add aliases search array and display best matches for aliases to user #30
Conversation
Note: This upgrades the required version of fuse.js to 6.6.2. This version allows weighting to be properly specified. This adds an aliases array for each command. A match in the alias has a weight of 2x compared to the name or description fields. If the user enables the displayHints option, the best matching alias will be appended to the end of the command name. This is useful for listing alternate terms for the one that is chosen as the command name. For example: name: emacs, alias: editor will show up as emacs (editor) if the user typed edit. The README.md has more info. This was inspired by the implementation of synonyms described at: https://blog.superhuman.com/how-to-build-a-remarkable-command-palette/#rule-4-build-flexibility-into-your-command-palette
|
This addresses #8. |
|
I forgot to mention that this also changes the sortFn used by fuse.js. I presort my command array by weight. I want the search results to take this into account. If two commands are close enough to each other (< 0.05 difference in scores) they don't sort by score but by their order in the commands array. With lower indexes sorting first. 0.05 is a magic number that produced the results I wanted in a couple of tests but I didn't try to optimize it (or understand the underlying algorithm for the scores). |
I think I have implemented the equivalent of / \(\w[\s\w-]*\)$/ in unicode. So non-latin languages should work. Also refactored the regexp out so it's not in 3 places in the file. Thanks to accidently finding the solution on https://www.roboleary.net/programming/2020/04/24/pimp-blog-reading-time.html.
This seems to do better. It weighs runs of consecuting matching characters hight. So searching for "ab" would rank cab higher than canb.
Also change a couple of variable names to match the new name of the former match_index function.
Removes one level of dereference from the inner loop. Means that changes to match_metric can't change the match. Also renamed variable that stores the metric from sum to metric. the values used to be a simple sum of differences. It isn't anymore.
sort so entry with highest match_metric will be first. If two match_metrics are equal, keep the order the entry is found in the aliases array.
Simple cleanup to clarify what's happening, shorten lines, reduce dereferencing.
I realized this patch replaced the standard fuse.js sort bunction. It is now an option to replace the sort function. Setting orderedCommands: true replced the default fuse.js sort function. This also includes a new replacement sort which is better than the original. The original would sort two items with scores of 0.03 and 2.2E-10 by their order in the commands array. The new function sorts by score if the score is < 0.009. For results that are close (< 0.05 apart), the results will be sorted by their position in the command array. This allows the dev to nudge results when the confidence in the scores is lower. This also pushes the debugOutput option down into App.svelte. In README.md, I moved the aliases and hints section lower and added a section on ordering matches.
| @@ -131,6 +131,7 @@ const c = new CommandPal({ | |||
| hotkey: "ctrl+space", // Launcher shortcut | |||
| hotkeysGlobal: true, // Makes shortcut keys work in any <textarea>, <input> or <select> | |||
| placeholder: "Custom placeholder text...", // Changes placeholder text of input | |||
| displayHints: false, // if true, aliases are displayed as command hints | |||
There was a problem hiding this comment.
Thankyou, I'm coming around to this option 👍
| function removeHints(items) { | ||
| if (! displayHints ) return; | ||
| items.map( (i) => { if ( i.hinted ) { | ||
| i.name = i.name.replace(hintRegexp, ''); |
There was a problem hiding this comment.
So this removes the hint if they have used it? Maybe a comment above here?
If so that's awesome 👍
There was a problem hiding this comment.
So this removes the hint if they have used it? Maybe a comment above
here? If so that's awesome +1
It just resets the command palette names to remove hints. It doesn't matter if they are
used or not. It returns the commands array name properties to their original state for
the next invocation.
This way pulling up the palette doesn't display the hints from the last invocation before the user even types anything.
LGTM changing the code now. Co-authored-by: Ben Winding <[email protected]>
If there is a way to make fuse.js apply a weight to an object, I didn't see it.
Working on it now. |
|
I think I got this right. I don't use git or GitHub much. My preferrd DVCS is Fossil because well, I'm a fossil 8-). Take another look. Also nice use of ChatGpt. what did you feed it to get it to generate the refactored code? |
Looks good, I'll fix the type errors in the next few days, probably makes sense to put this in a
It's amazingly flexible this, you can be pretty expressive and just paste code in below it |
|
One thing I would leave the comment that 0.05 and 0.009 are magic numbers with little if any research to justify them. |
Note: This upgrades the required version of fuse.js to 6.6.2. This version allows weighting to be properly specified.
This adds an aliases array for each command. A match in the alias has a weight of 2x compared to the name or description fields. If the user enables the displayHints option, the best matching alias will be appended to the end of the command name. This is useful for listing alternate terms for the one that is chosen as the command name. For example:
name: emacs, alias: editor
will show up as emacs (editor) if the user typed edit.
The README.md has more info.
This was inspired by the implementation of synonyms described at:
https://blog.superhuman.com/how-to-build-a-remarkable-command-palette/#rule-4-build-flexibility-into-your-command-palette