Probability tables for autocompletion

[jamshark70]

Currently, autocompletion options are arranged alphabetically. Sometimes, this means that frequently-used method selectors appear late in the list, after rare ones.

For example, type Array.fi -- probably you meant Array.fill, right? But fill is second in the list. fib is first. (fib? Did I ever use that method in 15 years?)

fill ctrl-u shows hundreds of uses of that keyword in the class library, while fib presents only one. It might be a reasonable way to build a probability table, at least initially: count references in the class library.

I haven't thought carefully about this. There might be drawbacks.

Future, non-urgent enhancement. Probably okay to file this under the IDE long-term project.

EDIT: Another example -- isN initially completes to isNaN rather than isNil. I could keep adding these... 🤣

[Sciss]

As a side note, the Firefox algorithm for address bar completion is incredibly good; it builds up this probability type of table by itself; for SC you would probably "ship" it with some pre-trained table that could then adapt (in the ideal implementation) to the user's look-ups. Interesting topic.

[scztt]

One thing to think about: once you’re beyond the first couple “most common” items in a large list, the rest of the list will be in random-ish order and won’t be easy to navigate/search visually. Beyond that, it has the potential to change each use, which is pretty disruptive when you’re accessing repeatedly. A good design might be one or two “most common”, followed by a separator plus a list with predictable order, e.g.:

__________________
Most common #1
Most common #2
__________________
A:method
B:method
Y:method
Z:method
__________________

Something like this would be a HUGE workflow improvement.

[telephon]

And you may want to use tab to autocomplete the word until the next choice. Like typing SinO + tab, you get SinOsc and the choices SinOsc and SinOscFB.