Is there a faster way to read large Vec<u8>?

[MeguminSama]

At the moment, say I have a struct like this:

#[derive(Debug, DekuRead, DekuWrite)]
#[deku(ctx = "size: usize", ctx_default = "0")]
pub struct Rfc {
	#[deku(bytes_read = "size")]
	pub data: Vec<u8>,
}

The problem, is that deku seems to loop through the reader for each u8 in bytes_read. This causes it to be very slow on large vectors. #[deku(read_all)] and #[deku(count = "size")] are also very slow.

At the moment, we're using our own read function to do something like this:

match reader.read_bytes(size, &mut buf) {
	Ok(ReaderRet::Bytes) => Ok(Rfc { data: buf }),
	_ => {...}
}

Which is significantly faster.

But I was wondering if there was a built-in way to do this with deku, instead of deku looping over each u8?

If this isn't a feature currently, I might consider implementing it if it's something you'd want in deku.

Thanks!

[wcampbell0x2a]

For read_all performance, check out the following MR. #441

Since deku makes small repeated reads, using a https://doc.rust-lang.org/std/io/struct.BufReader.html should reduce the read overhead.

[MeguminSama]

Thanks for getting back so quickly!

At the moment, our reader is already using a BufReader. We tried doing this in an attempt to speed it up, but unfortunately it's still much too slow when reading the vectors compared to our own read function.

Would some kind of #[deku(read_buffer = "size")] attribute be something you'd consider? Or is this out of scope for deku?

[wcampbell0x2a]

Definitely try out the merge request, it's really slow without that.

Would some kind of #[deku(read_buffer = "size")] attribute be something you'd consider? Or is this out of scope for deku?

Sure, I don't have the code in front of me, but I think we only store leftover as a u8, so we would need to store the leftovers in a Vec<u8> if needed. I'd like to use that only if you use read_buffer, since for embedded platforms you don't want allocations all the time.

[wcampbell0x2a]

I also don't know, it could be an improvement in our impl of Vec, I think it reads and evaluates one at a time currently.

[MeguminSama]

I will take a look, thanks :)

[wcampbell0x2a]

@MeguminSama see the latest commits on master, they should solve most or all of your performance problems