a fully working interpreter?

README.md

@@ -6,7 +6,40 @@ A PushGP implementation in Haskell
     * [ ] Write tests for every function.
     * [ ] tests/ are just copied from make-grade, need to write for this project.
 
-I would really like to benchmark the following three versions for speed: 
+## Design considerations
-1) My custom data one (this repo)
+The biggest design constraint is that for the exec stack (but not data stacks)
-2) direct independent lists for each stack (Rowan's miniHush)
+we have to be able to detect type at runtime.
-3) eval string (similar to my custumized version of propel clojure)
+
+A simple way to do this for the exec stack is as a list of custom data type.
+That custom Gene data type must have as many sub-types as there are types + fuction types.
+
+If the input stack is singular, then it needs a general Gene data type,
+but if there was an input stack per type, they could be specific.
+
+I would really like to benchmark some of the following three versions for speed: 
+
+1) Where some functions can act on all stacks (this repo),
+and thus every data stack is a list of a more general Gene type,
+elements of which are wrapped in data TypeGene so they can be identified in stack-general functions.
+To bind all the stacks for convenience,
+we could put each stack list in a tuple, or a custom data type, Data.Map or Data.HashMap.
+The exec stack will always need a more general Gene type,
+with Gene types wrapping each individual thing, for runtime identification.
+
+2) Where type-specific functions act on each stack independently, 
+and thus each data stack can have exclusive specific basic types,
+which are not wrapped in data TypeGene, because they do not need to be identified.
+To bind all the stacks for convenience,
+we could put each stack list in a tuple, or a custom data type,
+but not in a or Data.Map or Data.HashMap, as those require homogenous (K, V) pairs.
+The exec stack will always need a more general Gene type,
+with Gene types wrapping each individual thing, for runtime identification.
+
+3) Alternatively, for the exec stack, we could store strings, 
+and eval strings (similar to my custumized version of propel clojure)
+Regular and input stacks can stil be either TypeGene or basic types.
+This is clearly not ideal.
+
+4) For the exec stack itself,
+typeable, data generic, ghc.generic, data.dynamic, heterogeneous lists, etc. could also help,
+to detect the type of variables at runtime, but I would rather stick to language basics at first.

src/Push.hs

@@ -1,46 +1,31 @@
 module Push where
 
-import Data.List (foldl')
-
 -- import Debug.Trace (trace, traceStack)
 
--- GeneModular or Gene?
+-- The exec stack must store heterogenous types,
--- Should we use a StateFunc or *Func for each push type?
+-- and we must be able to detect that type at runtime.
--- Start with whole StateFunc since it is monolithic (easier to start),
+-- One solution is for the exec stack to be a list of [Gene].
--- then generalize and abstract with an apply method that itself takes a simpler function and the state?
+-- The parameter stack could be singular [Gene] or multiple [atomic] types.
-{-
-data GeneModular
-  = IntGene Int
-  | FloatGene Float
-  | BoolGene Bool
-  | StringGene String
-  | IntFunc [([Int] -> [Int] -> [Int])]
-  | StrFunc [([String] -> [String] -> [String])]
-  | BoolFunc [([Bool] -> [Bool] -> [Bool])]
-  | FloatFunc [([Float] -> [Float] -> [Float])]
--}
-
 data Gene
   = IntGene Int
   | FloatGene Float
   | BoolGene Bool
   | StringGene String
-  | StateFunc (State -> State -> State)
+  | StateFunc (State -> State)
   | Close
-  | Input Gene
 
 --  | Block [Gene]
--- If we do plushy,
+-- If we do plushy (as opposed to just detecting the Close itself,
 -- then we may need to make a structually recursive data structure for the "program" data structure
 -- exampleGenome = [Program] rather than [Gene], or just include the Block above?
 
 data State = State
   { exec :: [Gene],
-    int :: [Gene],
+    int :: [Int],
-    float :: [Gene],
+    float :: [Float],
-    bool :: [Gene],
+    bool :: [Bool],
-    string :: [Gene],
+    string :: [String],
-    input :: [Gene]
+    parameter :: [Gene]
   }
 
 emptyState :: State
@@ -51,38 +36,67 @@ emptyState =
       float = [],
       bool = [],
       string = [],
-      input = []
+      parameter = []
     }
 
-stackUpdate :: [Gene] -> State -> State
+-- Each core func should be: (State -> State -> State)
-stackUpdate newstack@(StateFunc _ : _) (State _ i f b s p) = State newstack i f b s p
+-- but each core function can use abstract helper functions.
-stackUpdate newstack@(IntGene _ : _) (State e _ f b s p) = State e newstack f b s p
+-- That is more efficient than checking length.
-stackUpdate newstack@(FloatGene _ : _) (State e i _ b s p) = State e i newstack b s p
+-- Everntually, this can be part of the apply func to state helpers,
-stackUpdate newstack@(BoolGene _ : _) (State e i f _ s p) = State e i f newstack s p
+-- which should take the number and type of parameter they have.
-stackUpdate newstack@(StringGene _ : _) (State e i f b _ p) = State e i f b newstack p
-stackUpdate newstack@(Input _ : _) (State e i f b s _) = State e i f b s newstack
-stackUpdate _ state = state
-
-unpackIntGene :: Gene -> Int
-unpackIntGene (IntGene item) = item
-
--- Start with monolithic intAdd function:
 intAdd :: State -> State
-intAdd state =
+intAdd (State es [] fs bs ss ps) = State es [] fs bs ss ps
-  let result = sum (map unpackIntGene (take 2 (int state)))
+intAdd (State es [i] fs bs ss ps) = State es [i] fs bs ss ps
-      dropped = drop 2 (int state)
+intAdd (State es (i : is) fs bs ss ps) = State es ((i + head is) : drop 1 is) fs bs ss ps
-   in stackUpdate (IntGene result : dropped) state
 
--- Later, generalize a function called applyFuncToState,
+--  let result = sum (take 2 (int state))
--- which takes each simpler atomic function, and the state,
+--      dropped = drop 2 (int state)
--- and applies the function to the state, for example:
+--   in updateIntStack (result : dropped) state
--- intAdd :: (Int, Int) -> Int
+
--- applyFuncState :: AtomicFuncTypes -> State -> State
+-- For safety, pattern match on [] and i:is or check for <2 long list after take 2?
--- this would change Gene to something like GeneModular above.
+
+-- Optionally, split this off into independent functions
+parameterLoad :: State -> State
+parameterLoad (State es is fs bs ss []) = State es is fs bs ss []
+parameterLoad (State es is fs bs ss (p : ps)) = case p of
+  (IntGene val) -> State es (val : is) fs bs ss ps
+  (FloatGene val) -> State es is (val : fs) bs ss ps
+  (BoolGene val) -> State es is fs (val : bs) ss ps
+  (StringGene val) -> State es is fs bs (val : ss) ps
 
 -- Wow, a one-liner for interpreting a paretheses-free genome...
 -- Does not handle any data elements in genome yet,
 -- but condition could be added to the lambda.
 -- Need to update this when adding parethetical blocks too.
-interpretGenome :: State -> [(State -> State)] -> State
+-- interpretFuncOnlyGenome :: State -> [State -> State] -> State
-interpretGenome state = foldl' (\acc f -> f acc) state
+-- interpretFuncOnlyGenome = foldl' (\acc f -> f acc)
+-- While this is not usable, it illustrates we want this pattern:
+-- foldl (strict, cumulative accumulator), and not this pattern:
+-- foldr (greedy/lazy incremental or quit early)
+
+-- Loads a genome into the exec stack
+loadProgarm :: [Gene] -> State -> State
+loadProgarm newstack (State _ i f b s p) = State newstack i f b s p
+
+-- Takes a Push state, and generates the next push state via:
+-- If the first item on the EXEC stack is a single instruction
+--     then pop it and execute it.
+-- Else if the first item on the EXEC stack is a literal
+--     then pop it and push it onto the appropriate stack.
+-- Else (the first item must be a list) pop it and push all of the
+--     items that it contains back onto the EXEC stack individually,
+--     in reverse order (so that the item that was first in the list
+--     ends up on top).
+interpretExec :: State -> State
+interpretExec (State [] is fs bs ss ps) = State [] is fs bs ss ps
+interpretExec (State (e : es) is fs bs ss ps) =
+  let poppedState = State es is fs bs ss ps
+   in case e of
+        (IntGene val) -> interpretExec (State es (val : is) fs bs ss ps)
+        (FloatGene val) -> interpretExec (State es is (val : fs) bs ss ps)
+        (BoolGene val) -> interpretExec (State es is fs (val : bs) ss ps)
+        (StringGene val) -> interpretExec (State es is fs bs (val : ss) ps)
+        (StateFunc func) -> interpretExec (func poppedState)
+
+-- The safety of interpretExec on empty stacks depends on the functions it calls.
+-- Need to make interpretExec strict, right?

src/main.hs

@@ -3,18 +3,4 @@ import GP
 import Push
 
 main :: IO ()
-main = do
+main = do pure ()
-  let exampleGenome = [intAdd, intAdd]
-  let exampleState =
-        State
-          { exec = [IntGene 5, FloatGene 3.4, BoolGene True, StringGene "hi"],
-            int = [IntGene 1, IntGene 2, IntGene 3],
-            float = [FloatGene 1.2, FloatGene 1.7],
-            bool = [BoolGene True, BoolGene False],
-            string = [StringGene "Hello", StringGene "Push"],
-            input = [Input $ IntGene 1, Input $ StringGene "Hi", Input $ BoolGene True, Input $ FloatGene 1.3]
-          }
-  -- This is an example of applying one function (head exampleGenome produces intAdd) to the exampleState:
-  assert ([3, 3] == map unpackIntGene (int (head exampleGenome exampleState))) pure ()
-  -- This function applies an entire genome to the starting state, and produces the final state:
-  assert ([6] == map unpackIntGene (int (interpretGenome exampleState exampleGenome))) pure ()