Make vim bindings opt-in

[jonas-schulze]

as an alternative to #41799.

If accepted, the second commit should be backported to v1.7.

[jonas-schulze]

This should be good now. #41834 (comment) remains relevant, though.

[timholy]

If you want to fix it this way, we'll have to backport #41576 as well. The safest option is to revert from 1.7 and do all this in 1.8. Thoughts?

[vchuravy]

If you want to fix it this way, we'll have to backport #41576 as well. The safest option is to revert from 1.7 and do all this in 1.8. Thoughts?

I am in favor of that.

[jonas-schulze]

If you want to fix it this way, we'll have to backport #41576 as well. The safest option is to revert from 1.7 and do all this in 1.8. Thoughts?

There is no urge, I was just hoping to get that feature in half a year earlier (or what else the release schedule is). 🙃

Regarding the backport: I would just disallow vim bindings for RadioMenus for 1.7 (i.e. not definining any of accepts_*(::RadioMenu)) such that the keybindings won't "become breaking" in 1.8.

[timholy]

Personally I think we should wait until Julia 1.8. The issue is this: this is an extension API. Whatever we pick has to be something we're happy supporting for the long term. This design seems reasonable, but it adds three additional methods that need to be defined to support a very simple feature, and it would be package-specific rather than user-specific. There might not be a better way to do this, but are we sure?

I would argue that the stabilization process of 1.7 is not the right time to make that decision; better to "do no harm" than to rush something.

[KristofferC]

+1 for revert and get some extra time to think about it. The three extra methods also stuck out to me and I think it deserves a bit more thought.

[jonas-schulze]

Ok. So should #41835 be merged into master (such that vim bindings have to be redone from scratch) or just into release-1.7 (such that this PR can remain as it is)?

[vchuravy]

Ok. So should #41835 be merged into master (such that vim bindings have to be redone from scratch) or just into release-1.7 (such that this PR can remain as it is)?

You can rebase this PR on a revert of #41835

[JeffBezanson]

From triage: seems like this should not be on the 1.7 milestone? Let's plan for it in 1.8.

[timholy]

@jonas-schulze, if you do rewrite this for 1.8, just couple of thoughts:

• One could check for additional configuration via hasfield in the Config object. You might be able to declare a navigation_keys field.
• The hardest part to deal with well is the custom keypress function. In a world where we can configure the key bindings for the user, hard-coded choices seem non-ideal.

Perhaps you could mimic my own strategy in rewriting the old API, and declare a further subtype of _ConfiguredMenu{C} that becomes the model of what we'd change the interface to when we can break it.

[ViralBShah]

Should plan for 1.9 now if we want to.