jcs090218 / flx-rs Goto Github PK

Original discussed in jcs-elpa/flx-rs#3

Here is another attempt to implement feature cache, see #3. I am not a rust expert, so I always look for improvement or better attempts.

(flx-score
"~/.emacs.d/straight/repos/orderless/orderless.el" "odrless")
(222 26 28 30 31 32 33 34)

(flx-score
"Code/Everyone/Xamarin/EMJavaBindings/EMJavaBindings/Additions/DrawableSoftReference.cs"
"odrless")
(138 1 62 63 68 69 70 85)

(flx-rs-score
"~/.emacs.d/straight/repos/orderless/orderless.el" "odrless")
(112 23 28 30 31 32 33 34)

(flx-rs-score
"Code/Everyone/Xamarin/EMJavaBindings/EMJavaBindings/Additions/DrawableSoftReference.cs"
"odrless")
(180 1 62 63 68 69 70 85)

A simple query like

#[test]
fn test() { score("Test", "d"); }

panics in find_best_match when trying to find "d" in str_info.

#[test]
fn test() { score("Testd", "d"); }

works since then str_info built from the search string contains the query string char(s).

Seems like a pretty straight forward bug. But I didn't really dig enough into the code to know what the right fix would be.

We may want to expose a way somehow (maybe (Score( str: &str, query: &str, group-separator: &str))) so that the heatmap fn can boost the score of file/directory names.

pub fn get_heatmap_str(scores: &mut Vec<i32>, str: &str, group_separator: Option<char>) {

Original flx implements this through setting up the heatmap-fn on the file cache that's created.

https://github.com/lewang/flx/blob/e3b3f0533e44c5250ce73d728b59a7e96c692b5d/flx.el#L232
(defun flx-get-heatmap-file (filename)
  "Return heatmap vector for filename."
  (flx-get-heatmap-str filename ?/))

If we had the cache (and we're ok with not exposing the cache to elisp), we could add an extra param/boolean? to flx-rs-score like

(flx-rs-score x string USE-FILESYSTEM-HEATMAP-FN)

for it to autopick the right heatmap function in score.

If we have a cache and we are exposing that cache to elisp, it would probably suffice to have the elisp side pass the map over. I think this might be tricky though since the cache that's in elisp resembles more of an object than a true hashmap. (It's more like an object that contains a cache, a heatmap, and a heatmap-fn).

If we don't have the cache, it may make sense to allow the score function take in a group-separator string (or list of strings/not super sure if that makes sense).

e.g.

(flx-rs-score x string GROUP-SEPERATOR-CHAR)

Rust panics with
thread 'test_nok' panicked at 'attempt to multiply with overflow', src/search.rs:256:40
on longer strings. Maybe somewhat surprisingly, both the query as well as the search string seem to contribute to this panic.

#[test]
fn test_ok() {// base, works
    let s = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz1234567890";
    score(s, "ABCDE");
}

#[test]
fn test_ok2() {// s increased by one, query reduced by one, still works
    let s = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz1234567890A";
    score(s, "ABCD");
}

#[test]
fn test_panic() {// s increased by one, query left as is, panics
    let s = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz1234567890A";
    score(s, "ABCDE");
}

I'm not sure whether only short strings are supported or this is a legit bug.