Internal Language IDP4
03/11/2016
If we want to support functions ...
Why differentiate between predicates and functions?
Predicate = Function to Bool
MINIMAL LAnguage constructs
- Function Application
- Lambda Abstraction
- Set of constants
Using this we can easily recover FO
\forall x: P(x) \Rightarrow Q(x).
∀x:P(x)⇒Q(x).
IDP3
QuantForm(forall,+,x)
BoolForm(disjunctive,+)
PredForm(-,P)
PredForm(+,Q)
VarTerm
Var(X,Type)
VarTerm
Var(X,Type)
\forall x: P(x) \Rightarrow Q(x).
∀x:P(x)⇒Q(x).
IDP3
QuantForm(forall,+,x)
BoolForm(disjunctive,+)
PredForm(-,P)
PredForm(+,Q)
VarTerm
IDP4
Ap
Forall
Lambda
Ap
Implies
Ap
Ap
Ap
P
Var(0)
Q
Var(0)
Var(X,Type)
VarTerm
Var(X,Type)
Challenges
- What set of builtins do we want?
- Forall, Exists, And, Or, Not, Implies, ....?
- Do we need N-ary application?
- And, Or, +, *, min, max, ... ?
- Use lists/sets for N-ary language constructs
- Binary version shorthand for list constructor
- De Bruyne indices
- Easy to make mistakes
- Idempotent printing/parsing cycle?
p \land q \land r
p∧q∧r
Ap
And
Ap
P
And
Ap
Q
Ap
R
Ap
And
Ap
[]
Ap
P
:
Ap
Q
:
Ap
Ap
R
:
Ap
Tentative list construction
From Fo to FO(.)
- Arithmetic
- Trivial
- Types
- Not needed in internal language
- Partial Functions
- Mainly a semantics issue, no representation issue
- Aggregates
- No fundamental difference between aggregate and quantifier
The easy ones
Constructors
Definitions
The Hard ones:
Constructors
- Integration of GADT's in Lambda Calculus is well known
- My stance: copy what they're doing
- Missing: good insight in standard implementations
Extra Language Construct:
Case-construct
case x of (a,b) -> a + b
Could be simulated as
return=c \leftarrow \exists a ,b: x = (a,b) \land c = (a + b)
return=c←∃a,b:x=(a,b)∧c=(a+b)
Definitions
- Symmetry between Functions and Predicates breaks down
- My Proposal: represent as specialized letrec
- of type: U -> Bool, needed for uniform prop<->pred
Extra Language Construct:
Define-construct
let { p () = q () q () = p () z () = p () || q () }
let { transCl e (x,y) = e (x,y) || exists (\z ->
e (x,z) && transCl e (z,y)
}
Some Remarks
- No explicit polymorphism
- Consequence:
Compiling of polymorphic code will always be impossible
- Consequence:
- No explicit typing information could prevent sanity check of transformations
- For some constructs, we will need smart specialization algorithms to IDP3-like data structures
- What about the generators?
Lang constructs
- Application
- Abstraction
- Builtin Constants
- Variable References
- Case
- Definition
Let? (easily simulated internally using abstraction)
Internal Language IDP4
By Ingmar Dasseville
Internal Language IDP4
- 1,066