implement Deref trait for SecretKey

[toxeus]

The Deref trait is implemented for SharedSecret and SerializedSignature. I was wondering if you have considered implementing that trait also for PublicKey and SecretKey already?

[apoelstra]

What would they Deref to? The internal representation is opaque and subject to change.

[toxeus]

I thought they would Deref to &[u8].

[toxeus]

SecretKey already allows access to the internal representation https://docs.rs/secp256k1/0.19.0/secp256k1/key/struct.SecretKey.html#method.as_ref. I think implementing Deref would be useful and consistent with the types mentioned above.

[elichai]

A. The internal representation of PublicKey is Implementation Defined, so to impl Deref you'd have to serialize it each time.
B. I actually think we should decrease things magic implementations like Deref, not increase, because they give a bunch of functions that don't really makes sense for SecretKey (on the other hand I do love abusing it sometimes because you magically get a bunch of functions for free)

[toxeus]

I agree with A. Let me adjust the issue title to reflect that.

About B: Deref is widely used in Rust and part of the language. Personally, I like it. I can understand that some perceive it as "magic" but this can be argued about many language features really and if you strictly go down that path you end up with golang ;)

Still, the inconsistency between SharedSecret and SerializedSignature on one hand and SecretKey on the other still exists. I'd like to understand when "magic" is warranted and when not.

[elichai]

About B: Deref is widely used in Rust and part of the language. Personally, I like it. I can understand that some perceive it as "magic" but this can be argued about many language features really and if you strictly go down that path you end up with golang

Don't get me started on "magic" in golang :) (comments are magic, the amount of return values from maps/channels is magically chosen by how many lvalues you have, cgo is 100% pure magic, which pointers the GC tracks is magic, should I continue? :) )

[apoelstra]

I think Deref in Rust is uniquely magical :)

[apoelstra]

In any case -- I think we should probably go in the direction of removing access to the internal representation (or providing a convert_to_ffi method in its stead). The internal representation is almost certainly not what the user wants unless they are doing magic FFI stuff (and then I would argue that they should never have started using the non-FFI types to begin with..). See #248 for example of where people got burned by the internal representation not resembling an actual public key.