[magic-akari]

No description provided.

[magic-akari]

It's hard to implement NewCipher for Rc4 since Rc4 can use variable key lengths.
What do you think about it? @tarcieri

[tarcieri]

@magic-akari you can make the NewCipher impl generic around a key size, and use a marker trait to determine the set of valid key sizes

See how AesGcm is similarly generic around a NonceSize as a roughly analogous example:

https://docs.rs/aes-gcm/latest/aes_gcm/struct.AesGcm.html

[magic-akari]

stream-ciphers/rc4/src/lib.rs

Lines 128 to 146 in 0b9ebe8

	#[test]
	fn test() {
	let mut rc4 = Rc4::new(b"Key".into(), &Default::default());
	let mut data = b"Plaintext".to_vec();
	rc4.try_apply_keystream(&mut data).unwrap();
	assert_eq!(data, [0xBB, 0xF3, 0x16, 0xE8, 0xD9, 0x40, 0xAF, 0x0A, 0xD3]);
	}
	#[test]
	fn test2() {
	use cipher::consts::U3;
	let mut rc4 = Rc4::<U3>::new_from_slices(b"Key", Default::default()).unwrap();
	let mut data = b"Plaintext".to_vec();
	rc4.try_apply_keystream(&mut data).unwrap();
	assert_eq!(data, [0xBB, 0xF3, 0x16, 0xE8, 0xD9, 0x40, 0xAF, 0x0A, 0xD3]);
	}

HI, @tarcieri ,
I tried to implement NewCipher, but it would be a pain for users to use.

User have to write Rc4::new(b"Key".into(), &Default::default()); or Rc4::<U3>::new_from_slices(b"Key", Default::default())

<U3> cannot be inferred and must be written manually.

While as a user, I prefer the following codes.
What do you think about it?

stream-ciphers/rc4/src/lib.rs

Lines 119 to 125 in 6321600

	fn test() {
	let mut rc4 = Rc4::new_from_slices(b"Key").unwrap();
	let mut data = b"Plaintext".to_vec();
	rc4.try_apply_keystream(&mut data).unwrap();
	assert_eq!(data, [0xBB, 0xF3, 0x16, 0xE8, 0xD9, 0x40, 0xAF, 0x0A, 0xD3]);
	}

[tarcieri]

@magic-akari the root problem there is really generic-array, and yes, its ergonomics are bad. Long term we'd like to get rid of it, and in the medium term perhaps we can replace it with RustCrypto/utils#707

However, I think it would still make sense to impl NewCipher.

Users should define a type alias with the expected key size, or the crate can define some aliases for common key sizes:

pub type Rc4_128 = Rc4<U16>;

To me this is preferable to trying to infer the key size from the parameter, since unless it's constrained someone might accidentally pass the wrong/unexpected key size for a given application without noticing.

[newpavlov]

Can you please update this crate to cipher v0.4? I will try to review it next week.

[magic-akari]

HI, @newpavlov
I tried to update it. I have to say it's hard to implement StreamCipherCore as it require a BlockSize.
Do you have any idea about it?
What BlockSize should Rc4 use?

[newpavlov]

Well, strictly speaking RC4 has block size of one byte. Implementation of StreamBackend::gen_ks_block would look rougly like this:

fn gen_ks_block(&mut self, block: &mut Block<Self>) {
    // `block` has type of `GenericArray<u8, U1>`
    block[0] = self.state.prga();
}

Alternatively you could implement the StreamCipher trait directly.

[magic-akari]

@newpavlov can you review it?

[tarcieri]

I can do a first pass this weekend

[tarcieri]

Looks good, thank you!

I'm going to do some minor followup cleanups after merging.

[tarcieri]

See #299 for fixups