Witness files should not require explicit type annotations

[apoelstra]

When we attach witnesses to programs we know the type that we're expecting and don't need the user to hint it. (At least, not since #160 which brought in a much-improved Value API.)

Probably still useful to allow hints, which can catch errors and help guide the compiler in more complex cases where it needs to shrink types or something. But in the majority of cases witnesses are things like fixed-size public keys or signatures where the type annotation doesn't provide any more information than the length of the hex string.

[apoelstra]

@roconnor-blockstream points out that in the case of "pruned values" we can require the user to provide explicit unit values, and have the witness parser always attempt to parse "the full type" except maybe that units can be put in place of anything.

In that case I don't think type annotations would ever be necessary except as sanity checks.

[uncomputable]

Witness files have type annotations to ensure that the file is unambiguous. Are these sanity checks not desirable?

[apoelstra]

They're fine if they're optional.

I'm not sure what you mean by "unambiguous". The rust-simplicity code knows exactly what type it wants.

[uncomputable]

I meant the witness file is unambiguous on its own. But I agree: Witness files are kept together with the program files, so their types can always be resolved.

[apoelstra]

To be clear -- I would like to keep these annotations in generated .wit files and make sure they always work. They just shouldn't be required since they're a bit of a PITA for people hand-editing the files.