Subsetting

A complete font may carry Latin, Latin Extended, Cyrillic, Greek, Vietnamese and a few thousand symbols. A page in English uses about a hundred of those, and a visitor downloading the rest is paying for alphabets that will never be painted. Cutting the file to the ranges you need is usually the single biggest saving available on a page that uses web fonts.

The trap is cutting too hard. A subset that keeps basic Latin and drops Latin Extended breaks Turkish, Polish, Czech and every other language whose accents live there — and it breaks silently, because the browser substitutes another font for the missing characters and only somebody reading that language notices. Turkish is the usual casualty: the dotless ı and the dotted İ are in Latin Extended-A, not in basic Latin.

The second trap is subsetting a font you then install on a desktop. A web subset has no business being an installed font; you will find the missing characters weeks later, in a document.