Incorrect prioritization of filename suggestions in quick open dialog (Ctrl+P)

[waldyrious]

Does this issue occur when all extensions are disabled?: Yes

• VS Code Version: 1.74.2
• OS Version: Ubuntu 20.04

Steps to Reproduce:

1. Clone the mdn/content repository
2. Open the quick open dialog by pressing Ctrl+P
3. type "ftu index" or "glossary ftu index"
4. Observe that the files/en-us/glossary/ftu/index.md is not the first result. Instead, other results that split the "ftu" into separate letters (like e.g. files/en-us/glossary/turn/index.md) get higher priority.

[TylerLeonhardt]

Does this repro a similar behavior when your second word has more than 3 characters? So like stun index? I think the problem here is that matching the first character in the path is weighted higher than an exact word match inside of the path.

cc @bpasero if you have thoughts.

[bpasero]

You can print diagnostics by commenting out all the code that uses DEBUG and/or DEBUG_MATRIX:

vscode/src/vs/base/common/fuzzyScorer.ts

Lines 22 to 23 in 688900a

	// const DEBUG = false;
	// const DEBUG_MATRIX = false;

I noticed it was not very clear so I have #171308 opened to improve the output.

Running so (with my changes) produces (for ftu):

I think the issue here is that we give the initial f a higher score since it is at "beginning of word" and thus wins over the f at the end of the path (by a few points). Note that we consider workspace relative paths for matching, hence the path starts with f.

I think the algorithm here fails to recognise that the ftu at the end of the path would produce an overall higher score given the consecutive bonus over the scattered ftu, where f matches at the beginning. I am not 100% sure why that is, because as far as I remember we match each input character over all possible combinations and then take the highest score. Would be something maybe to investigate and follow up.

[bpasero]

Matrix:

        f	 i	    l	    e	    s	    /	    e	    n	    -	    u	    s	    /	    g	    l	    o	    s	    s	    a	    r	    y	    /	    f	    t	    u	    /	    i	    n	    d	    e	    x	    .	    m	    d
f	M1/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	M0/S10	
t	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M1/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	M0/S12	
u	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M0/S0	M2/S19	M0/S19	M0/S19	M0/S19	M0/S19	M0/S19	M0/S19	M0/S19	M0/S19	M0/S19

[bpasero]

Previously reported as #56667 and #69648