How to decode an extrinsic in the client?

[nazar-pc]

Hey folks, what is the idiomatic way to decode a particular extrinsic out of OpaqueExtrinsic in the client?

Currently I have a few nested data structures with indexes matching those generated by proc macros and strategic offset of 3 bytes from the beginning of OpaqueExtrinsic's encoding (that I can't really explain, some of it comes from Vec<T>, but the rest - no idea), but it is extremely ugly, there must be a better way.

Maybe I could add some method to runtime API that will be able to take OpaqueExtrinsic as input and return Option<T> if decoding matches what I want it to be? All examples I found in Substrate repo do the opposite.

[Voxelot]

The most straightforward way is to deserialize as an UncheckedExtrinsic that you've configured for your runtime. For example, you could add a deserialize_opaque_extrinsic fn to your main runtime/lib.rs.

use parity_scale_codec::{Decode, Encode};
use sp_runtime::generic;

pub type Address = sp_runtime::MultiAddress<AccountId, ()>;
// Where Call comes from `construct_runtime!`
pub type UncheckedExtrinsic = generic::UncheckedExtrinsic<Address, Call, Signature, SignedExtra>;

pub fn deserialize_opaque_extrinsic(opaque_ext: OpaqueExtrinsic) -> Result<UncheckedExtrinsic, parity_scale_codec::Error> {
    let bytes = some_opaque_extrinsic.encode();
    UncheckedExtrinsic::decode(&mut bytes.as_slice())
}

Once you have an UncheckedExtrinsic, you can then match on the function to figure out which pallet call it was or inspect the signature to find the public key of the signer etc...

[nazar-pc]

Hm, that should work indeed, don't know why I didn't think about it. Thanks!

[bkchr]

Decoding an extrinsic from the client is a bad idea, because there could be a runtime upgrade and then you are not being able to decode the extrinsic anymore. The better solution is to use a runtime api to call into the runtime to decode the extrinsic you are interested in and return fixed types.

[nazar-pc]

Yes, that is exactly what I ended up doing, thanks!