Expand classRegex search range
#840
Conversation
| start: document.positionAt(Math.max(0, positionOffset - 1000)), | ||
| end: document.positionAt(positionOffset + 1000), | ||
| start: document.positionAt(0), | ||
| end: document.positionAt(positionOffset + 2000), |
There was a problem hiding this comment.
Thanks for this PR @bradlc, it makes the situation with #837 much better.
Do you think if it’s possible to expose 2000 as an extension option? In our codebase, we have several instances of cva() which are pretty heavy on comments, so they end up being more than 2K characters long.
I guess that if we look for cva( + ), the regexp might still not work. Or is it necessary to rewrite it? What do you reckon?
There was a problem hiding this comment.
Yeah I'm open to adding it as a config option. Can you share an example from your project that wouldn't be caught within the new range?
There was a problem hiding this comment.
The project is closed sourced, so I can’t share the snippet, sorry. You can imagine chadcn → button.tsx (which is 1000 chars long), but with:
- a few more variants, including
compoundVariants - extra spacing between groups of lines
- comments explaining the difference between variants
- a few more class names with selectors (
hover:,sm:,data-[some-state=some-value]:,[&>svg]:and so on)
It’s quite easy to go beyond 2K symbols with all these.
There was a problem hiding this comment.
Here is an ‘aerial view’ of a real-life cva() call with 10,724 characters in-between 😁
After doing an admittedly quite small amount of benchmarking it seems that expanding the search range _does_ have a perf impact but it's not so much of one that we shouldn't increase it. Though this very likely depends on the regex in use. I've done a few major things in this PR: - Searching for class lists when hovering now starts at the beginning of the document. This important for users who are using things like `cva` (the same reason it was expanded for completions in #840) - The search range for completions inside a class attribute have been expanded to 15k characters. This is useful if you're using a really large cva definition in one class list. - Ditto for searching for custom regexes which is useful when using CVA with separate variables in a JSX/TSX document. Fixes #1032 Fixes #984 This is a mostly a stop-gap measure until I can land a better system for "parsing" documents that aims to make most class detection trivial and quite fast. Custom regexes won't necessarily be able to take advantage of that but if I can get some perf wins elsewhere I might be able to expand the search range even further. Ideally I'd figure out a way to run regexes in a worker thread and allow them to be "cancelled" when they take too long but that is a fairly major-ish undertaking.
Fixes #832
Fixes #837
This PR expands the
classRegexsearch range so that longer matches can be found (#837). Additionally the search range now always starts at the very beginning of the document, so that patterns match as expected. An example of why this matters is a simple string pattern:'[^']*'. Consider the following document:If we start searching from the beginning the matches would be:
'', 'hello world', but if we start from the second character the only match would be:' + '