--- Log opened Sun Mar 13 00:00:26 2022
12:22 < sanket1729> andytoshi: one API question about finalizepsbt. The current behaviour is that if finalization fails at psbt input i, all inputs till i-1 are finalized. Is this behavior okay? Should we change the psbt finalize to take &self and return a new Psbt struct
12:24 < sanket1729> It seems wasteful to create a new psbt when we only want to set new two fields and remove a bunch of fields.
12:26 < andytoshi> sanket1729: i think that all inputs should be finalized except the ones that fail
12:26 < andytoshi> so everything from i-1, and everything from i+1 onward
12:27 < sanket1729> How about the return err then?
12:27 < sanket1729> Should it Vec<Error>?
12:27 < sanket1729> for all inputs that fail?
12:27 < andytoshi> yeah, i think so
12:27 < andytoshi> and it should be a tuple rather than a Result
12:27 < sanket1729> What should be tuple?
12:27 < sanket1729> The return type
12:28 < sanket1729> Makes sense
12:29 < sanket1729> I think it might be better idea to drop the `finalize` and only expose `finalize_input`. Have the users loop over them
12:29 < sanket1729> So, we don't have to deal with these cases. And the user can cleanly decide based on what it needs. 
12:29 < andytoshi> i wouldn't mind having both
12:30 < andytoshi> `finalize` would be a pretty short function to write
12:30 < andytoshi> it just does the loop and collects the errors
12:30 < sanket1729> Fair point.
12:30 < andytoshi> we'd document that users who want more precise control need to call `finalize_input` themselves
12:31 < sanket1729> I like it
12:32 < andytoshi> dope
12:39 < sanket1729> andytoshi: I think the return type would still be Result<(), Vec<Error>>. What is tuple return type that you are suggesting?
12:43 < andytoshi> sanket1729: oops, i was thinking (Psbt, Vec<Error>)
12:43 < andytoshi> but actually we don't want to return the original psbt
12:43 < andytoshi> or any psbt
12:44 < andytoshi> so yeah, Result<(), Vec<Error>> sounds good to me
12:44 < andytoshi> I'd also be ok wiht Vec<Error> and the user has to check if the vector is empty
12:46 < sanket1729> I have a slight preference to Result<(), Vec<Error>> instead. Returning a result seems more idiomatic
12:48 < andytoshi> sounds good to me
16:59 < jeremyrubin> my preference sanket1729 is to do: Result<Psbt, (Psbt, Vec<(usize, InputError)>)>
16:59 < jeremyrubin> that way the API you can write is something like:
16:59 < jeremyrubin> let finished = psbt.finalize()?;
17:00 < jeremyrubin> and if you *do* get a modified PSBT, you can work with the error'd one returned as Err((Psbt, errors)) and you can see which ones actually errored in the vec w/ the index
17:00 < jeremyrubin> altho maybe InputError has an index field already
17:01 < jeremyrubin> if finalize takes psbt instead of &mut psbt i think it's more clear that either you're OK with your psbt getting mut-d, or you have to clone it and then can combine back (if passed in from somewhere else)
17:02 < jeremyrubin> but the returning of the PSBT as a part of the result really helps with making the error handling code cleaner (since you either get an Err((psbt, erros)) or Ok(psbt) v.s. some error you need to check elsewhere
--- Log closed Mon Mar 14 00:00:26 2022