Memoize result for a given input

[Ambrevar]

A great optimization, I believe, would be to memoize the source update results depending on the input.

Example:

• Type "foo", takes 10s to compute.
• Type "bar", takes 15s to compute.
• Type "foo" again.

With the current prompter, the last "foo" would take another 10s. With memoization, it would be instantaneous.

In terms of memory, I don't believe it's a big deal since prompters are short-lived objects.

[aartaka]

This works for most cases, but not all. For example, it doesn't work for search engine prompts: we fetch the suggestions from the remove server, and these can change any moment. So memoizing search engine suggestions is quite misleading.

What I'd have is memoizing the results for the prompts that have a constant (= non-lambda) constructor. That should be pretty safe, because the only thing that can break in such a case is a pre/post-processor. But it's quite hard to conjure such a pre/post-processor that would break things, so we're safe.

Still, I'm not sure about this one.

[Ambrevar]

The constructor is run only once during the prompter life-time ;)

Here I'm suggesting to memoize update: If the processors have no side effects, then it's a pure function, which makes memoization totally legit :)

Now what about side effects? As far as I'm aware, no Nyxt prompt has any.
But in any case, the solution is simple: just add a memoize-p prompter option (default to T), this way if the user wants to set impure filters, they disable memoization.

Thoughts?

[aartaka]

Yeah, boolean toggle is the best solution here. But I'm somewhat cautious about setting it to T by default—seems like a potential foot-shooting behavior.

I mean, in case of Nyxt we can add a proxy source class with memoize-p set to T, but in the prompter itself I'd rather have it at NIL to not cause Heisenbugs downstream.

[Ambrevar]

Fair enough, it's super easy to subclass prompter anyways :)