[ufo]

This is good for hand-written html but annoying for machine-written html.

[ericfontaine]

machine would just have to count '/t' tabs at beginning of each line to parse. Is that any slower than scanning every character for '<'?