Crosscut

The postfix syntax in previous prototypes was supported by a stack-based evaluation model, which turned out to be problematic. The new model I came up with after briefly trying, then dismissing, a more traditional approach, is also using postfix syntax. But it's not supported by a stack-based evaluation model.

This model takes inspiration from functional languages like Haskell or the ML family (and thus indirectly from lambda calculus). In those languages, functions accept exactly one argument and return one result. Compared to the arbitrary number of inputs and outputs that the stack-based model enables, this is a welcome simplification.

And I do believe that this limitation fits postfix syntax quite well. It makes both control flow and data flow very obvious. You see code like a b c, and you immediately know that a is executed first and produces something that is then processed by b, which then in turn gives its result to c. There's more to it, because this alone would be very limiting. But that's a topic for tomorrow.