Thoughts on adding a frecency argument (global, channel, cli) much like history

[lalvarezt]

The purpose is that when enabled, items selected on previous runs will be favored (matcher boost) over the default matcher scoring.

If you always launch dotfiles and select .zshrc, it makes sense that if you type zsh then the first thing on the list is the file you opened many times before

@alexpasmantier thoughts?

[alexpasmantier]

This does make sense and IMO should be a channel opt-in. I'm not on my computer right now and can't remember if the history data file contains channel information?
If it does, then we already have all we need to implement this without any additional writes to disk 😁

[alexpasmantier]

String comparison gives you O(mn) while hashing gives you O(n) where m is the size of the input list and n is the size of the frecency list.

So I would argue using a very simple hashing algorithm will be faster.

[lalvarezt]

to be clear, I was referring to the frecency list only (entries).

I chose to keep two "lists":

• dataset is an FxHashSet<String> that gets filled as the items are loaded, O(1) for contains
• entries has the list of N most recently used items as a Vec<FrecencyEntry>

pub struct FrecencyEntry {
    /// The actual selected entry (file path, command, etc.)
    pub entry: String,
    /// The channel that the entry belongs to
    pub channel: String,
    /// Number of times this entry was selected
    pub access_count: u32,
    /// Timestamp of the last access
    pub last_access: u64,
}

The main reason here was that we have to iterate over the list anyways (calculate the new score, filtering by channel/global, sort). So it makes more sense to keep it as a vec instead of a hashset

[alexpasmantier]

Yes, that's what I have in mind too. I'm just saying that checking if DatasetHashset contains each frecency item will be faster than having two lists of String and doing manual string comparison (which is O(nm)).

[alexpasmantier]

So the example you're giving is indeed the way to go IMO (hashing every dataset item).

[lalvarezt]

have a look #671, don't be too hard on me 😅