Cannot implement custom History: ReedlineError is private

[ClementNerma]

Currently you can't implement your own History type as multiple trait functions return a Result<_, ReedlineError>, but ReedlineError (which comes from the private result module) is not available to use outside of the crate.

Would it be possible to make it available publicly?

Thanks!

[sholderbach]

Yes I think there is nothing speaking against that in my view. The trait should be implementation agnostic/tolerant.

While we are at it, it might make sense to make it either more general or replace it with a more specific name, as most of reedline just returns std::io::errors for the fallible interactions with the terminal. (Most of the reedline errors at the moment don't seem to actionable for the library users)

[ClementNerma]

I just opened a PR for this at #661

[sholderbach]

Looking at the current state of the history system, I think we still have some serious gremlins lurking there.

See the unfinished error handling around how we deal with the different capabilities of SqliteHistory and FileBackedHistory (.expect("TODO: error handling") doesn't sound like a long term solution). Also assumptions around how big the history minimally is seem to pop up as nushell/nushell#10826. We have some work to do here to make things more robust and easier to understand.

So I don't want to give any stability promises with regards to the API of the History trait and related parts.

I don't want to discourage you from experimenting with implementing your own History and reporting what pops up.
But keep in mind that we may be a moving target or prioritize robustness over total flexibility in the future (as some of the issues we see right now may be due to premature flexibility)

[ClementNerma]

I'm actually building a shell on top of reedline. But it's a personal project that isn't designed for public use before it's entirely finished and heavily tested, so even if there are major API breakages it's not a problem for me :)