Add fuzzing setup for Message::decode

[wiktor-k]

See fuzz/README.md for details on how to run it.

[wiktor-k]

@kpcyrd Sorry for disturbing you but if you don't mind could you look at this fuzzer setup and see if anything is obviously broken?

Thanks in advance! 🙇 and have a nice day! 👋

[wiktor-k]

Thank you very much for your suggestions @kpcyrd 🙏

[kpcyrd]

You're welcome :)

The way the data-slice-magic works - a slice looks like this in memory:

| pointer | length |

but since the slice itself is also immutable, you can only read from both fields.

However, you can create a new slice with identical pointer and length like so (this is a very cheap operation):

&data[..]

This slice you then "own" (kinda), you still can't change the data it points to, but you can change the pointer and length.

| D A T A F O O B A R A S D F |
  ^
  | pointer | 14 |

So if you read 5 bytes from your slice, the pointer would be incremented by 5, and the length decremented by 5. The slice acts as a read cursor essentially. :)

| O O B A R A S D F |
  ^
  | pointer | 9 |

The &mut is needed because it's expected by Message::decode, if it would take something like decode<R: Read>(mut reader: R) -> Result<Self> you could drop the extra &mut too. I'm not sure if that's a worthwhile idea though. Reader is implemented for impl Reader for Decoder<'_> and I'm not sure if the decode function would then insist on taking ownership of Decoder.

The patch looks good to me now. 👍

[wiktor-k]

However, you can create a new slice with identical pointer and length like so (this is a very cheap operation):

Ha, right. I was aware of the in-memory layout of fat pointers but didn't connect that &data[..] creates a new one 🤦‍♂️

The &mut is needed because it's expected by Message::decode, if it would take something like decode<R: Read>(mut reader: R) -> Result you could drop the extra &mut too. I'm not sure if that's a worthwhile idea though. Reader is implemented for impl Reader for Decoder<'_> and I'm not sure if the decode function would then insist on taking ownership of Decoder.

Alright. (Un)fortunately I don't control the Decode trait (since it comes from the RustCrypto's crate) and I've come with at different design for my other SSH crate that uses look-ahead parser (similar to programming languages).

The patch looks good to me now. 👍

Thanks a lot for your time and explanations. I really appreciate them! 🙇

I'll let the PR sit for a couple of days in case the other maintainers feel like commenting and if not merge it.

👋

[kpcyrd]

Sure, you're welcome! :)