Really? I'd say that a Forth compiler in Forth is going to be somewhat smaller than a Brainfuck compiler in Brainfuck.
Well probably... but no one said anything about bootstrapping being a requirement.
The LiSP book only describes self-hosting implementations while in Forth it's heresy to implement it in anything else than Forth or assembly. Well, perhaps Java implementations are excused for not being implemented in terms of themselves. :-)
Historically, the performance was pretty decent. I don't know how it stacks up against Hotspot today, but I note it does still seem to be a live project after >10 years.
We are talking about interpreters here, not compilers.
Binary Lambda Calculus has both Forth and Brainfuck beat,
with a BLC interpreter in BLC of only 206 bits (under 26 bytes).
Are you sure? From where I stand, it looks as though andrewvc and betterunix were both talking about compilers rather than interpreters, while dysoco didn't specify either, so I inferred that the compiler topic still stood.
I thought dysoco hinted at the brainfuck self-interpreter,
as linked to by peterwaller. There acutally is a brainfuck
compiler called awib (http://code.google.com/p/awib/), but
it's not that small (very artistic though). So I assumed
the discussion had drifted into interpreter territory...
Interpreter, sure. Why would a BF compiler be particularly easy?