Tailwind CSS v4.0 support

[rickyzhangca]

with the Tailwind CSS v4.0 beta, this error is given regardless of using the config file or the new CSS-first approach to import this plugin, appreciate anyone could take a look!

[plugin:@tailwindcss/vite:generate:serve] Package subpath './lib/util/flattenColorPalette' is not defined by "exports" in /Users/.../Documents/.../node_modules/.pnpm/tailwind-scrollbar@3.1.0_tailwindcss@4.0.0-beta.1/node_modules/tailwindcss/package.json

[rickyzhangca]

tailwindlabs/tailwindcss#15318 this should fix the bug, needs to confirm once it's out

[mst-mkt]

tailwindlabs/tailwindcss#15318 this should fix the bug, needs to confirm once it's out

This error is resolved at tailwindcss@4.0.0-beta.7, but at the same time the following different errors occur

Package subpath './lib/util/toColorValue' is not defined by "exports" in repo/node_modules/.pnpm/tailwind-scrollbar@3.1.0_tailwindcss@4.0.0-beta.7/node_modules/tailwindcss/package.json

This toColorValue function does not exist in the tailwind v4 code, so it may need to be addressed on the tailwind-scrollbar side

[rickyzhangca]

thanks!

[adoxography]

So, for full transparency here, I'm going to drag my heels on officially supporting v4 until I can get the package's test suite running using their new architecture. The way Tailwind is actually invoked has changed in v4 (i.e. it's not just a postcss plugin anymore), and that means that none of the tests are presently able to run.

Although it may seem like a simple little plugin, it's shockingly difficult to target pseudoclasses of pseudoelements (think ::-webkit-scrollbar-thumb:hover) in Tailwind. As a result, this plugin has to do some manipulation on the postcss AST that Tailwind generates, and that's a little dangerous; it's easy to create scenarios like #101 where that manipulation inadvertently messes up the CSS output from completely unrelated packages. I have tests in the test suite that should prevent any regressions, and I want confirmation from them before I go about releasing anything. Additionally, I'm not even sure that the approach the package uses will work at all in the new architecture, which could require a partial rewrite.

I'll finally have some down time at work this month, so I'm hoping to find time to do some reading on how the authors intend Tailwind to be invoked now