378: encode_to_vec() and encode_length_delimited_to_vec().

I think that will solve your underlying issue well, so I've marked this issue as closed by that PR. If you disagree, let me know, I'll unmark it.

[Sushisource]

@rubdos Awesome! That's perfect, and great point about this maybe not being the best for Into.

[wchargin]

In addition to admitting the with_capacity optimization, such a
dedicated encode_to_vec method can return Vec<u8> rather than
Result<Vec<u8>>, which saves callers from having to expect with a
comment indicating why this is always safe.

I see that the implementation in #378 does so, so imho this is another
good reason to support this feature:

• more concise
• expresses programmer intent better
• works in an expression context
• more efficient due to with_capacity
• more precise return type