Candidate extraction from Clojure/Script files sometimes produce false negatives

[eneroth]

What version of Tailwind CSS are you using?

v4.1.4

What build tool (or framework if it abstracts the build tool) are you using?

postcss-cli 11.0.1

What version of Node.js are you using?

v23.11.0

What browser are you using?

Chrome, Safari, Firefox

What operating system are you using?

macOS

Reproduction URL

#17087

Describe your issue

First of all, a huge thank you for including Clojure.

The issue stems from classes written as keywords that contain fractions. For example,

($ :div {:class [:flex :gap-1.5 :p-1]} …)

In this case, the candidates extracted from :gap-1.5 seem to be :gap-1 and 5, which will end up not matching any class present in the compiled HTML: <div class="flex gap-1.5 p-1">…</div>.

A suggested "good enough" solution is to, in addition to considering the keyword when broken down by period (.) separation, additionally consider the whole string as a candidate. I.e. such that it goes from considering the candidates,

["gap-1" "5"]

To

["gap-1" "5" "gap-1.5"]

This may end up producing some valid classes which are not actually present in the code (gap-1 in this case), but is still preferable, as it trades precision for accuracy.

Another, more elaborate solution would be to join candidates pair-wise and accept them iff they produce a valid class. Eg., the candidates ["gap-1" "5" "p-1"] are extracted. When the first pair is joined with ., they produce the candidate gap-1.5, which is therefore accepted in place of the previous two candidates, producing ["gap-1.5" "p-1"]. In the next iteration, the join produces ["gap-1.5.p-1"], which is nonsense, and is therefore rejected.

More discussion can be found towards the end of #17087.