doc: Keypair constructors #779
Conversation
Clarify the conditions for returning the InvalidSecretKey error. Fixes rust-bitcoin#758
The constructor of a secret key should return InvalidSecretKey in case of error. I guess there was a typo / copy-and-paste error that was never tested.
| @@ -891,7 +891,7 @@ impl Keypair { | |||
| match from_hex(s, &mut res) { | |||
| Ok(constants::SECRET_KEY_SIZE) => | |||
| Keypair::from_seckey_slice(secp, &res[0..constants::SECRET_KEY_SIZE]), | |||
| _ => Err(Error::InvalidPublicKey), | |||
| _ => Err(Error::InvalidSecretKey), | |||
There was a problem hiding this comment.
Oh yes, the entire error type needs a lot of work...
There was a problem hiding this comment.
I think we should move deserialization/serialization into Rust and then we can provide sensible error types. Obviously in this case the variant was totally wrong, but in general the error types in this library are crappy because they're just thin wrappers around the numeric error returns from the C code.
There was a problem hiding this comment.
...lol, and half this API is "take a &[u8] then length-check it" and should be replaced by an array-based on.
There was a problem hiding this comment.
Ye, I think we need to do all checks except "is on the curve"
There was a problem hiding this comment.
I remember that there was an initiative to replace slice-based methods with array-based ones. IDK if this was Miniscript or secp. In any case, I'm happy to write a follow-up PR to move the constructors to array parameters.
There was a problem hiding this comment.
@uncomputable that would be awesome.
If you want a fun weekend project you can also try to pull some parts out of #703 to replace point deserialization code with Rust. (That PR is out of sync and also IMO it does too much at once; we should practice Rust-ifying a couple functions and defining sensible APIs and error types.)
Clarify the conditions when
Keypairconstruction fails and correct the error variant that is returned in one of the constructors. Fixes #758