Also wanted to add that while I think this is very clever, and I am a big fan of Mr Tatham's work, that by the time we're talking about the 'advanced' version at the end we're edging towards using a stack-based system in the form of a context object and at that point it feels like we're just a leap and a jump to stack-based coroutines and full-on cooperative multithreading.

Also, by the time you're passing a coroutine context around anyway, you could refactor (say) the decompressor around a decompression context and the code would stay nice...

It's definitely interesting though, and it's been a few years since I read that coroutines page.