mosra
Fuzzy searching?
I was wondering if there was any interesting in allowing fuzzy search?
Some of the background for this is that I mantain a C API where there's a lot of prefixes to keep things unique (product initials + "class" name + function name).
Right now it appear that the search works only on the start of each object being searched.
For example:
Compare to if I just type "editing":
I'm very fond of how Sublime Text (and many other editors like VSCode etc) let you search. In the example above, if I typed "tee" it would value the upper case letters in the symbols so that "TextEditingEvents" ranked high.
I have in the past experimented with fts_fuzzy_match for such functionality for some projects I'm working on. (https://github.com/forrestthewoods/lib_fts/) It's been working rather well.
More details on the logic here: https://www.forrestthewoods.com/blog/reverse_engineering_sublime_texts_fuzzy_match/
Any interest in this?
See here for a WIP implementation of a similar thing by @sizmailov: https://github.com/mosra/m.css/pull/149
I wanted to implement something like this (in particular the {tee, texee, texede} -> TextEditingEvent variant) when doing the original search implementation but I put it aside because it wasn't strictly needed for the MVP. The search is implemented as a trie, so I'm not sure if the libs you linked would be of any use here, but I think I could still dig up the original implementation somewhere and finish it -- if my time allows, I guess you get the idea based on the frequency I reply on the issues here :sweat_smile:
I mantain a C API where there's a lot of prefixes to keep things unique (product initials + "class" name + function name)
With the change I did for #127, I finally have my hands free to add some config option allowing this (a similar case is wanting to search without get_ / set_ prefixes).
See here for a WIP implementation of a similar thing by @sizmailov: #149
Oh, that's interesting. I'm subscribing to that thread. Even though that's not complete fuzzy search, extracting words by camelCase or under_score separation still helps.
The search is implemented as a trie, so I'm not sure if the libs you linked would be of any use here
I did give it a quick try a few weeks ago. But I got stomped trying to understand the structure of the search data. The article you mentioned explains a lot!
Having good debug visualization is key to understanding the data.
Indeed! Does m.css come with the tools to debug the search data?
if my time allows, I guess you get the idea based on the frequency I reply on the issues here
No worries, I fully understand the challenges of maintaining a project like this. Not expecting you to do anything. I started this thread because I had tried to provide a PR myself for such functionality, but unfortunatly I got rather lost in the data structure and how data was obtained.
With the change I did for #127, I finally have my hands free to add some config option allowing this (a similar case is wanting to search without get_ / set_ prefixes).
Yes, we have a number of Get/Set prefixed functions. (well, prefixed before the function name, but after the product and class prefixes. One thing I wanted to look into was sorting that still kept Get/Set functions next to each other without having to resort to documentation markup. I might revisit that later on.
Does m.css come with the tools to debug the search data?
Yes, the documentation/test_doxygen/test_search.py can be used to visualize the search data the same way as was done in the article, including colors. I hope it still works properly, didn't touch the visualization code for over two years :)
One thing I wanted to look into was sorting that still kept Get/Set functions next to each other
Unless I misremember how the lookup and result population behaves, that'll happen automagically when the get/set prefixes get stripped. Or .. you mean in the doxygen-generated output? Disable the SORT_MEMBER_DOCS option, that one is enabled by default and absolutely useless in my opinion -- with it disabled, you'll get the functions ordered the same way as in the file (and there I assume you keep getter and setter pairs together).