v1.7.3

Fixes

Re-hash content-hashed assets after optimization (#6)

optimize rewrote content-hashed assets (Vite/Rollup [hash], webpack [contenthash]) in place while keeping the original filename. With Cache-Control: immutable on those assets, a returning browser kept serving a stale cached copy after a redeploy - old cp-* rules against new cp-* references - silently breaking styling.

optimize now re-hashes every content-hashed asset it modifies: the file is renamed to a new content-hash filename and every reference to it (HTML, JS, CSS url(), and bundler manifests) is rewritten. Naming is deterministic, so identical input produces identical output.

• On by default. Only renames files whose names already look like content hashes; stable names like styles.css are left alone.
• Pass --no-rehash to keep the original filenames (for example, when running classpresso before the bundler applies its own hashing step).
• New rehashAssets config option (default true).

Auto-exclude non-flattenable classes to prevent silent CSS loss

Classes that cannot be expressed as flat CSS properties - variant prefixes (md:, hover:, dark:), container queries (@sm, @[800px]), and combinator utilities (space-y-*, divide-x-*) - are now excluded automatically so patterns containing them are skipped instead of losing styles. Controlled by the new excludeNonFlattenableClasses config option (default true).

Notes

Why classpresso, not gzip: the win is at the render engine, not the wire. Every class on an element is another selector the browser matches during style recalculation. Collapsing a dozen utility classes per element into one cp-* class means fewer lookups and a smaller DOM. Gzip only collapses repeated bytes in transit, then the browser inflates them back to the same token count - so transfer-size deltas do not capture what the tool does.

Full changelog: v1.7.2...v1.7.3