Like most things in Epoch, conditionals are implemented as entities. (If you missed it, there's a full treatment on how the entity system works on the Epoch wiki.) An entity is basically just a blob of code with some attached semantics. For instance, entities can be assigned meta-control code, which is a chunk of code implemented in a library (i.e. outside of the target Epoch program itself, generally in C++ for now) that affects how an entity behaves.
A conditional has a simple piece of meta-control code: it pops a boolean value off the stack, examines it, and instructs the virtual machine to either execute the entity's code, or skip it. This is done by returning a control flag from the meta-control code. Meta-control code is also free to do things like create local variables, modify the values of variables, and so on; that will be useful later on when it comes time to implement loops.
Entities also have parameters just like function calls; in the case of conditionals, they have a single boolean parameter. This is the origin of the value that gets popped off the stack.
The final flow of Epoch code involving a conditional looks something like this:
- Raw Epoch code: Evaluate the conditional expression
- Raw Epoch code: Push the (boolean) value on to the stack
- Meta-control code: Pop the conditional value off the stack
- Meta-control code: If the value is true, return a flag that indicates the Epoch code body of the conditional entity should be executed
- Meta-control code: Otherwise, return a flag that indicates the code should be skipped
- Virtual machine: set the instruction pointer to either inside the entity, or just beyond its end, depending on the meta-control code's response
- Raw Epoch code: execute the condition's body, if applicable
- Raw Epoch code: continue as normal
Elseif and else blocks will be implemented using a technique I'm referring to as entity chaining. As a simplification of the chaining process, this basically allows one entity to tell the VM to chain into the next entity in the code; so if the conditional expression is false, for instance, the initial if entity can chain into the following elseif entities (if any) and then on into the final else (if any). There's some trickery going on in the VM to make sure that the code doesn't execute any more expression evaluations than it has to, but overall the concept is pretty straightforward.
This may seem like a lot of hoops to jump through for just if statements; but the real benefit of the entity system will show up when it comes time to implement loops. Iteration in particular will be very elegant in this system, as meta-control code can use the parameters to the loop entity to control the start and stop conditions, the direction of iteration, and so on. Combined with the ability to define custom infix operators, this can lead to very clean syntax like for(n in [1..10]) { foo(n) }. I'll need some kind of "yield" mechanism to make iterators really powerful, but that shouldn't be too terribly hard.
So... progress on flow control inches forwards, slowly but surely. Soon this unit test will be passing:
entrypoint : () -> (){ debugwritestring("Enter a number:") integer(foo, cast(integer, debugreadstring())) if(foo == 0) { debugwritestring("You entered zero") } elseif(foo == 1) { debugwritestring("You entered one") } elseif(foo == 2) { debugwritestring("You entered two") } else { debugwritestring("You entered something else") }}
And that will make me very happy.
Heh. I implemented this for the first run through Tangent. It was tricky to say the least. I didn't follow the good .NET way it gets implemented, but even that should have a number of gotchas to be aware of.
Plus I'm not sure how much your entity setup will hinder that development.