Abstraction of HTML interfaces?

I've been looking at PEAR a lot recently, in particular the HTML related classes. However, I always find myself reimplementing the PEAR code because it does everything I want, but it wants to print the code itself. Most of the code I work on now is template driven. But it still uses things like forms and the such and it would be very nice to use some of the form validation stuff, if it didn't want to do the output itself. Has there been any thought about abstracting the HTML part of it away so it can easily be integrated into a template system, or a simple, "echo-it-yourself" type of class? JE