Last time I tried v4 the automatic content detection was pretty bad. It would falsely detect tailwind classes if it seen a similar name in a string in a code file. It seemed pretty dumb.
That is more or less how it worked in v3 already, because any string could be used as a CSS class.
"The way Tailwind scans your source code for classes is intentionally very simple — we don’t actually parse or execute any of your code in the language it’s written in, we just use regular expressions to extract every string that could possibly be a class name."
The only difference in v4 is that scanning is so fast, they just scan roughly everything by default. You can still opt into manually specifying which files to scan.
In v3, one can use `blocklist` in the config to eliminate false-positives, but with the config file gone, this seems no longer possible, so I won't be upgrading.
In one project, roughly 75% of the detections were false-positives, unnecessarily bloating the CSS.
