Color me flabbergasted.
That's good and all, but in the "Why it's fast" section of the linked article, they say:
It was? Oh my... They also say that:
SquirrelFish’s bytecode engine elegantly eliminates almost all of the overhead of a tree-walking interpreter. First, a bytecode stream exactly describes the operations needed to execute a program. Compiling to bytecode implicitly strips away irrelevant grammatical structure. Second, a bytecode dispatch is a single direct memory read, followed by a single indirect branch. Therefore, executing a bytecode instruction is much faster than visiting a syntax tree node. Third, with the syntax tree gone, the interpreter no longer needs to propagate execution state between syntax tree nodes.
Why, yes, indeed!
For the record, Rhino has been doing this for ages - AST is compiled to internal "JS bytecode" format that strips away grammar, and then interprets it. This works like this since, well, around the turn of the millenium. (Actually, Rhino can even kick it another notch and can also optionally compile the JS bytecode to Java bytecode, eliminating the interpreter altogether. Which bytecode then the JVM JIT compiler can further compile to machine code at its own discretion.)
On the positive side, they're finally moving on to a better solution, and I congratulate them on the undoubtedly hard decision to finally take the leap toward a more modern execution architecture.