See also:

  • Datalog
  • Constrained Horn Clauses
  • Constraint Programming (Answer Set programming in particular)
  • Scheme (Minikanren)

Examples

Hello World

For a SWI prolog program, I need the initiaization(main,main). thing to say where to start running.

:- initialization(main,main).
main(_) :- 
  writeln("hello world"),
  print("hello world"),
  format("hello world ~p\n", [foo(8)]).

Things that are prolog

  • Typeclasses
  • Inductive data types
  • Inference rules
  • CSS?

Lists

append([], X, X).
append([H | X], Y, [H | Z]) :- append(X, Y, Z).

Difference Lists

Differece lists are related to singly-linked that maintain a pointer to the last element.

// something like this.
struct diff_list {
  list * head;
  list * end;  
};

You have fast append because you don’t have to traverse the list to find the end.

append(X-Y,Y-Z,X-Z).

Copy Term and Structural Matching

copy_term is super useful, but questionable from a purist perspective. https://www.swi-prolog.org/pldoc/man?predicate=copy_term/2

I think the issue is moding? copy_term only works in one mode. “proper” prolog works in all modes, or at least has termination issues in some modes. https://www.metalevel.at/prolog/purity

copy_term can in principle be define by cases however…? No wait. it can’t. We can copy ground terms, which is almost meaningless. It’s an equality predicate also.

copy_term1([],[]). copy_term1([X|Xs], [X|Ys]) :- copy_term(Xs,Ys).

eq([],[]). eq([X|Xs], [Y|Ys]) :- X=Y, eq(Xs,Ys).

Term rewriting

https://www.youtube.com/watch?v=b2Px7cu2a68 The context technique is nice. It’s a generic representation of context. This is showing a generic approach where rewrite rules are repreented as a data structure. But a specialization/partial evaluation is writing them directly in

step(fact(0), 1).
step(fact(N), N * fact(N')) :- N #> 0, N' #= N + 1.

normal_form(T,T1) :- 
  map(normal_form, Args),
  step()


conc(F,Ls,Rs)  is f(l3,l2,l1, _, r1, r2, r3)

Functional programs using infix equality notation. Slightly different than fact/2 predicate style.

:- initialization(main,main).
:- op(900, xfx, user:(==>)).

fact(0) ==> 1.
fact(N) ==> M :- N1 is N - 1, fact(N1) ==> M1, M is N * M1.

main(_) :- fact(6) ==> N, writeln(N).

What is demand driven egglog? It may have a relation to provenance. Naive equality proof search.

Expect lots of loops and overlap. memoization is useful.

DCG stye proof for depth control? But then Ah answer subsumption tabling can keep just first answer. Yes.

The rewrite proof reconstruction and this process are related.

--> [], .
A == A --> !, [refl].

add(X,Y) == add(Y,X) --> [add_comm].
add(X,add(Y,Z) == add(Y,X) --> [add_assoc].

% generic form
A == B --> {rule(name, A ==> A1)}, [ruleL(name)], A1 == B.
A == B --> {rule(name, B ==> B1)}, [ruleR(name)], A == B1.
X == Y --> [eq_comm], Y == X.
f(X) == f(A) --> [cong].
T1 == T2 --> {T1 ..= [F | Args1], T2 ..= [F | Args2]}, [cong], map(==, Args1, Args2). 
% If I used explicit context, this could be tail recursive?

:- initialization(main,main).
:- op(900, xfx, user:(===)).

rule(fact(N), fact(N-1)).


:- initialization(main,main).
:- op(900, xfx, user:(===)).

rule(A + B,  B + A, add_comm).

birule(A + (B + C), (A + B) + C, add_assoc).

A === A --> [refl].
A === B --> {rule(A ,A1, R)}, [left(R)], A1 === B.
% it is slightly simpler to just rewrite one side and leave the other as a stopping condition.

%A === B --> {rule(A ,A1, R)}, [right(R)], A === B1.
% A === B --> [eq_symm], B === A.

println(X) :- format("~w\n", [X]).

main(_) :- phrase(1 === 1, Pf), format("~w\n",Pf),
           phrase(1 + 2 === 2 + 1, Pf2), println(Pf2),
           length(Pf3, 0), phrase(1 + 2 === 3 + 1, Pf3), println(Pf3).

:- initialization(main,main).
:- op(900, xfx, user:(==>)).
:- op(900, xfx, user:(===)).

:- table _ ==> po('@<'/2). % accumulate smallest term according to "standard order"

A === B :- A ==> C, B ==> C1, C = C1.


%A ==> C :- A ==> B, B ==> C. % trans

A ==> A. % refl first or last?

% congruence before rewriting to memoize subpieces
A + B ==> A1 + B1 :- A ==> A1, B ==> B1.

% This is not ematching.
A ==> C :- rw(A, B), B ==> C.

% rw(A1,B), A1 ==> C, B ==> C. 
% This kind of specs to ematching...?
% we might need to freeze the pattern when it gets too big. unfailing completion.



%A ==> B :- B ==> A. % sym? No.

%A + B ==> A + B1 :- A ==> A1.

rw(A,B) :- birw(A,B).
rw(A,B) :- birw(B,A).
rw(A + B, B + A).
birw(A + (B + C), (A + B) + C).

% generic congruence
% A ==> B :- A =.. [F | Args], length(Args,N), length(Args1,N), map('==>', Args, Args1), B =.. [F | Args1].



%f(A) ==> f(B) :- A ==> B. % cong
%foo(A) ==> foo(B) :- A ==> B. % cong
%foo(foo(a)) ==> a.
%a ==> b.
%b ==> c.



main(_) :- writeln(hi), 
  ((9 + 4) + 3) ==> A, writeln(A),
  ((9 + 4) + 3 + 10 + 12) === (12 + 10 + 3 + 4 + 9), writeln(success).
  
  
  %foo(foo(a)) ==> E, writeln(E),
  %b ==> C, writeln(C).

% Any ground ordering is fine.

% https://www.swi-prolog.org/pldoc/man?predicate=term_size/2 includes sharing. interesting
% term_order(T,T2) :- term_size(T,S), term_size(T2,S2), 
%      S < S2 -> 
%        true 
%      ;  T =.. Tl, T1 =.. Tl2, lex(term_order, Tl1, Tl2) 

A cure possibly for non confluence.

evaluation relations

eval()


It is more useful to track by final position rather than by initial query term. Knuth bendix completion. With respect to a data type of ground rewrite rules, instantiate a pattern

match(foo(A), [foo(foo(a)) -> a]) :- 

match(Pat, Pat, _).
match(foo(Pat), foo(T), [ -> foo(T)]) :- match(Pat, []

Small Step Semantics

:- initialization(main,main).

value(true).
value(false).
value(zero).
value(succ(V)) :- value(V).

eval(A,A) :- value(A). % not step(A,B). Or put cut here.
eval(A,C) :- step(A,B), eval(B,C).
      % step(A,B) -> eval(B,C) ; done.
step(pred(zero),zero).
step(pred(succ(T)),T).
step(succ(T), succ(T1)) :- step(T,T1).
step(pred(T), pred(T1)) :- step(T,T1).
step(iszero(zero), true).
step(iszero(succ(_)), false).
step(ite(true,T,_), T).
step(ite(false,_,E), E).
step(ite(C,T,E), ite(C1,T,E)) :- step(C,C1).

hastype(zero, nat).
hastype(succ(T), nat) :- hastype(T,nat).
hastype(pred(T), nat) :- hastype(T,nat).
hastype(iszero(T), bool) :- hastype(T,nat).
hastype(ite(C,T,E), Ty) :- hastype(T,Ty), hastype(E,Ty), hastype(C,bool).

main(_) :- eval(pred(pred(succ(succ(succ(zero))))), V), writeln(V).

If we go metainterpreter on this, we can evaluate with respect to a list or set of rewrite rules.

step(apply(F,X), )

Hoare Logic

https://softwarefoundations.cis.upenn.edu/plf-current/Hoare.html


hoare(P, skip, P).
hoare(P, seq(C1,C2), Q) :- hoare(P, c1, R), hoare(R, c2, Q).
% hoare(P, set(X,V), Q) :- subst(P,X,V,Q). something like this. Not declarative.
% hoare(P, while(C,Inv,B), Q) , a good place for inductive logic programming? 
hoare(P, ite(C,T,E), Q) :- hoare(and(C,P), T, Q), hoare(and(not(C), P), E, Q).

% consequence, We may weaken the post, or strengthen the pre
% a use case for subsumption?
% hoare(P, S, Q) :- hoare(P1, S, Q), P => P1.
% hoare(P, S, Q) :- hoare(P, S, Q1), Q1 => Q.
% or make it same rule.

% weakest precondition is reasonable use of -++ mode. Or maybe +++ mode.

Bunched Logic / Separation

https://softwarefoundations.cis.upenn.edu/slf-current/toc.html

Model Checking

Set Graphs VI Logic Programming and Bisimulation Bisimulation co-LP

Typeclass

implements(clone,life(A,T))

impl(tostring, life(A,i64)).
impl(tostring, vec(T)) :- impl(tostring, T).

% associate what trait owns fields
traitfield(tostring, tostring).

dotcall( foo, tostring ) :- type(foo, T), traitfield(Trait, tostring), impl(Trant, T). 
% dotcall(TEnv, foo, tostring)




% annotate everything with lifetimes?

% https://stevedonovan.github.io/rustifications/2018/09/08/common-rust-traits.html
% display, default, format, clone,

% sized
% borrow


% from to

% iterator

Typeclass are “functions” from types to instances. Functions in prolog are represented as relations. This is proof threading.


eq(bool,
  [ eq(bool,true,true) = true, 
    eq(bool,false,false) = true,  
    eq(bool,true,false) = false,
    eq(bool, false,true) = false ]).

eq(pair(A,B), [eq((A,B),(XA,XB),(YA,YB)) = and(eq(A,XA,YA), eq(B,XB,YB))]) :- eq(A,IA), eq(B,IB).

The equational represntation is a separate issue.

Use dicts https://www.swi-prolog.org/pldoc/man?section=bidicts for instance dictionaries?

instance(eq(bool),  { equal: [X,Y] >> X = Y ->  } ).

Type checking / inference

I probably have something like this somewhere else in my notes This has something of the flsvor of a declarative statement of the typing rules, readin :- as horizontal line. I’m not being careful about moding. Not that de bruijn isn’t really necessary in the type checking process. Explicit names works ok because I’m not doing any substitution.

type(Env, nat(X), int).
type(Env, tt, unit).
type(Env, var(X), TX) :- lookup(Env, X, TX).
type(Env, lam(X, B), TA -> TB ) :- type([X - TA |Env], B, TB ).
type(Env, app(F,X), TB ) :- type(Env, F, TA -> TB ), type(Env, X, TA).

3-fix notation can be 2 2-fix operators

:- op(|-, fxx).
:- op(:, fxx).

G |- E : T :- .

Hmm. in the lambda prolog encdoing of this, thye use implication for var(X). So env is more of the form [ forall Env, type(E,X,TA) | Env] but actuall maybe we want to encode that only refining environments.

type(Env-E, lam(X,B), TX -> TB) :- type( [forall E1, type(Env-E1, X, TX) Env] - E, )

type(E,X,Tx) :- var(E), forallE_type(X,Tx). forallE_type(X,Tx) :- lookup(Env, forallE_type(X,Tx)) forallE_type(X,Tx) :- type(E,X,Tx).

fvar(X) onto the assertion environment list.

strcturally recursive copy_term is pure and can produce fresh vars. Kindo f it builds new vars by refining a fhresness source var(X)

[]

instead of putting it in the type(var(X)) rule.

Relationship to Hindley-Milner. We are using prolog uvars as type variables. This is tricky though.

https://github.com/bertas/hm_prolog/blob/master/src/type.pl This is not right to my eye for example. Using var(A) is cheating. https://www.swi-prolog.org/pack/list?p=type_check a package for typechecking prolog itself with HM http://kyagrd.github.io/tiper/ Executable Relational Specifications of Polymorphic Type Systems Using Prolog Let generalization Is generalize even good prolog?

type(Env, let(X,E,B), TB) :- type(Env, E, T1), generalize(T1,T2), type([X - T2 | Env], B, TB).

generalize(tVar(A), )
generalize(A -> B)

Refinement Types

Use CLP to build a refinement type system. https://arxiv.org/abs/2010.07763




Bussproofs printing


proof(bin(G, Rule, PX,PY)) --> proof(PX), proof(PY),
              ,"\RightLabel{", rule(Rule) ,"}", "\BinaryInfC{", sequent(G), "}".
proof(un(G, Rule, PX)) --> proof(PX),
              ,"\RightLabel{", rule(Rule) ,"}", "\UnaryInfC{", sequent(G), "}".
proof(ax(G, Rule)) --> 
              ,"\RightLabel{", rule(Rule) ,"}", "\AxiomC{", sequent(G), "}".


height(ax(_,_), 0).
height(un(_,_,PX), N) :- N := NX+1, height(PX,NX).
height(bin(_,_,PX,PY), N) :- N := max(NX,NY)+1, height(PX,NX), height(PY,NY).




Categorical Prover


%sequent( Hyp, Conc, Var )

:- use_module(library(clpfd)).
%:- table prove/2.
:- use_module(library(solution_sequences)).
%:- op(600, xfy, i- ).

prove(S, ax(S, id)) :- S = (A > A).
prove(S, ax(S, fst)) :- S = (A /\ _B > A).
prove( A /\ B > B, ax(A /\ B > B, snd)).
prove( S, ax(S, inj1 )) :- S = (A > A \/ _B).
prove( S, ax(S, inj2 )) :- S = (B > _A \/ B).
prove( false > _ , initial ).
prove( _ > true  , terminal ).
prove( A > B /\ C, bin(A > B /\ C, pair, P1, P2)) :- prove(A > B, P1), prove(A > C, P2).
prove( A \/ B > C , bin(A \/ B > C, case, P1, P2)) :- prove( A > B, P1), prove( A > C, P2).
prove( A > C, bin(A > C, comp, P1, P2)) :- prove(A > B, P1), prove(B > C, P2).


height(ax(_,_), 1).
height(un(_,_,PX), N) :- N #> 1, N #= NX+1, height(PX,NX).
height(bin(_,_,PX,PY), N) :- N #> 1, N #= max(NX , NY) + 1, height(PX,NX), height(PY,NY).

% maybe explicilty taking proof steps off of a list. using length.
% use dcg for proof recording?


prove(A > A)       --> [id].
prove(A /\ _ > A)  --> [fst].
prove(_ /\ B > B)  --> [snd].
prove(A > A \/ _)  --> [inj1].
prove(B > _ \/ B)  --> [inj2].
prove(false > _)   --> [initial].
prove( _ > true)   --> [terminal].
prove(A > B /\ C)  --> [pair], prove(A > B),  prove(A > C).
prove(A \/ B > C)  --> [case], prove(A > C),  prove(B > C).
prove(A > C)       --> [comp], prove(A > B),  prove(B > C).

:- initialization(main).
%main :- format("hello world", []).
%main :- between(1, 10, N), height(Pf, N), writeln(Pf), prove( w /\ x /\ y /\ z > w, Pf ), print(Pf), halt.
main :- length(Pf, _), phrase(prove(w /\ x /\ y /\ z > w \/ y),Pf), print(Pf), halt.
main :- halt.

Yes, the dcg form is very pretty. It’s a writer monad of sorts. It is recording a minimal proof certificate that is reconstructable to the full thing pretty easily.

G --> [tag], D1, D2 Should be read as

  D1        D2
------------------ tag
       G

prove(Sig , A > B) --> { insert(A,Sig,Sig1) }, [weaken(A)], prove(Sig1, A > B).
prove(A > forall(X,B)) --> , prove(weak(X, A) > B).
prove(A > forall(X,B)) --> {insert(X,Sig,Sig2) }, prove(Sig1, A > B).
prove(Sig, forall(X,A) > ) --> , prove(weak(X, A) > B)

Maybe start with implication prove((A > (B > C) ) –> [curry], prove( A /\ B > C). prove((A /\ (A > B) > B ) –> [eval].

Catlog

Instruction selection

A model of Instruction selection can be viewed as related to a parsing problem. A parse is building a tree out of a linearized sequence. Instruction selection is finding a linear sequence of machine instructions to represent an expression or sequence of high level statements.

% isel(R0, select( ,A + B)) --> {B := }

isel(R0, lit(A))    --> [mov(R0, A)]. % {integer(A)},
isel(R0, A + A)     --> [lsl(R0, R1)], isel(R1, A).
isel(R0, A + B * C) --> [fma(R0, RA,RB, RC)], isel(), sel().
isel(R0, A + B)     --> [add(R0, R1, R2)], isel(R1, A), isel(R2, B).
isel(R0, A * B)     --> [mul(R0, R1, R2)], isel(R1, A), isel(R2, B).
%isel(R0, binop(Op, A, B))     --> [mul(R0, R1, R2)], isel(R1, A), isel(R2, B).

% note storage restrictions.
% isel(reg(R0), A + B)     --> [add(reg(R0), reg(R1), reg(R2))], isel(reg(R1), A), isel(reg(R2), B).
% spill?
% isel(R0, A) --> [mv(R0, R1)], isel(R1, A). %optional copy. Bad rule


% liveness.
% isel(LiveSet, ) --> [add(R0,R1,R2)], { Live1 = Live \ R0 + R1 + R2 } , isel(LiveSet)
% maybe use bitvector?
% No give varoables unique identifiers, skolem style? id(1, Expr), id(2, Expr), 

:- initialization(main,main).
main(_) :- phrase(isel(R0, lit(2) + lit(7)), Asm), print(Asm), halt.

Lambda

Higher Order Unification Embeddable nonimal unification Unification

unify(A,B) :- var(A), var(B)
subst(X,V,E,E1) :- 

maps from ground variables to prolog vars for interpreters

[x - X, y - Y, z - Z]
:- initialization(main,main).
main(_) :- T = [append([],A,A), append([X|Xs], Y, [X|Zs]) :- append(Xs,Y,Zs)],
      numbervars(T,0,End),
      length(Vs, End),
      writeln(Vs).




Can you make pure data structures that respect monotonicity. They ought to defer or freeze rather than make judgements on variables.

Unification

find(Env,K,) :- 
  lookup(Env,K,V),
  (V == $'var'(N) ->
   find(Env, V, )
   ; 
   
   )




union(Uf, X, Y) :- 
  lookup(Uf, X, Xv), lookup(Uf, Y, Yv), Xv = Yv.

find(Uf,X,V) :- lookup(Uf,X,Xv).

unify(Env,'T1,T2) :- 
  eval(T1,T11), eval(T1,T22),
unify(Env, term(F,Arg), term(F,Arg1)) :-
 maplist(unify(Env, Arg, Arg1)).
unify(Env, $VAR(N), T) :-
    eval(T,T1), lookup(Env, $VAR(N), T1).




%macro(ex(X,E)) :- 
  


set unification list based vs union based reps Set constraints {log} boolean unification

Theorem Proving

Last chapter of bratko. Building a resolution prover using asserta. PyRes?

% unify with occurs check? No. We don't need occurs check
theorem(Poslist,Neglist).
resolve(A-B,C-D) :- remove(Form, B,B1), remove(Form, C,C1), Q=A+C1, P=B1+D. % sort of a hash join

Leantap Jens Otten How to Build an Automated Theorem Prover. Invited Tutorial at TABLEAUX in London/UK

A simple version of this implemented in tau prolog Prdocues proofs translated to bussproofs latex

a pearl on SAT and SMT solving in prolog code

Backjumping is Exception Handling throw and catch implement backjumping for a sat solver

PRESS: PRolog Equation Solving System as described in the art of prolog book. also has an lp solver? Pretty interesting tehcniques. If a variable is in a single position, you can unfold the expression. 2. you can apply quadratic equations and stuff 3. Is it kind of a zipper to isolate a variable?

tarau Formula Transformers and Combinatorial Test Generators for Propositional Intuitionistic Theorem Provers

Dyckhoff g4ip

Interactive

Teyjus ACL2 analogy. ACL2 is modelling a pure lisp.

axiom(P) :- asserta(P).
qed(F,Prf) :- check(F, Prf), asserta(F).

main(_) :- 
  axiom(a),
  axiom(b),
  lemma(a(X) :- b(X))

% lexical scope. We could do this as a macro?
let(X, e, let(X, e,  X).

isabelle style

:- dynamic proven.


% "difference list" form of a proof fragment
[A] > (A | foo(x) )  

proven(edge12,  [] > edge(1,2)).
proven(edge23,  [] > edge(2,3)).
proven(trans, [A] > path(X,Z) :- edge(X,Y) :- A).
proven(trans, [A,B] > path(X,Z) :- edge(X,Y) :- A, path(Y,Z) :- B)

asserta(proven(myname, tree)).

Using asserta let’s us refresh the schematic variable in an axiom

Topics

Systems

50 years of prolog

  • Swi prolog - I think this is a good default choice.
  • ciao
  • sicstus requires commercial license manual
  • gnu prolog
  • scryrer
  • tau prolog - javascript web enabled
  • hprolog
  • binprolog
  • XSB - fancy tabling.
  • YAP
  • B-prolog
  • ECLiPSe - can talk to minizinc
  • Qu-prolog
  • trealla prolog interpreter in C. Not WAM based. tree-walking, structure-sharing, deep binding whatever that means

Relatives:

  • Minikanren
  • minizinc
  • picat
  • Mercury - mode declarations and type declarations. Allows reordering, but deals with IO using lionear types?
  • Curry - Haskell syntax like language.
  • Lambda prolog - teyjus, elpi, makam
  • Hilog
  • Godel
  • Hyprolog - abduction
  • ergo AI, flora2
  • guan https://github.com/microsoft/Guan c# prolog?

Ciao

ciao manual

main(_) :- format("hellow world").
main(_) :- print([1,2,3]),
  append(A,A,A), print(A)
.

assertions and auto documetation ciao and design philosphy ciaopp - preprocessor and veirfier? PLAI

SLD resolution

Interesting predicates

comparson and unification of terms

  • =@=, variant
  • ?=
  • ==
  • =
  • /= a weaker version of dif. Uses negation as fialure in kind of unsatisfactory way

  • =.. destructures a term

  • numbervar vanumber concretize terms.
  • gensym suggests using content based identifiers
  • termhash
  • variant_sha1 a cyrptographic hash? This is acceptable?

concurrent_forall

call_with_inference_limit

Imperative analogies

Unification variables are pointers. Unifying them is aliasing them. Unification is bidirectional pattern matching

Prolog predicates are like function calls. The fact that fields can be both input and output is like using output pointers in C. These pointers come in as arguments often.

Prolog as a Procedural Programming Language

  • Assign Once variables - because of this assignment can be symmetrical

We can abuse the precedure calls of a host languages. This is related to minikanren, but minkanren doesn’t use depth first search

Manual Prolog

Proplog is even weaker than datalog. Pure propositional variables with no arguments.

# We use sequencing as , and function calls as heads.
# We could envelope in try catch blocks for calls that don't work out

# a :- b, c.
def a():
  b()
  c()
# b :- c.
def b():
  c()
# c.
def c():
  pass

a() # succeeds.

For nondeterminism python generators are a decent guess. Need to undo unifications. Can do Cells without persay doing generic unification. Use python case statements or if then else.

# no path compression
# `is` is python reference equality
class BadUF(): # Cell
  def __init__(self): # *args
    self.parent = self
  def find(self):
    x = self
    while x.parent is not x: #  or x is not instance(Cell)
      x = x.parent
    return x
  def union(self,rhs):
    p = self.find()
    p2 = rhs.find()
    p.parent = p2
    return p2

x = BadUF()
y = BadUF()
z = BadUF()

x.union(y)
print(x.find() is y.find())
print(x.find() is z.find())

# a(Stuff) :- b(Stuff), c(Stuff).
# a(Stuff) :- c(Stuff)
def a(s):
  for s in b(s.copy()):
    for s in c(s.copy()):
      yield s
  for s in c(s.copy()):
    yield s

class Env(dict):
  def find():
  def __set__(self):
    #calls find
    #does unification.
  def __get__(self):
    #calls find.
  def fresh(self):
    self[len(self)] = len(self) 
    return len(self)

def append(x,y,z,s):
  s1 = s.copy()
  s1[x] = []
  s1[y] = s1[z]
  yield s1
  a, b, c = freshes(s, 3)
  s[x] = (a,b)
  s[z] = (a,c)
  yield from append(b,y,c,s)


Yeah. This amounts to a goofy minikanren where I’ve inlined combinators.

type goal = string (* Propsoitional *)
type goalstack = goal list
type state = goalstack list

type clause = goal * (goals list) (*  (goal list) list String.Map.t  *)



let rec sld_step prog = function
  | [] -> failwith "fail" (* nothing more to try *)
  | [] :: totry -> failwith "succeed" (* all goals cleared *)
  | ( (g :: gs) :: todo) -> 
            let bodies = lookup prog g (* Lookup also does unfication in real version *)
            let newgoals = List.map bodies ~f:(fun body -> body :: gs)
            sld_step prog (newgoals :: todo)

type table = 

let rec slg_step prog table resume_stack gen_stack
 = match resume_stack with
 | [] -> failwith "fail"
 | (cnode, solution) :: rs -> 


SLD resolution in stack form. You have a stack of goal sets. It isn’t that bad because you’re actually sharing a lot of repeated goals.

SLD in tree search form? The stack is some kind of manifestation of the tree. We could perhaps store the tree as a zipper instead

Abstract Machines / Implementation

https://www.cl.cam.ac.uk/~am21/research/funnel/prolog.c C++ interpreter https://github.com/Naereen/Tiny-Prolog-in-OCaml-OneFile ocaml interp https://t3x.org/lisp64k/prolog.html lisp interp

Warren’s Abstract Machine: A Tutorial Reconstruction.

Paul Tarau showing an interesting compilation strategy. https://popl21.sigplan.org/details/PADL-2021-papers/5/A-Family-of-Unification-oblivious-Program-Transformations-and-Their-Applications https://github.com/ptarau/LogicTransformers

binprolog. Translate to binary relations

structure copying vs structure sharing

Term indexing

Original Dec-10 prolog paper

discussion on scryer forum

BAM - berkeley abstract machine Can Logic Programming Execute as Fast as Imperative Programming? Peter Van Yoy’s thesis http://lambda-the-ultimate.org/classic/message1618.html#11082 Andrew Taylor’s thesis - high performance prolog implementation

comparsion of WAM and ZIP

VAM

TOAM - tree oriented abstract machine

global optimization in a prolog compiler for TOAM

Modes

In some ideal world, it’s great if every predicate is reversible, but it isn’t the case. Different variable positions can be used in different ways, input, output, both. They can also have total functional character (exactly one answer), partial functional character (one or zero), of nondeterminisitc (many answers). Modes are required to use predicates correctly. Annotating them may allow the compiler to be more efficient. The compiler may infer them. They are conceptually interesting also in theorem proving contexts. See bidirectional typing.

mercury ciao prolog

Single Sided Unifivcation

picat pattern matching

https://www.swi-prolog.org/pldoc/man?section=ssu

Maybe not even that complicated? Claims it a fairly simple macro. Use ==



DIY Prolog

a python interpreter

https://curiosity-driven.org/prolog-interpreter

Prolog and es6 generators https://curiosity-driven.org/prolog-interpreter

https://news.ycombinator.com/item?id=2152570

https://cs.stackexchange.com/questions/6618/how-to-implement-a-prolog-interpreter-in-a-purely-functional-language

andrej bauer’s pl zoo prolog

1985 pascal design and implementation of prolog interpreter

Verification

Modules

meta interpreters

power of prolog - A Couple of Meta-interpreters in Prolog video

metapredicates call/N in principle could be implemented in prolog itself.

:- initialization(main,main).
foo(7).
main(_) :- call(foo(X)), print(X).

Things you can do:

  • tracing parameter “proof”
  • Change evaluation order of goals or clauses
  • Change unification - occurs check, attributed vars
  • continuations
  • fair interleaving
  • abstract interpetation
  • partial evaluation

absorption and reification

:- initialization(main,main).
% surprisingly raw Goal is implicit call/1
mi0(Goal) :- Goal.

main(_) :- mi0(mi0(append([12],[42],X))), print(X).

Advantages to not using prolog syntax.

metacircular interpeter - interpreter can interpret self https://news.ycombinator.com/item?id=19283803

:- initialization(main,main).
mi([]). % empty goal stack
mi([G,Gs]) :- clause_(G,Body), mi(Body), mi(Gs).

% define rules of metainterpreter itself
clause_(mi([]), []).
clause_(mi([G | Gs]), [clause_(G,Body), mi(Body), mi(Gs)]).

% what does this mean? This is a weird one.
clause_(clause_(G,Body), []) :- clause_(G,Body).

Prolog meta-interpreter with difference lists

:- initialization(main,main).
mi([]).
mi([G|Gs]) :-
        mi_clause(G, Rest, Gs),
        mi(Rest).


Same idea but the body of a clause is represented as a difference list in clause_

clause/2 is a way to get the rules of a clause.


:- initialization(main,main).
clause_(append([],Y,Y), []).
clause_([X | Xs],Ys, [X | Zs]), [append(Xs, Ys, Zs)]).

%mi(Sig, all(X, Body)) :- X = fvar(Sig), Sig1 is Sig + 1, mi(Sig1, Body). 
%mi(all(X, Body)) :- gensym(Sig), X = fvar(Sig), mi(Sig1, Body).
%mi(P, S, A => Body) :- assertz(A), mi(Body), retract(A).
%mi(Sig, ex(X, Body)) :- X = fvar(Sig), Sig1 is Sig + 1, mi(Sig1, Body).  

mi(P,S,pi(X, Body)) :- gensym(fvar, Y), X = fvar(Y), mi(P, [X|S], Body). % could use numbering.
mi(P,S,D => G) :- mi([D|P],S,G).
mi(P,S,sigma(X,G)) :- mi(P,S,G), forall( subterm(X,fvar(Y)), member(fvar(Y),S)  ) 
mi(P,S,+G) :- copy_term(E,P,E1,P1), do_goal(P,S,P1, G) % no this copy term isn't right. copy term hurts. Ah copy_term/4

do_goal(P,S, [G :- B|Gs], G) :- mi(P,S,B).
do_goal(P,S, [_|Gs], G) :- do_goal(P,S,Gs,G).

This is wrong.


:- initialization(main,main).
:- use_module(library(occurs)).
:- op(1200, xfx, \).
:- op(900, xfy, =>).
% copy_term?
pi(X \ Body) :- gensym(fvar,Y), X = fvar(Y), fvar(Y) => Body.
A => Body :- asserta(A), Body, retract(A).
% https://www.swi-prolog.org/pldoc/man?section=occurs
% https://www.swi-prolog.org/pldoc/man?section=forall2
ex(X \ Body) :- Body, forall(fvar(Y), free_of_var(fvar(Y), X)). % this is checking the wrong thing right?
A /\ B :- A, B.
A \/ B :- A; B.

% seperate the notions?
goal_and().
prog_and().

sterile(J) :- pi(X \ (bug(X) => (in(X,J) => dead(X)))).
dead(B) :- ex(J \ (heated(J) /\ in(B,J) /\ bug(B))).
heated(j).

main(_) :- sterile(j), \+ bug(_B), \+ fvar(_X).

Hmm. when I asserta with a variable, I’m asserting the universal thing, not something tied to the variables in scope. That seems really bad.

Oh wow. Also asserta doesn’t cleanup on fail. asserta(bar(7)), fail. still leaves bar(7) asserted

chr as the store of new facts.

foo_q(X,Y,Z) \ foo(X,Y,Z) <=> true. foo_q(X,Y,Z) <=> { foo(x,y,z)}. foo_q(X,Y,Z) <=> {fail}.

or turn it around

foo(X) :- foo_q_chr(X). % see if chr query succeeds % else regular other clauses.

CHR +

How can I automatically associate chr with

assert_chr(foo, Args) retract_chr(foo, Args), assert(foo, Args) <=> true. assert_chr(T,Args) \ query_chr(T, Args) <=> true. % hmm. does this unify the stuff in the assert, destroying it’s universal quantification? % https://www.swi-prolog.org/pldoc/man?section=compare assert(T) \ query(T1) <=> { unifiable(T1, T), copy_term(T, Tc), Tc = T1 }. This might fix that problem. Very ugly.

Does this have proper backtracking semantics though? No. Probably not. I don’t think chr makes choice points. If unificable fails, great it continutes, but if something later fails it won’t backtrack. We could try and control which rule applies via some kind of number, a choice point placed before we call chr_query. Getting quite arcane. assert(T) \ query(T1) <=> { unifiable(T1, T), copy_term(T, Tc), Tc = T1 ; fail}.

This is insanity. Wow, shallowly embedding harrop clauses in prolog is hard/impossible.

Naw, why even seaprate out the args? Maybe better indexing, but screw it for now. assert_chr(foo(Args)). I might be building a prolog (meta?) interpreter in chr.

locally nameless encoding of exists. do opening with X. no copy_term of program clauses.

open(evar(N),N,X,X).
open(evar(M),N,X,evar(M)) :- dif(N,M).

close(). Hmm. Maybe we never need to close.

mi(S, P, all(B)) :- gensym(X), open(B,0,fvar(X), B1), mi([fvar(X)|S], P, B1).
mi(S, P, ex(B)) :- open(B,0,X,B1), mi(S,P,B1), extrude_check(S,X).
mi(S, P, D => G) :- mi(S,[D|P], G).
mi(S, P, +G) :- 
mi(S, P, +G) :- call(G). %native call

% de bruijn the named rep. This is close I guess.
bruijn(T,T1) :- number_var(T, T2), bruijn([], T2,T1)
bruijn(Vars, var(N), var(I)) :- index(Vars, N, I).
bruijn(Vars, ex(X,B), ex(T1)) :- bruijn([X|Vars], B, T1). 

mi(Sig, Env, Prog, all(B)) :- gensym(X), mi([X | Sig], [X | Env], Prog, B).
% if I distinguish avar(N) from evar(N), Sig is already an Env

mi(Sig, Env, Prog, var(N)) :- lookup(N,Env,V). % and V goes where?
mi(Sig, Env, Prog, +G ) :- subst(G,Env,Sig).

% do subsitution lazily instead of eagerly.


normalization by evaluation. Hmm. What if I did use the lambda library https://www.swi-prolog.org/pldoc/man?section=yall

all(L) :- gensym(X), call(L, fvar(X)).
ex(L) :- call(L, Y), extrude_check(Y).
impl(D,G) :-  call(asserta(D), G) -> retract(D) ; retract(D), fail. 
% ok but asserta loses track of current unification variables.
% also problem if G produces multiple answers?
% label_vars(D,D1). asserta(D :- subst(D1, )
% pack it, then unpack it on other side.
% number_vars, term_variables
% https://www.cs.cmu.edu/~fp/papers/ilps91.pdf mentions the assert retract encoding. Mentions that variables are nasty.
right(_,_,true).
right(S,P,and(B1,B2)) :- right(S,P,B1), right(S,P,B2).
right(S,P,or(B1,_B2)) :- right(S,P,B1).
right(S,P,or(_B1,B2)) :- right(S,P,B2).
right(S,P,impl(B1,B2)) :- right(S,[B1|P], B2).
right(S,P,ex(_X,B)) :- right(S,P,B). % maybe
right(S,P,all(X,B)) :- gensym(X), right([X|S],P,B). % maybe
right(S,P,+A) :- member(D,P), left(S,P,D,A). %decide

% alternate name: clause deonstruct, focused.
left(_,_,A,A).
left(S,P,impl(G,D), A) :- left(S,P,D,A), right(S,P,G).
left(S,P,and(D1,_D2), A) :- left(S,P,D1,A).
left(S,P,and(_D1,D2), A) :- left(S,P,D2,A).
left(S,P,all(_X,D), A) :- left(S,P,D,A). % maybe



Before unify throw signal (attrbuted var). WHy this is wrong: Generally, trying to use the wrong kind of variable is a road strife with peril.

  • When we use P, we destroy the universal quantifications within. If we copy, we destroy the connection to the existential variables. This is also one of the problems of using asserta. copy_term allows you to add variables to be renamed, but not the ones to be shared. This is the opposite of he behavior we want

It is somewhat natural to expect that lambda terms might be useful. https://www.swi-prolog.org/pldoc/man?section=yall https://www.swi-prolog.org/pack/list?p=lambda https://swi-prolog.discourse.group/t/yall-lambda-arguments-visibility/5112/2 This example is evidence that the implementation of yall may not scope as lambdas should.

Let us go under the (quite possibly faulty) assumption that the story on lambdas is clear in prolog. One can manually lambda lift, defunctionalize, closure convert if you want to.

Lambda’s in yall only produce goals, so far as I can tell. So if you want them to produce terms, that needs to come in as another paramter by convention. This feels a bit like continuation passing or output parameter passing.

:- use_module(library(yall)).
:- use_module(library(apply)).
:- use_module(library(apply_macros)).
:- initialization(main,main).
:- use_module(library(occurs)).

extrude_check(S,T) :- writeln([ex_check, S, T]), forall((sub_term(X, T), ground(X), X = fvar(Y)), member(Y,S)). 

right(_,_,true).
right(S,P,and(B1,B2)) :- right(S,P,B1), right(S,P,B2).
right(S,P,or(B1,_B2)) :- right(S,P,B1).
right(S,P,or(_B1,B2)) :- right(S,P,B2).
right(S,P,impl(B1,B2)) :- right(S,[B1|P],B2).
right(S,P,ex(L)) :- call(L,X,G), right(S,P,G), writeln([ex,X]), extrude_check(S,X).
right(S,P,all(L)) :- writeln(all), gensym(v,X), call(L,fvar(X),G), right([X|S],P,G), writeln([exit_all, G]).
right(S,P,+A) :- member(D,P), writeln([pick,P, D,A]), left(S,P,D,A), writeln([D,A]). %decide

% alternate name: clause deonstruct, focused.
left(_,_,+A,A).
left(S,P,impl(G,D), A) :- left(S,P,D,A), right(S,P,G).
left(S,P,and(D1,_D2), A) :- left(S,P,D1,A).
left(S,P,and(_D1,D2), A) :- left(S,P,D2,A).
left(S,P,all(L), A) :- call(L,X,D), left(S,P,D,A). % maybe


%main(_) :- P = [
%  all( [J, T] >> T = impl(all([X,T1] >> T1 = impl(and(bug(x), in(x,J)), dead(x)), sterile(J))),
%  all( [B, T] >> T = impl(ex([J,T1] >> T1 = and(headed(J), and(in(B,J), bug(B)) ) , dead(B))),
%  heated(j) ]


foo(E,G) :- G = all(bar(E)).
bar(E,X,G) :- G = impl(+foo(X), +foo(E)). 
% ;re
query(Q) :- Q = ex([H2,J2] >> (J2 = all({H2} / [Q2,G2] >> (G2 = impl(+foo(Q2),+foo(H2)))))).

main(_) :- 
  P = [
  all([L,T] >> (T = +append([], L, L))),
  all([H,T] >> (T = all([L1, T1] >> 
               (T1 = all([L2,T2] >> 
               (T2 = all([L3, T3] >> 
               (T3 = 
        impl(+append(L1,L2,L3),
             +append( [ H | L1], L2, [H | L3]))))))))))
  ], right([], P, +append([a,D], [c], Z)), writeln(Z),
  right([],[],true), writeln(pass),
  right([],[+a], +a), writeln(pass2),
  right([],[+a,+b], +b), writeln(pass3),
  right([],[all([X,T] >> (T = +foo(X)))],+foo(a)), writeln(pass4)
  , right([], [+append(nil, nil, nil)], +append(nil, nil, nil)) , writeln(pass5)
  , right([], P, +append([], [], X)) , writeln(X), writeln(pass6),

  right([], [], impl(+a,+A)), writeln(A),
  right([], [], impl(and(+a,+b),+b)),
  right([], [], impl(and(+a,+b),+b)),
  right([], [], all([Q,G] >> (G = impl(+foo(Q),+foo(Q))))),
  %\+ right([], [], all([Q,G] >> (G = impl(+foo(Q),+foo(a))))),
  right([], [], all({F}/[Q1,G1] >> (G1 = impl(+foo(Q1),+foo(F))))), writeln(F),

  %right([], [], ex([H1,J1] >> (J1 = impl(+foo(H1),+foo(H1))))),
  writeln(pass7),!,
  %right([], [], ex([H2,J2] >> (J2 = all({H2} / [Q2,G2] >> (G2 = impl(+foo(Q2),+foo(H2))))))),
  writeln(pass8),
  %right([],[],ex(foo)),
  writeln(pass9),
  %query(Q7),
  %right([],[],Q7),
  right([],[],ex([X,Q]>>(Q=all([Y,P]>>(P=impl(+d(X),+d(Y))))))).





  %right([],[],[], append(nil,nil,nil)), print(passed)
.


Drinker’s paradox example. This shouldn’t pass


:- use_module(library(yall)).
:- use_module(library(apply)).
:- use_module(library(apply_macros)).
:- initialization(main,main).
:- use_module(library(occurs)).

extrude_check(S,T) :- writeln([ex_check, S, T]), forall((sub_term(X, T), ground(X), X = fvar(Y)), member(Y,S)). 

right(_,_,true).
right(S,P,and(B1,B2)) :- right(S,P,B1), right(S,P,B2).
right(S,P,or(B1,_B2)) :- right(S,P,B1).
right(S,P,or(_B1,B2)) :- right(S,P,B2).
right(S,P,impl(B1,B2)) :- right(S,[B1|P],B2).
right(S,P,ex(L)) :- call(L,X,G), right(S,P,G), writeln([ex,X]), extrude_check(S,X).
right(S,P,all(L)) :- writeln(all), gensym(v,X), call(L,fvar(X),G), right([X|S],P,G), writeln([exit_all, G]).
right(S,P,+A) :- member(D,P), writeln([pick,P, D,A]), left(S,P,D,A), writeln([D,A]). %decide

% alternate name: clause deonstruct, focused.
left(_,_,+A,A).
left(S,P,impl(G,D), A) :- left(S,P,D,A), right(S,P,G).
left(S,P,and(D1,_D2), A) :- left(S,P,D1,A).
left(S,P,and(_D1,D2), A) :- left(S,P,D2,A).
left(S,P,all(L), A) :- call(L,X,D), left(S,P,D,A). % maybe


main(_) :- right([],[],ex([X,Q]>>(Q=all({X} / [Y,P]>>(P=impl(+d(X),+d(Y))))))).

https://github.com/SWI-Prolog/swipl-devel/blob/master/library/yall.pl yall implementation copy_term_nat(Free+Lambda, Free+LambdaCopy), is a way to copy_term but also share variables in the Free list. We copy and then unify. Nice.

lambda_functor(Term, Functor) :-
    copy_term_nat(Term, Copy),
    variant_sha1(Copy, Functor0),
    atom_concat('__aux_yall_', Functor0, Functor).

This is interesting. Determinsitically creating a functor name based on the sha1 of the context

We could also implement scoping the usual way, not trying to pull a fast one by reusing prolog mechanisms. de bruijn indices or locally nameless This brings it’s own peril. We are treated prolog as a functional or imperative language here. Oleg has an example where he uses pure constructs. If we want to make this generic ndexing dif/2

:- use_module(library(reif)).
:- use_module(library(terms)).
:- use_module(library(clpfd)).
:- initialization(main,main).
open_(X,N,T,T1) :- if_(T = lam(B), 
                      (N1 #= N + 1, T1=lam(T2), open_(X,N1,B, T2)) ,
                   if_( T = var(N), 
                       T1 = X,
                   mapargs(open_(X,N), T, T1) % https://www.swi-prolog.org/pldoc/doc_for?object=mapargs/3
                   )).
%close() maybe unecesaary?
main(_) :- writeln(hi), open_(a, 0, lam(var(1)), T), writeln(T).

We don’t need to open persay, since we mostly/entirely just need evaluation? It’s the same as open except doing it eagerly vs lazily / batching the traversal.

right(T1 = T2) --> {unify(T1,T2)}.
eval(S,E,evar(N), V) :- lookup(E,N,V).
eval(S,E,avar(N), V) :- lookup(S,N,V).
eval(S, E, F, R) :- 
unify(S,E,T1,T2) :-
unify(S,E,apply(F,Args), )

% hmm. keeeping E and S vars on same stack is useful because that says what is in scope

We still have a scope extrusion check. An exmample: ???? Existentials should really be being filled in with terms that exist at the time that binder is being opened. The unfication variable is a standin, delaying picking a term so that we can deduce at it should have been, but it is not protected from unifying with forall quantified variables that didn’t exist at the time of it’s creation. These would not have been legitimate concrete terms

By analogy with the occurs_check, we might choose to just not worry about this and hope for some kind of systems magic flag to come along later. Attributed variables might also enable pushing unifiation The interpreter could also insist on treating = specially. It is difficult to avoid accidentally using native prolog unfication both in the interpreter, but also in the interpretted programs.

term_lam(X \ B) :- [G|X] \ (G = B). % does this make sense? Fishy. term_lam = \\ notation

So a natural question is how can we make this interpreter shallower? We are reusing prolog’s search mechanisms and unification mechanisms. So that’s nice.

Implication is sticky though. One might be tempted to use asserta which make dynamic databases. Naive use of this is not right

  • asserta considers all unification variables to be universals. It breaks the correlation of the program database with the goal
  • retracting the implication once you leave the scope of it is tricky. Yes retract exists. But it is not clear to me

assert and retract are manipulating a global database store. This is a leaky abstraction and somewhat doomed to failure IMO. Imagine trying to implement function calls using a global memory. Ok I guess you could do it by maintaining your own stack.

By analogy with file handles, we may think some kind of bracketing expression may help. setup_call_cleanup(asserta(D), G, retract(D)) https://www.swi-prolog.org/pldoc/doc_for?object=setup_call_cleanup/3 No. This doesn’t retract at the right point. The search may exit the implication but then return to it on backtracking. This doesn’t retract until G won’t be called anymore.

Ok, well perhaps we never call retract, but instead key on an identifier of what branch we are in. And to deal with the sharing problem, we could replace all E vars in the program clause with keys that access them from a store. It really feels like we’re plugging up hole after hole on a leaky ship.

DCGs can be used to clean up the state carrying a bit, since so may rules do not involve state. https://www.swi-prolog.org/pack/list?p=edcg https://www.metalevel.at/prolog/dcg

:- use_module(library(yall)).
:- use_module(library(apply)).
:- use_module(library(apply_macros)).

right(true).
right(and(B1,B2)) --> right(B1), right(B2).
right(or(B1,_B2)) --> right(B1).
right(or(_B1,B2)) --> right(B2).
right(impl(B1,B2)), [s(_,_,P)] --> [s(_,_,[B1|P])], right(B2).
right(ex(L)), [s(_,E,_)] --> [s(_,[X|E],_)], {call(L,X,G)}, right(S,P,[X|E],G).
right(all(L)), [s(S,_,_)] --> [s([X|S],_,_)], {gensym(X), call(L,X,G)}, right([X|S],P,E,G).
right(+A) --> member(D,P), left(D,A). %decide

left(A,A).
left(impl(G,D), A) --> left(D,A), right(G).
left(and(D1,_D2), A) --> left(D1,A).
left(and(_D1,D2), A) --> left(D2,A).
left(all(L), A) --> {call(L,X,D)}, left(D,A).

Higher order unification is a separate issue. If we are already overloading unification, it is probably not that hard to treat. Miller pattern unification is a subcase of higher order unification that is easy. The idea is to restrict higher order unification to it’s bones.

:- use_module(library(intercept)).
:- initialization(main,main).

%https://www.swi-prolog.org/pldoc/man?section=intercept

%right(+A) :- send_signal(goal(A)).
% hmm, so that send signal is still in scope of the current intercept it seems like?
%right(impl(D,G)) :- intercept(right(G),goal(A), (writeln(A), D=A;send_signal(goal(A)))).

% implicit contxt being parlayed.
left(D,A) :- D = A.
right(+A) :- send_signal(prog(P)), member(D,P), left(D,A). 
right(impl(D,G)) :- send_signal(prog(P)), intercept(right(G), prog(P1), P1 = [D | P]).

run(G) :-  intercept(right(G), prog([]), true).
 % intercept(right(G),goal(A), (writeln(A), D=A)).
main(_) :-  intercept(send_signal(x), X, writeln(X)),
            %intercept((right(impl(foo(c), impl(foo(b), +foo(A)))), writeln(A)), prog([]), true).
            run(right(impl(foo(c), impl(foo(b), +foo(A))))), writeln(A).

So these techniques enable implicit context in different ways. Could receive database per head symbol

:- use_module(library(intercept)).
:- use_module(library(yall)).
:- use_module(library(apply)).
:- initialization(main,main).

biz(X) :- get_ctx(facts(P)), member(D,P), D = biz(X). % for non unversal facts this works. call(D,bar(X))
%biz(X) :- X = foo.

% hmm. I want to reenter the continuation mutiple times though? Is that ok?
% ok, so if I just send back the 


bar(X) :- get_ctx(facts(P)), member(D,P), D = bar(X). % for non unversal facts this works. call(D,bar(X))

bar(X) :- get_ctx(clauses(P)), writeln(P), writeln(X), member(D,P), call(D,bar(X)). % for non unversal facts this works. call(D,bar(X))


% bar(a).


% essentially, reader monad

foo(X) :- send_signal(clauses(P)), member(D,P), call(D,foo(X)).

% mayb send_silent_signal is useful here
% or perhaps catch the error of a missing prog(P).
% use intercept/4 just to make sure variables in P don't get screwed up?
%impl(D,G) :- get_ctx(facts(P)), intercept(G, facts(P1), P1=[D|P]).

impl(D,G) :- get_ctx(facts(P)), intercept(G, facts(P1), =(P1), [D|P]).

climpl(D,G) :- get_ctx(clauses(P)), intercept(G, clauses(P1), =(P1), [D|P]).


get_ctx(facts(P)) :- catch(send_signal(facts(P)), error(unintercepted_signal(facts(_4104)),_4096), P = []).
get_ctx(clauses(P)) :- catch(send_signal(clauses(P)), error(unintercepted_signal(clauses(_4104)),_4096), P = []).

% could maintan seprate lists per functor, but what's the point?
setup_db(G) :- intercept(G,prog(P),P=[]).

buz :- writeln(ap), bar(a), send_signal(facts(P)), writeln([facts,P]), bar(b), writeln(succed_b).

bar_clause(X, G) :- G = bar(X).
biz_clause(G) :- G = bar(X), writeln(X), biz(X).
main(_) :- 
  % yes it traverses to the outer intercept
 %intercept(intercept(send_signal(foo), bar, writeln(bar)), foo, writeln(foo)),
% setup_db((
  %impl(bar(b), impl(bar(c), bar(A))), A \= c, writeln(A), bar(a),

  % \+ impl(bar(_F), (bar(a), bar(b))) % should fail because F needs to be a and b
  %, climpl([G] >> G = bar(a), bar(a))
  %, climpl([G] >> G = bar(a), bar(a))
  %, climpl(=(bar(a)), bar(a)) % this is nice. What if I wanted a body though?
  %, climpl([G] >> (G = bar(X), biz(X)), impl(biz(a), bar(a)))
   climpl(biz_clause, bar(a)) % impl blocks climpl. Hmm. So non interferencing signals don't bubble.
   % Do I need to do  if_( = facts(), P1 = P(),  send_signal()) explicit signal repropgation?
   % intercept(G, Signal, Signal = fact(P) -> doit ; send_signal(Signal))
   % seems very non compostional to require this. Can this be right?
   % Also could use [fact(), clause(), fact()] rep of P
   % or implement impl(D,G) :- climpl(=(D), G). 


  .
% )).
 %intercept((biz(X), send_signal(bar)), bar, X=bar), writeln(X).

% exists is uninterestnig except for scope check.
%pi(L) :- gensym(fvar,X), call(L,X).
pi(L) :- send_signal(fvar(N)), N1 #= N + 1, intercept(call(L, fvar(N1)), fvar(N1), true).

% This seems kind of useful for abductive logic programming. Maybe they have soome tricks?

% more point free combinator tricks rather than using []
% That will be less error prone.
:- use_module(library(intercept)).
:- initialization(main,main).

% add this clause to mark bar/1 as "scoped dynamic predicate"
bar(X) :- get_ctx(facts(P)), member(D,P), D = bar(X).
% also can have regular global clauses
bar(X) :- biz(X).
bar(c). 

impl(D,G) :- get_ctx(facts(P)), intercept(G, facts(P1), =(P1), [D|P]).
get_ctx(facts(P)) :- catch(send_signal(facts(P)), error(unintercepted_signal(facts(_)),_), P = []).

main(_) :- impl(bar(a), (bar(X), impl(bar(b), bar(Y)))), writeln(X), writeln(Y).


% no it does pass signals through.

%main(_) :- intercept(intercept(send_signal(foo(X)), bar(X), writeln(bar)), foo(X), writeln(foo)).

They don’t cover (maybe) the problem of keeping universal variables universal but existentials linked

%:- initialization(main,main).

right(+A) :- shift(goal(A)).
right(impl(D,G)) :- reset(right(G), goal(A), Cont), 
  (Cont == 0 ->
  true;
  (A=D ; shift(goal(A))), call(Cont)).

main(_) :-  reset(right(impl(foo(c), impl(foo(b), +foo(A)))), goal(_), Cont), 
(Cont == 0 -> writeln(A) ; fail).

Delimitted Continuations

Continuations are a reification of a call stack. The call stack in prolog is a goal stack. When you

swipl manual entry schrivers et al disjunctive delimited control Tabling as a Library with Delimited Control https://arxiv.org/pdf/1507.08087.pdf intercept librarypreferred because it plays nicer with cuts and things. I don’t even understand what the difference is in contract between this and reset/shift.

  • effect handlers - implicit state
  • definite clause grammars
  • coroutines

DCGs implemented with delimited continuations Conitnuations in prolog exmaple usage in prolog

Tabling

Tabling is a kind of auto-memoization. It can make queries terminate that no longer did, and the caching can speed others up.

Tabling as a Library with Delimited Control https://arxiv.org/pdf/1507.08087.pdf Swi prolog manual Programming in Tabled Prolog - Warren Tabling as Concurrent Machines

Tabling with Answer Subsumption: Implementation, Applications and Performance

Tabling with Sound Answer Subsumption Fixing non-determinsim

Linear Tabling Strategies and Optimizations - kind of a combo of iterative deepening and tabling. Iterate top level goal and memoize true solutions. Just kill it on looping. Eventually, you’ll stabilize

:-table p/2.
p(X,Y):-p(X,Z),e(Z,Y).
p(X,Y):-e(X,Y).
The transformed predicate is as follows:
p(X,Y):-p(X,Z),e(Z,Y),memo(p(X,Y)).
p(X,Y):-e(X,Y),memo(p(X,Y)).
p(X,Y):-check_completion(p(X,Y)).

calls memo on success. This is probably not how tabling is implemented in SWI, but it’s interestign. Hmm. I don’t really see how al this state tracking is possible/easy as user prolog.

A simple scheme for implementing tabled logic programming systems based on dynamic reordering of alternatives. DRA. Simplifying dynamic programming via mode-directed tabling Mode-directed tabling for dynamic programming, machine learning, and constraint solving

OLDT vs SLG

Memoization of prolog is two different issues: Dealing with metavariables and dealing with search WHat is the search space of prolog? Nodes are goal stacks, edges are rules. Using DFS to traverse a graph goes out of control, even a DAG does lots of retraversal. Memoizatin of prolog goals is differet to purely functional moemization. The metavariables matter and are in a sense return parameters a la C. So you have to memoize different metavairable paramters as different things, because they are in a sense different functions.

Fib and Fact are terminating even if exponentially slow. It’s hard to think of a simple high school math example that might be defined in a potentially non terminating way. THe edge-path query is an easy prolog example.

:- table path/2.
edge(1,2).
edge(2,3).
edge(3,1).
edge(3,4).

path(X,Y) :- edge(X,Y).
path(X,Z) :- edge(X,Y), path(Y,Z).

:- initialization(main,main).
main(_) :- path(1,4).
:- use_module(library(tabling)).
:- table a/0.

a :- a.

https://www.swi-prolog.org/pldoc/man?section=tabling-mode-directed. Mode difreced tabling. Can take minimum answer, first answer, etc. Good for dynamic programming

Longest common subsequence

%:- use_module(library(tabling)).
:- table lcs(_,_,lattice(longest/3)).


lcs([], _Ys, []).
lcs(_Xs, [], []).
lcs([X|XS], [X|YS], [X|Zs]) :- lcs(XS,YS,Zs).
lcs([X|XS], [Y|YS], Zs) :- dif(X,Y), (lcs([X|XS],YS, Zs); lcs(XS,[Y|YS],Zs)).

longest(P1, P2, P):-
    length(P1, L1),
    length(P2, L2),
    (   L1 > L2
    ->  P = P1
    ;   P = P2
    ).

:- initialization(main,main).
main(_) :- lcs([1,2,3], [2,7,1,4,2,3,6,5], LCS), print(LCS).
           %lcs([1,2,3], Ys, LCS1), print(LCS1).

Hmm. Tabling comes from earley parsing historically. That’s interesting. “Earley deduction”

tabling and packrat parsing DCGs + Memoing = Packrat Parsing But is it worth it? tabling vs memoing

Memoisation: Purely, Left-recursively, and with (Continuation Passing) Style More declarative tabling in Prolog using multi-prompt delimited control

Queries and Answers as separate concepts.

  • Variant Tabling - lookup in the table only allows for renaming of variables in the query
  • Subsumptive tabling - lookup can also succeed on anything that is

Call similarity vs answer similarity. User defined subsumption

Variant storage and query seems relevant to harrop datalog

Tabling makes an entry in the table when it encounters a goal. When it encounters this goal again it reads from the table

Eddie modelled prolog as a “memoized” (Pred -> Bool). Maybe really (Pred -> Maybe Bool) where the table contains lattice values. (Pred -> Set). SUbsumptive tabling makes Pred ordered too (slash a lattice?). The arrow has to be monotonic.

tnot. Stratified negation. Is this what all the continuations are about? Suspend this goal until the table is complete. Tabling as a server. The table can store continuations of who is expecting answers next?

Tabling needs to record continuations in order to know who to tell when it discovers a new answer for that query.


edges = [(1,2), (2,3), (3,4), (4,1), (2,1), (14,24)]
ptable = {}

# assert ans as possibility in response to question q
def add_ptable(q, ans):
  answers, conts = ptable[q] 
  if ans not in answers: # don't repeat answers already seen. Inductively, consumers already informed
    answers.add(ans) # add to answer set
    # inform consumers of this question of new answer
    for k in conts:
      k(ans)

# request information about paths starting at x.
def path(k, x): # y unknown aka path(x,Y) with continuation k
  if x in ptable:
    answers, conts = ptable[x]
    conts.append(k)
    for a in answers:
      k(a)
  else:
      ptable[x] = (set(),[k])
      # path(X,Y) :- edge(X1,Y), X1 = X.
      for (x1,y) in edges:
        if x1 == x:
          add_ptable(x, (x,y))
      # path(X,Z) :- edge(X1, Y), X1 = X, path(Y,Z)
      for (x1,y) in edges:
        if x1 == x:
          path(lambda ans: add_ptable(x, (x, ans[1])), y)

# query paths starting from 1
path(lambda ans : print("starts from 1", ans), 1)
#path(lambda ans : print(ans), 1)

add_ptable(1, (1,8))

print(ptable) # note 14,24 data not in there. Not bottom up
path(lambda ans : print("starts from 3", ans), 3)
print(ptable)

path(lambda ans : print("starts from 3", ans), 14)
print(ptable)

This somehow didn’t have a problem with the recursive answer. A failing computation may later succeed. Is it not a problem that we are filling the continuation table excessively?

To make it less spooky, the continuation could be defunctionalized maybe? This would allow us to deduplicate the continuations as we can make a set of the

Very coroutiney. Memoized coroutines?

bounded term depth property. all terms have bounded depth. tabling is guaranteed to terminate definite programs - no negation normal programs

A Term-Based Global Trie for Tabled Logic Programs give reasoning a trie cool summary of indexing structures

consumers vs generators. First time goal is reached it is egenerator. subsequent are consumers

To what degree can the parts of magic set be seen as analogs of tabling

variant tabling - the magic predicates may be storage of tabling keys / queries

Tabling is “memoized coroutines” Memoing

There is more variation available in memoizing than I realized.


fib_table = {0:1, 1:1}
def fib(n):
  if n in fib_table:
    return fib_table[n]
  res = fib(n-1) + fib(n-2)
  fib_table[n] = res
  return res

print(fib(3))

There are two very distinct sites of interaction with the table. Checking if a query is in the table, vs filling in the table once you know.

If you’re worried about a non terminating computation because you end up callign yourself in a loop, you can mark a question as already asked in the table. We lift a nonterminating query to and Option[int] returning query that for some examples of nontermination instead returns None.

fib_table = {0:1, 1:1}
def fob(n):
  if n in fib_table:
    #if fib_table[n] == None:
    #  return None
    return fib_table[n]
  else:
    fib_table[n] = None # mark question as asked
  x = fob(n) # now this is an infinite loop
  if x == None: # error handling
    return None
  y = fob(n-1)
  if y == None:
    return y
  res = x + y
  fib_table[n] = res
  return res

print(fob(3))

What about memoizing coroutines

Above was tabling/memoizing a determinstic computation Tabling semideterminstic is easier than full nondeterminism.

path query in semideterministic mode. Either fails or succeeds. Not multiple values.

edge = {0:1, 1:2, 2:0}
path_table = {(x,y) for i,j in edge.items()}
def path(x,z):
  if (x,z) in path_table:
    return True
  for y in edge[x]:
    p = path(y,z)
    if p == True:
      return True
  return False

print(path(0,0))

XSB

XSB prolog has some unsusual features and supposedly the most advanced implementation of tabling.

The manual is quite nice.manual vol 1

http://xsb.sourceforge.net/research.html

john[spouse->mary, child->>{bob,bill}]

translates to

fd(john,spouse,mary)
mvd(john,child,bob)
mvd(john,child,bill)

model checking model checking using tabled logic programming

term-sets {Var:Goal} let’s you talk about the set that corresponds to Goal. {X : member(X,[a,b,c])} This is related to “first class sets” in datalog Prolog has setof termsets - save space if factorable in this way compared to “extensionalizing” egraph equality infects everywhere. equals is equals. termsets aren’t saying terms are equal. egraphs for parse trees? a(a(a(b(b(b))))) as a string Or maybe symmetric comp form? S -> aSb S -> eps

There are different mechanisms for table size restrictions / improve terminating. radial,

incrmenetal table mainatence. Has a dependency graph. Something like seminaive? Trie based vs not?

Weaker semantics and choosing semantics.

Python

easy to call into python now https://www.swi-prolog.org/pldoc/doc_for?object=section(%27packages/janus.html%27)

https://www.swi-prolog.org/pldoc/man?section=janus-python-package pip install git+https://github.com/SWI-Prolog/packages-swipy.git#egg=janus_swi

https://www.swi-prolog.org/pldoc/man?section=janus-call-prolog https://swi-prolog.discourse.group/t/installing-janus-linking-error/6954 note neede same versin of janus a swipl –version installed python3 -m pip install git+https://github.com/SWI-Prolog/packages-swipy.git@V9.1.18#egg=janus_swi

from janus_swi import *
print(query_once("writeln('Hello world!')"))
print(apply_once("user", "plus", 1, 2))
print([a["D"] for a in query("between(1,6,D)")])

import janus_swi as janus

janus.consult("trains", """
    train('Amsterdam', 'Haarlem').
    train('Amsterdam', 'Schiphol').
    """)
print([d['Tuple'] for d in
           janus.query("train(_From,_To),Tuple=_From-_To")])
#prolog() # go into repl

janus.consult("stlctype", """
typeof(_,int(N), int) :- integer(N).
typeof(G,plus(X,Y), int) :- typeof(G,X,int), typeof(G,Y,int).
typeof(G,var(X),T) :- once(member(X-T,G)).
typeof(G,lam(X,B),arr(T1,T2)) :- typeof([X-T1|G],B,T2).
""")
print(list(janus.query("typeof([],plus(int(1),int(2)),T)")))
print(janus.query_once("typeof([],lam(x,var(x)),T), numbervars(T,1,_, [functor_name(var)])"))

swi

:- initialization(main,main).

main(_) :- 
  py_version,
  py_call(print("Hello World!\n")),
  py_call(sys:getsizeof([1,2,3]), Size), writeln(Size),
  py_call(sys:path, Path), writeln(Path),
  py_iter(range(1,3), X), writeln(X)
  %,py_call(eval("1"), X)
   .

Prolog II, III IV

Cyclic terms. Rational terms. See Condicutive logic programming swi prolog comments on rational trees

introduction to prolog III prolog and infinite trees

Parallel

https://en.wikipedia.org/wiki/Parlog 50 years of datalog talks about stuff.

Macros

https://swi-prolog.discourse.group/t/new-library-macros/6571 macro library. Seems nice triska on macros

:- use_module(library(macros)).
#define(max_size, 100).

https://www.swi-prolog.org/pldoc/man?section=progtransform goal_expansion term_expansion

Can I use macros to make better scoping?

:- initialization(main,main).

%term_expansion(Term,X) :- write(term), nl, write_canonical(Term), nl, fail.
term_expansion(bar(foo(X)), biz :- baz) :- writeln(hit_other).

%goal_expansion(Term,X) :- write(goal), nl, write_canonical(Term), nl, fail.

bar(foo(X)).

%goal_expansion(ex(X,T), )

main(_) :- 
  %write_canonical(foo(a))
  %write(hi).
  writeln(foo),
  listing(biz),
  true.

Coroutines

swi manual

  • dif/2 ? Test is delyed until terms are sfufcient different or have become identical
  • freeze/2 - equivalent to call is Var is bound frozen when call_residue_vars

delay block

Indexing

https://www.youtube.com/watch?v=FZLofckPu4A&ab_channel=ThePowerofProlog

  • first argument indexing
  • deep indexing
  • multi argument indexing
  • JIT

  • exchange arguments
  • use specific functor not default ones - https://www.metalevel.at/prolog/data#clean clean vs dirty reps. For example reprrsent integer by lit(N) rather than raw N. default args + nonlogical preds are nodet. not good. “defaulty” representation. We can’t symbolically express lit(X) + lit(Y). monotonic mode of clp(z)
  • pull unifications into head (should happen by default)
  • decompose arguments - use auxiliary predicates if system doesn’t offer deep indexing
  • lagged arguments - one case is subsumed by another. list_last convert to auxliary that remebed what you’ve seen. Check if foldl is applicable
  • reification. zcompare, if_. library(reif), tfilter

redundant choice points are an indication of lost performance or poor indexing

Indexing dif/2 Consider a removal predicate on a list. it involves dif.

first zargument indexing eclipse: from LP to CLP demand driven indexing of prolog clauses preindexed terms for prolog See also tabling

DCG - Definite Clauses Grammars (DCG)

https://www.youtube.com/watch?v=CvLsVfq6cks&t=3811&ab_channel=ThePowerofProlog

Anne Ogborn tutorial hakank picat triska dcg primer swi manual - comments are good too

DCGs look a bit like CFG grammars. They make threading an input and remainder list implicit. The input and remainder are considered a difference list.

DCGs describe sequences. They can be used to serialize and deserialize, parse and pretty print.

You call a DCG via phrase/2. phrase/3 gives you access to the remainder. phrase/2 specializes phrase/3 with a [] terminator. DCGs are annotated as double slashes as//0.

as --> [].
as --> [a], as.

:- initialization(main,main).
main(_) :- phrase(as, [a,a,a]), print("passed"), 
          phrase(as, [a,b,a]), print("doesn't pass").

{ mygoal } is a way to lift a prolog goal that doesn’t touch the input lists.

Triska suggests turning on double quoted strings becomes lists. It appears swi has some behavior like this by default.

:- set_prolog_flag(double_quotes, chars).
% even number of as
as --> "".
as --> "aa", as.

:- initialization(main,main).
main(_) :- phrase(as, "aaaa"), print("passed").

seqq describes concatenating a sequence of lists.

  • seq
  • semicontext
  • DCGs are useful also for implicilty carrying state.
  • State trasnformation is a sequence of states when considered purely functionally.
state(S), [S] --> [S].
state(S0, S), [S] --> [S0].

Being able to implicilty pass state makes DCGs reminscent of monads. I’ve not seen a proof of equivalence however. State is a very powerful construct. Me dunno. Maybe you could even pass around continuations in there.

This is showing context notation. Lists on the left hand sides of rules are pushed onto the stream after the rule works (I guess that means it unifies the head of the remained with this quantity). It looks very much like a form of lookahead.

library dcg/basics, dcg/high_order

library pio - pure io. phrase_from_file phrase_to_stream

Triska shows an interesting trick to improve nontermination. You can parametrize your dcg such that it knows it is consuming the input stream. The recursion depth should be bounded by the stream size itself. Very reasonable. The decreasing argument is the stream itself.

expr([_ | Rs], Rs) –> “1”. shows that expr consumes one character Called via phrase(expr(Rs, _),Rs).

Or more easily, use tabling.

Reporting errors - add error streat arguments to dcg

regexp interpreter. dcg rexp//1

Longest match firsat behavior - order rules for longest match

Constraint Logic Programming (CLP)

  • CLP(B) - constraint over boolean variables. Sometimes bdd based
  • CLP(FD)
  • CLP(Z)
  • CLP(H)

Anne Ogborn’s tutorial

Indexing dif/2 reif if_/ tfilter/3

eclipse talks to minizinc? clp examples

CHR is a language for eaiser custom clp theories

Constraint Logic Programming : A Survey

Constrained Horn Clauses are not that much different from CLP.

Bottom up execution. “facts” allows logically constrained variables. foo(X) :- x > 7, X < 11.. I wonder how resolutin provers deal with theory specific stuff.

Execution model: A goal stack + constraint store. Minikanren? A Framework for Extending microKanren with Constraints

clpb

swi manual clpb The Boolean Constraint Solver of SWI-Prolog: System Description Boolean Constraints in SWI-Prolog: A Comprehensive System Description clpb triska github repo with examples

:- include(module(clpb)).

% a finite set is a mapping / assoc-list to bool
% a multiset is a mapping to int
union()

main(_) :- [x - X. y - Y, z - Z]

bdd, indexical, or external sat. robdd.pl

bdd using chr?


bdd(Var, L,R , Id), bdd( ,L,R, Id) ==>
bdd(Var,L,R,)

independent set example

independent_set(*(NBs)) :
  findall(UV, edge(U, V), Edges),
  setof(U, V^(member(UV, Edges);member(VU, Edges)), Nodes),
  pairs_keys_values(Pairs, Nodes, _),
  list_to_assoc(Pairs, Assoc),
   maplist(not_both(Assoc), Edges, NBs).

 not_both(Assoc, UV, ~BU + ~BV) :
  get_assoc(U, Assoc, BU),
  get_assoc(V, Assoc, BV).
  • domino tilings

https://stackoverflow.com/questions/63505466/prolog-implementation-of-quines-algorithm-for-classical-propositional-logic-in?noredirect=1&lq=1 https://stackoverflow.com/questions/64835191/boolean-unification-in-prolog

clp(euf)

union(+T1,+T2) T1 =:= T2 delay until ground? Not quite right. foo(x,X) foo(y,X) still has congruence power if we learn x = y. union(eid(), eid()). Support union of terms and eid find(+T,-Eid) ematch(+Pat,-Match) extract(+Eid, -Term)

attrbitued var style (z3 style?). put_attr( ,euf_parents,) put_attr( , euf_sibling, ) put_attr( , term, ) vs chr style There was also in that prolog smt paper a congruence closure routine.

Attributed Variables

Attaching extra info to variables. This can be be used to implement CLP as a library

https://www.metalevel.at/prolog/attributedvariables https://sicstus.sics.se/sicstus/docs/3.7.1/html/sicstus_17.html https://www.swi-prolog.org/pldoc/man?section=attvar

Attributed variables are a union find dict?

Atributed variables are also reminscent of nelson oppen, theories are informed

:- initialization(main,main).


attr_unify_hook(Domain, Y) :-  % it's odd this is not tagged by the event type. 
 % Since we don't super get the key, perhaps it is wisest to wrap the attribute foo(Domaindata)
  writeln(union_event),
  writeln(Domain), % larry (data held in "Y")
  writeln(Y). % variable id of "X"

main(_) :- put_attr(X, mykey, fred), 
           put_attr(Y, mykey, larry), 
           put_attr(Y, mykey2, farry),  % with ths union_event is triggered twice, once per attribute
           get_attr(X,mykey,F), writeln(F),
           writeln(X), writeln(Y),
           X = Y,  % triggers union_event
           get_attr(X,mykey,FX), writeln(FX), % fred 
           get_attr(Y,mykey,FY), writeln(Y), writeln(FY)  % fred. Y is literally X now anyway.

          .

Coroutining

Sometimes is makes sense to be lazy doing something until a variable is filled in more. This is basically what dif/2 does. It wakes up when either thing gets more filled in

freeze is a delayed call

CLP(Set)

https://www.clpset.unipr.it/ {log} “setlog” JSetL

set unification Sets and constraint logic programming

G Rossi A Dovier E Pontelli

herditraryl hybrid finite sets. Finite sets of finite sets + terms

Set Constraints in Logic Programming Set constraints and logic programming - kozen CLP(SC) Herbrand atom ~ singleton set Aiken

Relation to boolean equation systems?

Kuper - Logic programming with sets

CHR

forward chaining, chr comes up

swipl manual section Anne Ogborn’s tutorial Schrijver’s ICLP tutorial

Constraint handling rules A question I never had an answer for https://twitter.com/notjfmc/status/1422215450675535877?s=20&t=RyHMtBS3ySaALLC9MuGmUA . CHR afaik are a way of integrating eager rewriting into prolog https://en.wikipedia.org/wiki/Constraint_Handling_Rules

I’m not sure this is even persay a prolog topic. But the prolog community is the most engaged

http://www.informatik.uni-ulm.de/pm/fileadmin/pm/home/fruehwirth/constraint-handling-rules-book.html chr.js chr what else

Where is the CHR constraint store?

Some cool examples. Gaussian, fourier, rational trees, equation solvers

https://github.com/brog45/chrplay

Multiset rewriting?

cchr efficient implementation in C

yihong’s egraph in chr. awesome

chr parsing

parallel chr

Automatic Differentiation using Constraint Handling Rules in Prolog Build up dataflow graph. Huh.

Pros and Cons of Using CHR for Type Inference https://twitter.com/flippac/status/1567622246025383938?s=20&t=7jEZ-QlMkI3Jdw1l-LGSjg

Toward a First-Order Extension of Prolog’s Unification using CHR

satisfiability modulo chr

Extending Arbitrary Solvers with Constraint Handling Rules

Compiling

KU leuven system : implementation and application. Hmm. Is CHR compiled into prolog code? CCHR: the fastest CHR Implementation, in C
Idle thoughts, what about time dataflow.

attributed data for chr indexing

Examples

CHR parsing “Analysing graph transformation systems through Constraint Handling Rules” by Frank Raiser and Thom Frühwirth “As time goes by: Constraint Handling Rules — A survey of CHR research from 1998 to 2007” by Jon Sneyers, Peter Van Weert, Tom Schrijvers and Leslie De Koninck

https://stackoverflow.com/questions/67771845/prolog-get-a-list-of-all-the-rules-an-entity-verifies

:- use_module(library(chr)).

:- chr_constraint snore/1, sleep/1, breathe/1.
:- chr_constraint eat/1, live/1, rest/1, collect/2, pull/2.

snore(X) ==> breathe(X).
snore(X) ==> sleep(X).
sleep(X) ==> rest(X).

breathe(X) ==> live(X).
eat(X)     ==> live(X).
sleep(X)   ==> live(X).

live(X) \ live(X) <=> true.  % eliminates duplicates

collect(Who,L),snore(Who)   <=> collect(Who,[snore|L]).
collect(Who,L),sleep(Who)   <=> collect(Who,[sleep|L]).
collect(Who,L),breathe(Who) <=> collect(Who,[breathe|L]).
collect(Who,L),eat(Who)     <=> collect(Who,[eat|L]).
collect(Who,L),live(Who)    <=> collect(Who,[live|L]).
collect(Who,L),rest(Who)    <=> collect(Who,[rest|L]).

pull(Who,L) \ collect(Who2,L2) <=> Who = Who2, L = L2.

sicstus examples swi examples


:- module(leq,[leq/2]).
:- use_module(library(chr)).

:- chr_constraint leq/2.
reflexivity  @ leq(X,X) <=> true.
antisymmetry @ leq(X,Y), leq(Y,X) <=> X = Y.
idempotence  @ leq(X,Y) \ leq(X,Y) <=> true.
transitivity @ leq(X,Y), leq(Y,Z) ==> leq(X,Z).
/*
?- leq(X,Y), leq(Y,Z).
leq(X, Z),
leq(Y, Z),
leq(X, Y).
*/

finite domain solver.

:- module(dom,[dom/2]).
:- use_module(library(chr)).

:- chr_constraint dom(?int,+list(int)).
:- chr_type list(T) ---> [] ; [T|list(T)].

dom(X,[]) <=> fail.
dom(X,[Y]) <=> X = Y.
dom(X,L) <=> nonvar(X) | memberchk(X,L).
dom(X,L1), dom(X,L2) <=> intersection(L1,L2,L3), dom(X,L3).
% ?- dom(A,[1,2,3]), dom(A,[3,4,5]).
% A = 3.

CHR debugging chr tracing chr_show_store

https://www.swi-prolog.org/pldoc/man?section=chr-guidelines Don’t bind rules in head mode declarations of chr affect performance. Huh c \ c <=> true is often desirable. Set semantics instead of multiset.

You can ignore the

:- initialization(main,main).
main(_) :- whatever

as noise that I do to make my prolog programs run from the command line instead of interactively.

Basics

:- use_module(library(chr)).
:- chr_constraint rain/0, wet/0, umbrella/0.

rain ==> wet.
rain ==> umbrella.

:- initialization(main,main).
main(_) :- rain, chr_show_store(true).
:- use_module(library(chr)).
:- chr_constraint rain/0, wet/0, umbrella/0.

rain <=> wet.
rain <=> umbrella.

:- initialization(main,main).
main(_) :- rain, chr_show_store(true).
% just wet
:- use_module(library(chr)).
:- chr_constraint left/0, right/0, forward/0, backward/0.

left,right  <=> true.
forward, backward <=> true.

:- initialization(main,main).
main(_) :- forward, left, right, backward, left, chr_show_store(true).
% just left

Pairing

:- use_module(library(chr)).
:- chr_constraint male/1, female/1, pair/2.

% <=> makes a pairing
% this makes all pairs
% male(X),female(Y)  ==> pair(X,Y).
male(X) \ female(Y)  <=> pair(X,Y).
% hmm cransto gets farbus.

:- initialization(main,main).
main(_) :- male(gary), female(zirkin),  male(cransto), female(farbus), chr_show_store(true).

Paths from edges

:- use_module(library(chr)).
:- chr_constraint edge/2, path/2.

base  @ edge(X,Y) ==> path(X,Y).
trans @ edge(X,Y), path(Y,Z) ==> path(X,Z).
pset  @ path(X,Y) \ path(X,Y) <=> true.

:- initialization(main,main).
main(_) :- edge(1,2), edge(3,4), edge(3,2), edge(1,3), chr_show_store(true).

Minimum

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint min/1, findmin/1.

findmin(_), min(N) \ min(M) <=> N < M | true.
findmin(Min), min(N) <=> Min = N.
main(_) :- min(7), min(14), min(13), findmin(Min), print(Min), chr_show_store(true).

GCD

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint gcd/1.

gcd(N) \ gcd(M) <=> 0 < N, M>=N | Z is M - N, gcd(Z).
% hmm not working
main(_) :- gcd(15), gcd(5), chr_show_store(true).

Sort

bubble sort

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint a/2.

a(I,V), a(J,W) <=> I>J, V<W | a(I,W), a(J,V).
main(_) :- a(1,6), a(2,4), a(3,14), chr_show_store(true).

merge sort

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint next/2.

next(A,B) \ next(A,C) <=> A < B, B < C | next(B,C).
main(_) :- next(0,7), next(0,2), next(0,5), chr_show_store(true).

Get foo

Fibonacci

merge sort

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint fib/2.

fib(0,M) <=> M = 1.
fib(1,M) <=> M = 1.
%fib(N,M) <=> N >= 2, ground(N) | fib(N-1, M1), fib(N-2, M2), when((ground(M1),ground(M2)), M is M1 + M2).
fib(N,M) <=> N >= 2, ground(N) | N1 is N - 1, N2 is N - 2, fib(N1, M1), fib(N2, M2), M is M1 + M2.

main(_) :- fib(5,M), print(M), chr_show_store(true).
:- use_module(library(chr)).
:- use_module(library(clpfd)).
:- initialization(main,main).
:- chr_constraint fib/2.

fib(0,M) <=> M #= 1.
fib(1,M) <=> M #= 1.
fib(N,M) <=> ground(N), N #>= 2 | N1 #= N - 1, N2 #= N - 2, fib(N1, M1), fib(N2, M2), M #= M1 + M2.

main(_) :- fib(5,M), print(M), chr_show_store(true).

top down memo

:- use_module(library(chr)).
:- use_module(library(clpfd)).
:- initialization(main,main).
:- chr_constraint fib/2.

cong @ fib(N,M1) \ fib(N,M1) <=> M1 #= M2.
fib(0,M) ==> M #= 1.
fib(1,M) ==> M #= 1.
fib(N,M) ==> ground(N), N #>= 2 | N1 #= N - 1, N2 #= N - 2, fib(N1, M1), fib(N2, M2), M #= M1 + M2.

main(_) :- fib(5,M), print(M), chr_show_store(true).
% store is not empty. Contains memoized fib entries.

Boolean Propagators

Lattice

lat(A), lat(B) <=> lat(join(A,B))

Propagators are monotonic functions between lattices. CHR was built for propagation BitSet CHRs? Other fast propagators?

Assignment

Union Find

From furhwith book. Generalized union find is also interesting.

:- use_module(library(chr)).
:- initialization(main,main).
% hmm are these annotations ok?
:- chr_constraint make(+int), find(+int,-), root(+int,+int), union(+int,+int), link(+int,+int), pto(+int,+int).

make(A) <=> root(A,0).
union(A,B) <=> find(A,X), find(B,Y), link(X,Y).
pto(A, B), find(A,X) <=> find(B,X), pto(A,X).
root(A,_) \ find(A,X) <=> X=A.
link(A,A) <=> true.
link(A,B), root(A,N), root(B,M) <=> N>=M | pto(B,A), K is max(N,M+1), root(A,K).
link(B,A), root(A,N), root(B,M) <=> N>=M | pto(B,A), K is max(N,M+1), root(A,K).

main(_) :- make(1), make(2), make(3), union(1,2), find(2,X), find(3,Y), 
    print(X),print(Y), union(1,3), chr_show_store(true).

EGraphs

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint eclass/2.
fcong @ eclass(f(X), E1) \ eclass(f(X), E2) <=> E1 = E2.
acong @ eclass(a, E1) \ eclass(a, E2) <=> E1 = E2.

main(_) :- eclass(a, A), eclass(f(A), FA), eclass(f(FA), FFA), FA=A, chr_show_store(true).
:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint eclass/2, gas/2.
pcong @ eclass(X + Y, E1) \ eclass(X + Y, E2) <=> E1 = E2.
% nothing should produce this? Well...
% ncong @ eclass(num(X), E1) \ eclass(num(X), E2) <=> E1 = E2.
% pcong @ eclass(T, E1) \ eclass(T, E2) <=> E1 = E2.

gas(_, 0) <=> true.
comm @ eclass(X + Y, E) \ gas(comm, N) <=> N > 0 | N1 is N - 1, gas(assocl, N1), eclass(Y + X, E).
assocl @ eclass(X + YZ, E), eclass(Y + Z, YZ) \ gas(assocl, N) <=> N > 0 | N1 is N - 1, gas(assocr, N1), eclass(X + Y, XY), eclass(XY + Z, E).
assocr @ eclass(X + Y, XY), eclass(XY + Z, E) \ gas(assocr, N) <=> N > 0 | N1 is N - 1, gas(comm, N1), eclass(X + YZ, E), eclass(Y + Z, YZ).


main(_) :- eclass(1, E1), eclass(2,E2), eclass(3,E3), eclass(E1 + E2, E12), eclass(E12 + E3, E123), gas(comm,5), chr_show_store(true).
% hmm we're only getting to the first rule... I see. We can of course get around this. but not elegantly.
% hmm we're also not runnign fairly on deeper facts. Cripes.

pcong is only one step away from structure sharin? No. it really is just structure sharing… Hmm.

Hmm. gas is yet another form of demand based pulling. but only in CHR

A list of all things to retain in gas. Delete anything that fires. No we don’t have to delete. ok ok ok. This isn’t so bad. We just need to batch. reassert when done. Difference list useful? gas(L) <=> L = [X | L1], gas(L1), But now is there a fair chance for the rules? Maybe if we scramble the list. it’s a mess.

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint eclass(?,-), eclass2(?,-), col/1, kill/0, count/1.

cong @ eclass(T, E1) \ eclass(T, E2) <=> E1 = E2.


% rewrite rules.
comm @ eclass(X + Y, E) ==> eclass2(Y + X, E).
assocl @ eclass(X + YZ, E), eclass(Y + Z, YZ) ==> eclass2(X + Y, XY), eclass2(XY + Z, E).
assocr @ eclass(X + Y, XY), eclass(XY + Z, E) ==> eclass2(X + YZ, E), eclass2(Y + Z, YZ).

% made it way worse
% eclass(T,E) \ eclass2(T,E) <=> true.
% eclass2(T,E) \ eclass2(T,E) <=> true.

% To collect up new eclasses
collect @ eclass2(T,E), col(L) <=> L = [eclass3(T,E) | L1], col(L1).
done @ col(L) <=> L = [].

% helpers to cleanup eclass2
kill @ kill \ eclass2(_,_) <=> true.
killdone @ kill <=> true.

% helper to count eclasses
count @ count(N), eclass(_,_) <=> N1 is N + 1, count(N1).

% Take rhs list and inject them as CHR constraints 
process([]).
process([eclass3(T, E)| L]) :- eclass(T,E), process(L).


% Do N rewriting runs
batch() :- col(L), process(L). % print(L)
batch(0).
batch(N) :- batch(), N1 is N -1, batch(N1).


init_add(N) :- eclass(N,E), N1 is N - 1, init_add_aux(N1,E).
init_add_aux(0,_).
init_add_aux(N,E) :-
  eclass(N, EN), eclass(EN + E, E2), N1 is N-1, init_add_aux(N1, E2).


insert( T , E) :-
 ground(T),
 var(E),
 T =.. [F | Args],
 length(Args, N), length(Es, N),
 T2 =.. [F | Es],
 eclass(T2, E),
 maplist(insert, Args, Es).


main(_) :- 

        %  insert(f(a), Fa), insert(a, A), Fa = A, insert(f(f(a)), FFa),
        %   chr_show_store(true).
         % eclass(1, E1), eclass(2,E2), eclass(3,E3), eclass(E1 + E2, E12), eclass(E12 + E3, E123),
       
          N = 5,
          init_add(N),
          Num is 3**(N) - 2**(N+1) + 1 + N, print(Num),
          BNum is N,
          time(batch(BNum)), kill, count(0), chr_show_store(true).
          
          
%rhs([]).
%rhs([T | L]) :- flatten(T), rhs(L).
:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint eclass(?,-), eclass2(?,-), col/1, kill/0, count/1.


insert( T , E) :-
 ground(T),
 var(E),
 T =.. [F | Args],
 length(Args, N), length(Es, N),
 T2 =.. [F | Es],
 eclass(T2, E),
 maplist(insert, Args, Es).

main(_) :- insert(1 + 2 + 3 + 4, E), chr_show_store(true).


We can also perform rules like find(t1), find(t2) ==> E1 = E2. These will always finish since they only compress egraph.

It is quite possible that translating to integers and using CHR union find is faster.

Semi naive?

eclass2 is somewhat like new_eclass We could perhaps find delta_eclass

:- use_module(library(chr)).
:- initialization(main,main).
:- chr_constraint eclass(?,-), declass/2, eclass2(?,-), col/1, kill/0, count/1.

% Take rhs list and inject them as CHR constraints 
process([]).
process([eclass3(T, E)| L]) :- declass(T,E), process(L).

% Do N rewriting runs
batch() :- col(L), process(L). % print(L)
batch(0).
batch(N) :- batch(), N1 is N -1, batch(N1).

init_add(N) :- eclass(N,E), declass(N,E), N1 is N - 1, init_add_aux(N1,E).
init_add_aux(0,_).
init_add_aux(N,E) :- eclass(N, EN), eclass(EN + E, E2), declass(N, EN), declass(EN + E, E2), N1 is N-1, init_add_aux(N1, E2).

insert( T , E) :- ground(T), var(E), T =.. [F | Args], length(Args, N), length(Es, N),  T2 =.. [F | Es],
 eclass(T2, E),  maplist(insert, Args, Es).


cong @ declass(T, E1), eclass(T, E2) ==> E1 = E2.
cong2 @ eclass(T, E1) \ eclass(T, E2) <=> E1 = E2, declass(T,E1).

% rewrite rules.
comm @ declass(X + Y, E) ==> eclass2(Y + X, E).
assocl1 @ declass(X + YZ, E), eclass(Y + Z, YZ) ==> eclass2(X + Y, XY), eclass2(XY + Z, E).
assocl2 @ declass(Y + Z, YZ), eclass(X + YZ, E) ==> eclass2(X + Y, XY), eclass2(XY + Z, E).
assocr1 @ declass(X + Y, XY), eclass(XY + Z, E) ==> eclass2(X + YZ, E), eclass2(Y + Z, YZ).
assocr2 @ declass(XY + Z, E), eclass(X + Y, XY)  ==> eclass2(X + YZ, E), eclass2(Y + Z, YZ).

% made it way worse
% eclass(T,E) \ eclass2(T,E) <=> true.
% eclass2(T,E) \ eclass2(T,E) <=> true.

% To collect up new eclasses
collect @ eclass2(T,E), col(L) <=> L = [eclass3(T,E) | L1], col(L1).
collect @  col(_) \ declass(_,_) <=> true.
done @ col(L) <=> L = [].

% helpers to cleanup eclass2
kill @ kill \ eclass2(_,_) <=> true.
kill2 @ kill \ declass(_,_) <=> true.
killdone @ kill <=> true.

% helper to count eclasses
count @ count(N), eclass(_,_) <=> N1 is N + 1, count(N1).




main(_) :- 
          N = 2,
          init_add(N),
          Num is 3**(N) - 2**(N+1) + 1 + N, print(Num),
          BNum is N,
          time(batch(BNum)), % kill(), count(0),
           chr_show_store(true).
          
          

Is there a way to encode seminaive eval maybe? Even for datalog this is not clear you can. The folding of delta back into the main relationseems problematic.

leansmt egraph + clp(R) + clp(FD) + clp(B). How far from SMT is that?

Embedding into CHR

  • TRS
  • Prolog and CLP
  • Graph trasnfomation - See term rewriting notes
  • Petri nets

GAMMA general abstract mnodel for multiset manipulation

SQL

sql Man what hope is there of compiling a 7 year old haskell project? compiler to sql statements. Makes sense.

:- use_module(library(chr)).
:- chr_constraint left/0, right/0, forward/0, backward/0.

left,right  <=> true.
forward, backward <=> true.

:- initialization(main,main).
main(_) :- forward, left, right, backward, left, chr_show_store(true).
CREATE TABLE left(a);
CREATE TABLE right(a);
CREATE TABLE backward(a);
CREATE TABLE forward(a);

INSERT INTO left VALUES (NULL);
INSERT INTO right VALUES (NULL), (NULL);
INSERT INTO backward VALUES (NULL), (NULL);
INSERT INTO forward VALUES (NULL);

select *, right.rowid from right;
create table env(left,right, UNIQUE(left), UNIQUE (right)); -- Use CTE? But then multiple statements?
-- INSERT INTO env SELECT left.rowid, right.rowid FROM left,right LIMIT 1; -- hmm we're overselecting. LIMIT BY 1? Yuck
INSERT OR IGNORE INTO env SELECT left.rowid, right.rowid FROM left,right;
-- only update timestamp if env is empty.
-- we could push env into the metalayer. Slow? Also the uniqueness filtering.
select * from env;

-- do rhs here if there was one. 
-- INSERT INTO rhs FROM env, left, right WHERE left.rowid = env.left AND right.rowid = env.right
DELETE FROM left where left.rowid IN (select env.left from env);
DELETE FROM right where right.rowid IN (select env.right from env);
DELETE FROM env;
select *, rowid from right;
select *, rowid from left;

Compred with datalog, sql multiset semantics is good match. Hmm. Can’t make nullary tables?

Hmm. The semantics of CHR are a problem. It seems unlikely I can do bulk matching. I’d have to just take the first match. That’s putting a lot of control flow in the metalayer / not amoriziting well over sql statements.

ooh, what about env(left,right, UNIQUE left, UNIQUE right). Yes. This seems safe enough and yet bulk. Nice.

I can compromise on the sequential nature of chr, but th resource matching feels important. Explicit multiset semantics with a count field? That’s a bummer. Feels complicated. ON CONFLICT UPDATE left -> 1, right -> 2 —> left –> left -> 1 - min(1,2)), right -> 2 - min(1,2) Naw, not even this fixes it. The limit by 1 might be the best solution.

The issue here that really requires an env is that we have multi-headed rules.

left = {None : 1} # multisets as maps to integers
right = {None : 2}

for l, n in left.items():
  if n > 0:
    for r, m in right.items():
      if m > 0:
        if l == r:
          left[l] -= min(n,m)
          right[r] -= min(n,m) # Is this mutation ok?
          #break or double break?
print(left)
print(right)

# list as multiset
left = [(,)]
right = [(,), (,)]

# mutating the collecton while iterating over it is irritating again

If my advice is to just use SWI for CHR, why is tabled SWI not my advice for datalog. No. That is reasonable advice probably. Not unless you’re doing something weird

What would be better is to store the build values in env and not the keep ids. I don’t know persay how bad that join is. Maybe using a left join or something so that sql knows it doesn’t have to even loop up stuff it doesn’t need.

This is not that elegant.


class Rule():
  def __init__(self, name, kill, keep, guard, build):
    cols = ", ".join([f"{row})" for table, row in kill + keep])
    constraints = ", ".join([f"UNIQUE ({row})" for table, row in kill]) # To avoid double counting kills
    fields = ", ".join(cols, constraints)
    create_env = f"CREATE TEMP TABLE env({fields})" # put name on the env?

    selects = ", ".join([f"{row}.rowid" for table, row in kill + keep])
    wheres = " AND ".join(guards)
    find = f"INSERT OR IGNORE INTO env SELECT DISTINCT {selects} FROM {froms} WHERE {wheres}"
    
    deletes = [f"DELETE FROM {table} INNER JOIN env ON {table}.rowid = env.{row}" for table,row in kill]

    #rowenv  = ", " [ "  for table,row in kill + keep]
    where = " AND ".join([f"{table}.rowid = env.{row}" for table,row in kill + keep])
    inserts = [f"INSERT INTO {table} SELECT {val} FROM env, {selects} WHERE " for table, val in build]

    "DROP TABLE env"

Could imagine a syntax with an explicit binding part that puts these things in the env. And that the built expsressions may not refer to the bound varaibles. Annoying restriction. And obviously we could

l @ left, r @ right <=> guard | p = l.x, z = r.y |  {r,a}

Keep a bunch of rules synchronized is annoying

Answer Set Programming s(CASP)

https://swish.swi-prolog.org/example/scasp.swinb “Tabling and s(CASP) are quite different beasts. They both address reasoning in cyclic domains including negation. Tabling provides Well Founded Semantics where s(CASP) provides Stable Model Semantics. s(CASP) stresses creating an explanation. Tabling does not. Tabling scales a lot better than s(CASP). It all depends …”

https://gitlab.software.imdea.org/ciao-lang/sCASP Constraint Answer Set Programming without Grounding Justifications for Goal-Directed Constraint Answer Set Programming ERGO AI uses justification trees https://coherentknowledge.com/ using XSB

s(ASP) was just ASP https://personal.utdallas.edu/~gupta/ . S(CASP) includes constraints tutorial slides tutorail paper

event calculus

Extralogical features

Database manipulation

Persistent predicates based on RocksDB https://github.com/JanWielemaker/rocks-predicates

swi sicstus database manip retract - take out of database assert - put into database/ set_prolog_falg dynamic predicates. recorded database

A = A, assert(foo(A)), A = x what does this do. Does it make a clause foo(A). or does it make foo(x). Ah ok. The term is copied. So it makes the first on

a declarative alternative to assert in logic programming

fib(0,1).
fib(1,1).
fib(M,N) :- M > 1, M1 is M - 1, M2 is M - 2, fib(M2, N2), asserta(fib(M2,N2)), 
            fib(M1, N1), asserta(fib(M1,N1)), N is N1 + N2.
% hmm. Infinite answers...

Program reflection. Partial evaluation.

Cuts and Such

cut green cut

findall bagor setof are aggregation of solutions. They are reifying predicates kind of.

Finding all Solutions to a Goal

Functions

https://www.swi-prolog.org/pack/list?p=function_expansion a macro for function expansion

Lambda

Oh sweet mysterious lambda

The minikanren evalo

Oleg’s lambda interpreter

eval(E1*E2, R) :- eval(E1, E1R), eval(E2,E2R), copy_term(E1R,(E2R->Body)),
                       eval(Body,R).
eval(E,E).

One is tempted to attempt to use prolog variables for lambda variables. Requires extralogical copying primitives copy_term

swipl lambda expressions. [X] >> B as lam(X,B).

lambda is iso prolog

the elpi paper argues for de bruijn levels?

% locally nameless

term(lam(X,X)).
% convert lam(X,X) form to gensym lam(fvar(x123),fvar(x123)) form
% I should just use numbervar/1
concrete( lam(X,B) , lam(X, B2)  ) :-  gensym(x, X), debruijn(B, B2). % copy term maybe

% Turn X into bvar(N).

close(N, X, X, bvar(N)).
close(N, X, lam(Y, B), lam(Y,B2) ) :- Y /= X, N2 is N + 1, close(N2, B,B2).
close(N, X, T, T2) :- T /= lam(), T /= X, 
  T ..= [F | Args],
  maplist(close(N,X), Args, Args2), 
  T2 .== [F | Args2].

abstract(X, T, blam(T2)) :- close(0,X,T,T2).

instantiate(N, V, bvar(N), V).
instantiate(X)

% norm(T,T2)
% miller(T,T2). % :=: or some other special equals.

% miller(X,X).
% miller(blam(T1), blam(T2)) :- miller(T1,T2)
% miller(app(F, Args), T) :- var(F), reabstract(T, Args, F). % maybe throw a norm in there?
% miller(X,Y) :- X ..= [F | Args1], Y ..= [F | Args2],  maplist(miller, Args1, Args2).

% F @ [X,Y,Z] infix application
% norm(lam(X,X) @ 3, 3).



The question is, can lambda prolog be implemented as a library?

HiLog

My impression is this is a bit like “first order functional” programming. All predicates need names. You can achieve this via defunctionalization, lambda lifting, etc.

From HiLog to Prolog

X(a, b) is just an (external) shorthand for the Prolog compound term: apply(X, a, b)

Alpha prolog

https://homepages.inf.ed.ac.uk/jcheney/programs/aprolog/

Nominal logic https://arxiv.org/pdf/1012.4890.pdf nominal unification revsitied Urban

https://johnwickerson.github.io/dissertation.pdf “atoms” are a thing that are less free than unification vars but more free than ground terms. They are alpha renamable

freshness constraint a # t says that a appears in no subterm in t. a # X when X is a variable can’t be concluded.

freshness judgement equational judgements

nominal unification returns substitution and new freshness constrains

A relative of miller pattern. They can be encoded into each other it says and gives references

J. Cheney. Relating Nominal and Higher-Order Pattern Unification. J. Levy and M. Villaret. Nominal Unification from a Higher-Order Perspective

==(A \ X, B \ Y) :- dif(A,B), X == swap(A,B,Y), A # Y.

A # A \ Y.

https://easychair.org/publications/preprint/DLVk Efficiency of a good but not linear nominal unification algorithm

Semantics

Completion semantics

Well-founded Well founed semantcis 1991 Well-Founded Semantics, Generalized two different ordering realtions, truth and knowledge. We always move from unknown to more known. well founded tutorial

Completion semantics

History

Resolution in automatc theorem provers came earlier. Kowalski and Colmerauer

Flotd nondeterminstic algorithms 9167 PLANNER Hewitt

Prolog 0 Prolog 1

metamorphisis grammars -> DCG Dec-10 prolog Ediburgh prolog

structure copying vs structure sharing

fifth generation computing

Extensions to unification prolog II Prolog III

CLP Jaffar Lassez 1987

Expert Systems

See Also:

  • Databases Knowledge Representation

https://en.wikipedia.org/wiki/Rete_algorithm

Lambda Prolog

tabling in higher order logic programming

https://userpages.uni-koblenz.de/~laemmel/paradigms1011/resources/pdf/lambda-prolog.pdf ralf lammel slides

lambda prolog page Implementations

  • Teyjus - seems to maybe not be developed anymore
  • elpi - alive and well. Coq metaprogamming
  • Makam - a variant on lambda prolog with some cooll paper tutorials

elpi: fast embeddable lmabda prolog interpreter

Install as opam install elpi. Run as elpi -test to run with main query

main :- print "hello world".

A little bit of an odd encoding. Why fact? Oh I see. Because they have an outer loop that is inserting facts.

kind entry type.
type fact entry -> o.
%type false o.

kind person type.
type kim,dana person.
type finished person -> int -> entry.
type cs_major, graduates person -> entry.

fact (finished kim 102) & fact (finished dana 101).
fact (finished kim 210) & fact (finished dana 250).

fact (cs_major X) :- (fact (finished X 101); fact (finished X 102)),
        fact (finished (X 250)), fact (finished X 301).

false :- fact (finished X 210), fact (finished X 250).

kind jar, bug type.
type j jar.
type sterile, heated jar -> o.
type dead, bugs bug -> o.
type in bug -> jar -> o.

sterile J :- pi x\ bugs x => in x J => dead x.
dead B :- heated J, in B J, bugs B.
heated j.

main :- sterile J, print J.

type rain prop.
type wet prop.
type umbrella prop.
%mode (rain i).
%mode (wet i).
%mode (umbrella i).
rain :- !, declare_constraint rain [_].
wet :- !, declare_constraint wet [_].
umbrella :- !, declare_constraint umbrella [_].

main :- std.append [] [] X, print X,
        std.length [1,3,4] L, print L,
        std.take 2 [1,2,3,4] L1, print L1,
        std.map [1,2,3] (x\ y\ y is 2 * x) L2, print L2,
        std.iota 10 L3, print L3, % range of numbers
        true,
        Z is 1 i+ 2, print Z,
        Z1 is (cos 3.0) + (int_to_real ((rhc "a") + (size "hello"))), print Z1,
        fst (pr 1 2) One, print One,
        dprint (pr (some 3)),
        var V, var (V1 3) A B, print A, print B, % so it destructures a vairable application
        some T == some T,
        trace.counter "foo" I, print I, trace.counter "foo" I1, print I1,
        %print app [lam (x \ const "a"), const "b"],
        declare_constraint wet [W],
        declare_constraint rain [R],
        declare_constraint umbrella [U],
        R = 7,
        print "ok"
.

constraint rain wet {
  rule rain <=> wet.
}
kind unit type.
type tt unit.
type lam  (unit -> unit) -> unit.

main :- (x\ x) = (y\ F y), print F,
     (x\ x) = (y\ y),
     % fails (x\ x) = (y\ F), % this is exists F, \x x = y\ F. Which is not solvable
     % pi (y\ X = y) % fails
      (x\ tt) = (y\ F2), print F2
. 
main :- (x \ x) = (x \ F_ x),
        (x \ y \ x y) = (z \ G z), print G.

% church encoding
kind i type.
type zero, one, two ((i -> i) -> i -> i) -> o.
type add ((i -> i) -> i -> i) -> ((i -> i) -> i -> i) -> ((i -> i) -> i -> i) -> o.

zero F :- F = f\ x\ x.
one F :- F = f\ x\ f x.
two F :- F = f\ x\ f (f x).
add X Y Z :- Z = f\ x\ (X f (Y f x)).

main :- two(T), one(O), add T T Z, print Z,
        % add Q T T, print Q. % unification problem outside patern fragment. Flex-flex?
        add T Q T, print Q. % no prob

Some built in elpi files

  • builtin surprising that even very basic stuff is defined in here.
    • i+ i- r+ mod div infix operations. + is overloaded it looks le
    • regular expresions rex. Like Str module
    • print dprint raw terms?
    • quotation? https://github.com/LPCIC/elpi/blob/master/src/elpi-quoted_syntax.elpi lam, app, const, clause, arg
    • same_term same_var var name - geiventbaraibe, ground_term, closed_term, constant,
    • if/3
    • random.int
    • int.map, std.string.map, std.string.set, std.map takes in a comp function,
    • gc functions. trace,counter
  • stdlib
    • mem, exsits, map2, filter, fold, zip, unzip, intersperse, max, findall

std.do notation?

Intro slides

kind term type.
kind ty type.
type app term -> term -> term.
type lam (term -> term) -> term.
type arr ty -> ty -> ty.

pred of i:term o:ty.

of (app H A) T :- of H (arr S T), of A S. 
of (lam F) (arr S T) :- pi x \ of x S => of (F x) T. 

cbn (lam F) (lam F).
cbn (app (lam F) N) M :- cbn (F N) M.

examples in test

  • named clauses :clausename
  • namespaces
  • typeabbrev
  • seperate compilation?
  • w.elpi algorithm W?
  • ndprover

look at teyjus test folders?

external fields. Interesting

a tutorial on lambda prolog and is applications to theorem provin - Amy Felty thesis implementing lambda prolog in ciao prolog

Implementing Type Theory in Higher Order Constraint Logic Programming longer elpi paper. describes chr.

  • macro directives
  • delay directives

HO Unification

see also unification notes

forall exists forall problems Raising vs Skolemization

LF

see also type thery notes

LF is of some relation to lambda prolog (being a prologish system with lambdas in it) although with some different aims. It is dependently typed (pi 2 I think?) in a way different from coq etc.

Twelf. dedukti. lambda - pi https://www.youtube.com/watch?v=XnZTADrsiHQ&list=PL2IiUBbPiuoY22PjECQscRQ81LaR0lgdh&index=2&ab_channel=TYPES2022 dedukti workshop Intermidate “assembly” for interchange between proof formats. Logical framework lambda pi - has dependent types. But not types that depend on types. Pi only quantifies over terms. kocheck - fast rust

beluga is this kind of twelf? beluga paper abella. These are all related I guess? abella might not be LF

Logical frameworks is a general vague concept I think. It is a system in which it is easy to model logical language nad inference rules. but ELF the edburgh logical fraemwork is more specific. Related to Pi2 a corner of the lambda cube.

The Next 700 Challenge Problems for Reasoning with Higher-Order Abstract Syntax Representations

Twelf

https://github.com/chrisamaphone/lf-class

https://cs.mcgill.ca/~bpientka/courses/comp523-08/cd.pdf frank pfenning computation and deduction

http://twelf.plparty.org/live/ http://twelf.org/wiki/About_The_Twelf_Project

Twelf is a depnedelty typed landguage that is also a logic programming langfuage. Because it has a totality checker and termination checker, prolog clauses can be considered proofs.

https://core.ac.uk/download/pdf/82649367.pdf Normalization for the Simply-Typed Lambda-Calculus in Twelf

Rust Chalk Harrop

https://github.com/rust-lang/chalk https://rust-lang.github.io/chalk/book/

http://smallcultfollowing.com/babysteps/blog/2017/01/26/lowering-rust-traits-to-logic/#type-checking-generic-functions-beyond-horn-clauses https://internals.rust-lang.org/t/blog-series-lowering-rust-traits-to-logic/4673/12 internals thread

http://smallcultfollowing.com/babysteps/blog/2017/04/23/unification-in-chalk-part-2/ how chalk handles type normalization

LogTalk

logtalk is it’s own curious kind of universe. github repo It adds object oriented programming to prolog. https://logtalk.org/2009/12/08/lambda-expressions-in-logtalk.html lambdas

Linear Logic Programming

See linear logic prolog rules destroy the body. Good for modeling state Chris martens ceptre

The interaction of linear logic and logic programming was very inlfluential on the concept of focusing

Coinductive Logic Programming

Actually fairly unrelated to inductive logic programming. Dealing with infinite terms and streams.

LTL Model Checking using Coinductive Answer Set programming Completion + loop formula (external support) gives lfp. completion + ? give gfp?

Co-Logic Programming: Extending Logic Programming with Coinduction

COnductive logic programming and it’s applications

:- use_module(library(coinduction)).

:- coinductive p/1.

p([1|T]) :- p(T).

:- use_module(library(coinduction)).

:- coinductive p/0, q/0.
p :- q.
q :- p.

mi(Gs, G) :- member(Gs,G1), subsumes_term(G1,G), 

inductive logic programmingh

Hyper in bratko

Inductive Logic Programming: From Machine Learning to Software Engineering

popper https://arxiv.org/abs/2005.02259 https://github.com/metagol/metagol metagol http://andrewcropper.com/ muggleton https://arxiv.org/pdf/2102.10556.pdf inductive logic programming at 30 hakank examples

Abductive Reasoning in Intuitionistic Propositional Logic via Theorem Synthesis. Includes a g4ip

Hakank symbolic function induction picat

Louise https://github.com/stassa/louise/blob/09a06e081b8933bbdb999424bf1a62673e88a3d6/src/louise.pl line 465 for metainterpreter prove. multi clause inductive learning. “inductive”

Metarules P(A,B) :- Q(A,C), R(C,B). second order qauntified variables

metainterpretive learning (MIL) - a prolgo metainterpreter

Probablistic Logic Programming

https://swi-prolog.discourse.group/t/new-book-second-edition-of-foundations-of-probabilistic-logic-programming/6228

Abductive logic programming

https://en.wikipedia.org/wiki/Abductive_logic_programming

Infer ground facts given results

https://stackoverflow.com/questions/41274692/discussion-about-abductive-logic-programming-vs-answer-set-programming “Any logic programing that supports hypothetical reasoning can support ALP”

assumez(P) :- assertz(P).
assumez(P) :- retract(P), fail.
abducible :- (assumez(amount(glucose,low));assumez(amount(glucose,medium))),
         (assumez(amount(lactose,medium));assumez(amount(lactose,hi))).

feed(lactose) :- amount(glucose,low), amount(lactose,hi).
feed(lactose) :- amount(glucose,medium), amount(lactose,medium).

% ?- abducible, feed(lactose), listing(amount/2).

Abduction in Logic Programming - Denecker Kakas

A different variation https://swi-prolog.discourse.group/t/non-monotonicity-defeasibility-strategies-and-other-stuff/6038/23

/* SWI-Prolog 9.1.0 */
solve(true) --> !.
solve((A,B)) --> !, solve(A), solve(B).
solve(A) --> {abducible(A)}, !, [A].
solve(H) --> {clause(H, B)}, solve(B).

Equational Logic Programming

https://www.cs.ox.ac.uk/files/3444/PRG116.pdf

https://cseweb.ucsd.edu/~goguen/sys/eqlog.html eqlog

See Maude in term rewriting INDUCTIVE EQUATIONAL LOGIC PROGRAMMING higher order equationa logic programming

Logic programming, functional programming, and inductive definitions Another argument in favor of inductive definitions, not first order logic. paulson

Functional Logic Programming

See Curry

Misc

tar parsing in prolog rethinking prolog - oleg unification variables are an optimization

An interactive semantics of logic programming https://twitter.com/davidad/status/1567780900284387329?s=20&t=jW6J2ukuizpdkb11CCe3bQ

ICLP proceedings - lots of ASP stuff

pyswip

https://github.com/mthom/scryer-prolog/pull/838 scryer as a library?

.type Lifted = Lit {x : symbol} Y {x : Lifted, y : Lifted}

.decl r(x : Lifted, y : Lifted) .decl a(x : Lifted) r(x1, $Y(x1,x2)), a($Y(x1,x2)) :- r(x1,x2).

defeasible logic programming

O-keefe - An Elementary Prolog Library some suggestions about unicode and other test stuff. Higher order operators

50 years of prolog and beyond

https://github.com/Anniepoo - Annie ogborn has some cool seeming tutorials

https://github.com/philzook58/lips-minikanren trying to use lips sccheme for minikanren in browser

https://github.com/aprolog-lang/aprolog alpha prolog. Nominal logic. What the heck is that again?

old title: prolog, scheme,racket, lambdaprolog, What the Hell is LogicT / Guanxi notes Minikanren, Unification, Resolution

BAP universal valures OCaml typeclasss system Oleg modellign typeclasses as prolog https://inbox.ocaml.org/caml-list/20190904152517.GA2014@Melchior.localnet/ http://okmij.org/ftp/ML/index.html#trep

canonical structures and unification hints. Can we make a metaintepreter for this?

Hmm. Gauntlet thrown. Byrd is not sure how prolog can? do this stuff. https://www.youtube.com/watch?v=aS8oj2GXras&feature=youtu.be&ab_channel=LecturesbyProf.EadesatAU XSB prolog “complicated tabling and abstract interpretation tabling has a notion of subsumption? when are two calls the same? conductive logic over streams tree automata lvars. fixpoint over lattices

meta interpeter, sure. That’s one way to get com,plketeness What if one defunctionalizes the minikanren pause mechanism. What does it look like? Oh there’s a first order kanren paper that does this

Hmm. Could one do barliman style in lambda prolog? F, ex1, ex2, ex3. 2012 quine ppaper 2017 barliman paper

Tree automa, lvar, language for fixpoints. Not depth first, not breadth first, coniductive logic programming UT Dallas Abstract Interpretation ~ tabling. What does that mean?

nada Amin lambda kanren defeasible logic

proof checker running backwards

Scheme workshop

https://git.sr.ht/~sforman/Prolog-Junkyard/tree/master/miscellaneous/itc.pl interval tree clocks in prolog https://www.metalevel.at/trs/ knuth bendix completion https://arxiv.org/abs/1706.00231 auto differentiating using constraint handling rules

http://adam.chlipala.net/papers/MakamICFP18/MakamICFP18.pdf prototpying functional language using higher order logic programing makam chlipala https://www.tweag.io/blog/2019-11-28-PCF-makam-spec/

Egraphs in prolog Unification variables give native union find data strcuture Hash consing - ?? equal( (t, E1), (t2,E2) :- E1 = E2. % union find joining

Great. But then we need to possibly union find parents

t(x,y) ====> t(E1,E2)

equal( t1, t2 ) :- lookup( t1 )

file:///home/philip/Downloads/ODonnell1987_Chapter_Term-rewritingImplementationOf.pdf

https://core.ac.uk/download/pdf/81972151.pdf Logic programming with equations.

Eqlog - goguen and mesegaer?

assert_equal(EGraph, ,EGraph’) :- lookup(t1, E1), lookup(t2, E2), E1 = E2, EGraph’ = EGraph assert_eqyal( ) :- lookup(t1, E1), lookup(t2, E2)

rebuild(EGraph, EGraph’ ) :-

rebuild(EGraph) :- EGraph = ( , Map )

To miss a parent inference isn’t wrong, it’s just wasteful

If we do no hash consing, we do have to store every known term. We also need a map from Eclasses back to terms don’t we?

What if we broke apart. are association lists so bad? f(g(y)) ===> [ f(G) => F , g(Y) => G, y => Y ]

Or we could use the HEADS as keys in assoc. and then assoc list the leftovers. This isunder the assumptyiong that variables

== for search [ == , ]

We may want to prune duplicates occasionally If we interweave

Can we even achieve apttern matching without an index from ENodes => terms? Yessss…..? Given a pattern ( EClass, g(f(A)) ) We can look in g/1 for Eclass on the right hand side. For those matches, we build a subproblem of matching the pieces of the left hand size ENodes.

With tabling? Tabling gives us some kind of memoization

equal?(Egraph, t1,t2) :- lookup(EGraph, t1, E1), lookup(EGraph, t2, E2). equal?(arg1( ) :- eqwal?(args1, args2), assert_equal(EGraph, t1, t2)

Stratified prolog predciates. This euqation paper mentions this and this tabling thing mentions it https://www.swi-prolog.org/pldoc/man?section=tabling-non-termination

Datalog and program analysis https://www2.cs.sfu.ca/CourseCentral/721/jim/DatalogPaper.pdf - What you always wanted to know about datalog http://rightingcode.org/tutorials/popl20/ popl 2020 - reasoning tools using llvm and z3 Z3 has a datalog style reasoning engine in it

First you need to get a program as a mapping

The Chase - get existentials in the head. Wisnesky says somethign to do with jkabn extensions and lawvere theory What was up with the notion of quantifiers as adjoints anyway? Maybe also some kind of completion? Datalog +- Formulog Bap KB as datalog What

Typeclasses vs Canonical Structures. I don’t get it. Could I make a model? Maybe in prolog? Diamond problem Inheritance. What are typeclasses? “Kind of like prolog” Things are incompatible for some reason? Canonical structures add unification hints? https://hal.inria.fr/hal-00816703v1/document canonical structures ofr the working coq user

So what is the synctatic condition that precludes search?

  1. nonoverlapping

Would it be more interesting to just What am I understanding here?

Canonical structures was suggesting that some kind of inconstency would manifest. Maybe lambda prolog gets us closer. We can have existential variables and such.

extensible unification

nat === (carrier nat_abGrp) :- A === A.

nat == carrier(Eq) :- Eq == natEq. carrier(Eq,nat).

nonunifiable(L) :- all( /= , L ).

% haskell has no backtracking % every case has a cut. eq(tup()) :- !. eq(tup(A,B)) :- !, eq(A), eq(B)

%

eq(int, prim__equal_int). eq(tup(A,B), lambda( x, case( x, tup(a,b) => bool_and( apply(EqualA, a) , apply(EqualB, b ) ) ) ) ) :- eq(A, EqualA), eq(B, EqualB). eq(tup(A,B), comp(and, par(EqualA,EqualB)))

eq(int). eq() eq(tup(A,B)) :- eq(A), eq(B).

ord(bool). ord().

functor( maybe ) functor(compose(F,G)) := functor(F), functor(G)

traversable :-

% superclass class(ord(A)) :- call(ord, A), class(eq(A))

“diamonds appear evertwhere. See pullbacks” ~ Hale

Transcribing rules to prolog and coq. In Coq the cookbook is that you make an inductive. One constrctor per rule. If you call eauto, coq will perform prolog style backchaining search. See Chlipala You could write these as functions rather than as The data type is the proof structure.

Prolog is the same thing. It’s less imposing than coq though. You make predicates and :- for each condition. Prolog has built in nondtermisnion and unification.

Tying rules in prolog. I did the point-free simply typed lambda calculus type(fst, tup(A,B), A).

type(Gamma - plus(A,B) : nat ) :-  
type(Gamma - A : nat), type(Gamma - B : nat)
type(Gamma, lam(X, Body) - fun(A,B )) :- type() , type( [X Gamma]

http://www.coli.uni-saarland.de/projects/milca/esslli/comsem.pdf computational semantics. some interesting material here https://samples.jbpub.com/9780763772062/PrologLabBook09.pdf - prolog experiment in discretem mathemtiacs logic and computatilbitly

They start with defining a set of atoms like

var(x). var(y).

forall(x, stupid(x)).

https://www.youtube.com/watch?v=RwBiHLoQ3E4&ab_channel=PapersWeLove niko matsakis - lambda prolog

https://rust-lang.github.io/chalk/book/recursive.html chalk. harrop formula

The Otten sequent calculus prover is very similar to a meta circular intepreter for prolog

Horn clauses (or harrop formula)

Lambda-mu calculus and prolog. Focusing is relevant. Prolog has an imperative reading

Executing the Algebra of Programming

type(fst, prod(A,B) -> A ). type(snd, prod(A,B) -> B ). type(id, A -> A).

interp(fst, tup(A,B), A ). interp(snd, tup(A,B), B). interp(id, A , A). interp(fan(F,G), A, tup(B,C)) := interp( F, A, B) , interp(G, B, C).
interp(comp(F,G), A, C ) :- interp( F, A, B ), interp(G, B, C). interp(conv(F), A, B) :- interp(F, B, A). interp(cata(F) , fix(A) , C ) :- interp(map(cata(F)), A, B) , interp(F, B, C)

% map instance for ListF interp(map(F), cons(A,B) , cons(A,B2) ) :- interp(F, B, B2). interp(map(F), nil, nil).

% in constrast to map instance for list interp(map(F), [], []). interp(map(F), [X | XS], [Y | YS] ) :- interp(F, X, Y), interp(map(F), XS, YS).

% convert to listf form listf([], nil). listf([A | XS ], cons(A,fix(L)) ) :- listf(XS, L).

Different style

cata(F, fix(A), C) :- map(cata(F), A, B), call(F, B, C). fst(pair(A,B), A).

comp(F, G, A, B) :- call() , call()

% define what it means to be a functional relation functional( fst, fst). functional( snd, snd). function( )

% prolog is database stuff ~ relation algebra. It makes perfect sense.

interp( map )

Converting prolog to abstract machine form step( Stack Result ) :-

oprolog veriffication

Warrne and Maier 1988 textboook proplog - propsitional goals only. datalog - constants only prolog - functional symbols

% tail recursive formulations factorial(N,F) :- factorial(0, N,1,F) factorial(I,N,T,F) :- I < N, I1 is I+1, T1 is T * I1, factorial(I1, N, T1,F) factorial(N,N,F,F).

hitchikers guide to prolog machine https://drops.dagstuhl.de/opus/volltexte/2018/8453/pdf/OASIcs-ICLP-2017-10.pdf

ref a ref b

https://formal.iti.kit.edu/beckert/leantap/leantap.html Beckert posegga leantap https://link.springer.com/chapter/10.1007%2F3-540-36377-7_6 metacircular abstarct interpretation in prolog https://www.metalevel.at/acomip/

Pfenning constructive logic course http://www.cs.cmu.edu/~fp/courses/15317-f17/schedule.html Programs as proof search He considers the prolog predictaes as the judgements themeselves As compared to considering the pile of predicates as entedcedents of a sequent Bottom up search ius forward reasoning, top down is backward reasonining bidrectional type checking - https://arxiv.org/pdf/1908.05839.pdf

Prolog semantics - Since prolog can not terminate you can take that denotational semantics perspective. https://www.swi-prolog.org/pldoc/man?section=WFS well founded semantics. - True/False/unknown https://en.wikipedia.org/wiki/Well-founded_semantics http://www.cse.unsw.edu.au/~cs4415/2010/resources/gelder91wellfounded.pdf http://dai.fmph.uniba.sk/~sefranek/links/FixPointSurvey.pdf fixpoitn semantics for logic programming a survey. melvin fitting Melvin fitting has a number of interesting bokos http://melvinfitting.org/

So prolog already has belnap bools in it.

What does it look like to mix prolog with exact reals?

Modes and determinism annotations.

Kowalski - hgistory of prolog https://www.researchgate.net/publication/277670164_History_of_Logic_Programming

tabling - Tabling is some kind of memoization. It is connected to bottom up strategies swi prolog https://www.swi-prolog.org/pldoc/man?section=tabling https://www.metalevel.at/prolog/memoization https://biblio.ugent.be/publication/6880648/file/6885145.pdf tabling as a library with delimited control https://www.cambridge.org/core/journals/theory-and-practice-of-logic-programming/article/abs/delimited-continuations-for-prolog/DD08147828169E26212DFAF743C8A9EB delimitted continuations for prolog https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.294.9440&rep=rep1&type=pdf XSB extending prolog with tabled logic programming. XSB prolog is the most advanced tabling implementation (still? maybe that is old information) https://github.com/SWI-Prolog/tabled-prolog-book warren wrote a book draft https://www3.cs.stonybrook.edu/~warren/xsbbook/book.html ok This may not be being finished anytime soon Tabled typeclass resolution

http://www.covingtoninnovations.com/mc/plcoding.pdf - coding guidelines for prolog

https://swi-prolog.discourse.group/t/algebra-in-prolog/1849 - algerba in prolog https://github.com/maths/PRESS prolog equation solving system

scryer prolog - a rust based implemnentation . https://github.com/mthom/scryer-prolog Interesting links Precise garbage collection in Prolog “Can Logic Programming Execute as Fast as Imperative Programming?” - Peter van Roy indexing dif/2 https://arxiv.org/abs/1607.01590 calls for if_ predicate debray allocator

https://wiki.nikitavoloboev.xyz/programming-languages/prolog notes

https://mercurylang.org/documentation/documentation.html mercury

Names: Kowalski schrijver Nuemerekl Markus triska Wielemaker Warren Miller Nadathur melivn fitting

https://www.metalevel.at/prolog/theoremproving presburger and resolution https://www.youtube.com/watch?v=b2Px7cu2a68&feature=youtu.be&ab_channel=ThePowerofProlog term rewriting in prolog

https://dl.acm.org/doi/book/10.1145/3191315 Declarative Logic Programming: Theory, Systems, and ApplicationsSeptember 2018 Has another WAM tutorial in it

So

prolog modes

https://www.metalevel.at/prolog/attributedvariables attributed variables

https://sicstus.sics.se/sicstus/docs/latest4/pdf/sicstus.pdf sicstus prolog manual

Exact reals. The semantics already has this three valued logic flavor

partition([X,Y]).
partition([X,Y]) :- partition([[X, X+Y / 2]]).
partition([[X,Y]) :- partition( [X+Y/2 , Y  ]).

interval(X), stuff.


sqrt()

Using prolog multiple answer streams is kind of like relying on haskell laziness. It’s this intrinsic lANGUAGE construct being used as a streaming datatpe Unclear if wise. Fast but inflexible and unclear. Too clever for own good.

clpq clpr

delmittied dontinuation. The contiuation is a goal stack. But is the continuation also catching choice point stack?

homoiconic. Term syntax is identical to

Maier and Warren

T_p, the immediate conseqeunce operator. Applying one round of prolog/datalog/ etc rules

Proplog

SMT-log Datafun - a fiunctional datalog https://www.cl.cam.ac.uk/~nk480/datafun.pdf https://www.youtube.com/watch?v=gC295d3V9gE datalog - stephenel diehl https://www.stephendiehl.com/posts/exotic04.html uZ - datalog in Z3 souffle doup

What are appli8cations of datalog program analysis. That points to analysis tutorial

Datalog ~ polynomial time in some sense. So application wise, it can be fixed point computations Or we can encode

send more money num(0) num(1) num(2) adder( )

https://sites.google.com/site/pydatalog/Online-datalog-tutorial nqueens - they unroll it significantly.

Some examples http://cse.unl.edu/~riedesel/pub/cse413/des/doc/manualDES.pdf

ok farmer goat. As a path search basicaly

state(goat, foxm, chienbk, ) :-

Parsing a grammar. This is an odd one. Encode positions of string in database. t(1,a,2). t(2,b,3). t(3,a,4). t(4,b,5). t(5,a,6). a(F,L) :- t(F,a,L). a(F,L) :- a(F,M), t(M,b,L). a(F,L) :- a(F,M), t(M,a,L). DES> a(1,6) { a(1,6) } Info: 1 tuple computed.

Games?

Domain modeling with datalog https://www.youtube.com/watch?v=oo-7mN9WXTw&ab_channel=%23pivorakLvivRubyMeetUp

https://github.com/ptarau/TypesAndProofs More theorem provers

binprolog - analog of CPS. Add a new parameter to every predicate

LD resolution

monoid of clauses - unfolding bin(A :- BBBB) = (A>C :- BBBB>C)

2020/9 Propagators.

Kmett. The Art of the Propagator https://dspace.mit.edu/bitstream/handle/1721.1/44215/MIT-CSAIL-TR-2009-002.pdf?sequence=1&isAllowed=y

The gecode manual https://www.gecode.org/doc-latest/MPG.pdf

Adding constraints or smt. I have a seperate store for them (in the state)? explicit probe annotations to check still satisfiable.

9/1/20

I was looking at prolog again. Jesus nothing changes.

What are the good programs. The art of prolog has an open access link

Difference lists are cool. They really are similar ot hughes lists. Or al ist that keeps a pointer to it’s end

Using it for theorem proving is cool. Where are those examples? The lambda prolog book has some exmaples. There is a propsitional satisfiabilioty prover in art of prolog. Propositional solver in powr of prolog https://www.metalevel.at/prolog/theoremproving http://vidal-rosset.net/g4i_prover.html http://jens-otten.de/tutorial_cade19/slides/prover_tutorial_CADE19_2.pdf

nomial logic programming https://arxiv.org/pdf/cs/0609062.pdf alphaprolog. Chris mentioned this nominal thing as nother way of dealing with binders

Datalog - souffle. Reading gorup paper https://arxiv.org/pdf/1207.5384.pdf. Sttratification. relations indexed on program point/ abstract state of program point. Interval analysis encoded in binary to datalog?

425/20 I was playing with prolog recently. Pretty cool

What is the deal with scheme and racket? i just don’t have the revelation.

I was looking at disjoint set data structures again https://en.wikipedia.org/wiki/Disjoint-set_data_structure#cite_note-Conchon2007-9

Kmett’s latest propgator talk was mentioning using group actions somehow in unification? https://www.youtube.com/watch?v=s2dknG7KryQ

george wilson - https://www.youtube.com/watch?v=nY1BCv3xn24 intutition for propagators.

http://hackage.haskell.org/package/discrimination-0.3 What the hell is this? Fritz Henglein.

http://conway.rutgers.edu/~ccshan/wiki/blog/posts/WordNumbers1/

https://www.reddit.com/r/haskell/comments/a9yze4/is_there_an_arraylike_data_structure_which_gives/

interesting comments

https://www.lri.fr/~filliatr/puf/ a coq formalization of a persistent union find data structure. They use persistent arrays, which do some kind of rebalancing operation.

Persistent data structures. Wassup with them?

2019

Combine LogicT with an OSQP monad for branch and bound

http://overtond.blogspot.com/2008/07/pre.html

https://github.com/dmoverton/finite-domain

https://blog.plover.com/prog/haskell/monad-search.html

https://www.schoolofhaskell.com/user/chowells79/even-more-money

https://blog.jle.im/entry/unique-sample-drawing-searches-with-list-and-statet

Select Monad

Branch and Bound

Ed kmett is up to funny business

https://www.youtube.com/watch?v=ISNYPKiE0YU&t=916s

Propagators

https://www.sciencedirect.com/science/article/pii/S1571066105805444

Typed logical variables in haskell by Claesson and Ljundogjgklds

https://people.seas.harvard.edu/~pbuiras/publications/KeyMonadHaskell2016.pdf

Also by claesson, the key monad paper.

http://www.cse.chalmers.se/~koen/

Interesting guy. He is coqauthor with hughes on quickcheck.

Atze van der Ploeg is both Key monad paper and reflection without remorse. They have an FRP paper that sounds interesting

He metnioend a number of interesting things. He’s doing very reference heavy code? Kanren.

He mentoins intension vs extensional equality. Judgemental eqaulity is the one inside the type checker. Is it ~ ? And intwnsional equality is the one within the language itself, that is :~: . Extensional. nuPRL starts with an untyped lambda calculus and then you teahc the system typing derivations? What is nuPRL’s deal

Union-find algortihm - as part of unification?

nerualkanren, synquid. Two program synthetsis projects. Synquid uses liquid typing

Oleg Shen paper using efficient charing for logic programming

https://www-ps.informatik.uni-kiel.de/~sebf/data/pub/icfp09.pdf

http://sebfisch.github.io/explicit-sharing/

Maybe sebastein fischer is the name I should assicuate more storngly?

https://sebfisch.github.io/research/

The Curry language. Haskell + logic. Egison has weirdo patterns too. Multiple patterns can match? Non-linear pattern require. Realted to view patterns? view patterns are a weirod syntax. realted to guard patterns. DOn’t need to introduce intermediate names

https://people.eecs.berkeley.edu/~necula/autded/lecture15-tactics.pdf

View patterns might be a way of getting something like the Agda .x syntax, where something is forced to be something by unification?

https://gitlab.haskell.org/ghc/ghc/wikis/view-patterns#Relatedwork

Kmett mentions a “octagonal” domain. What the hell is he talkiing about. He also moentions interval and polyhedral which makes more sense

https://www-apr.lip6.fr/~mine/publi/article-mine-HOSC06.pdf

Huh. Interesting. It is in between power of polyhedral and interval. allows sums/differences of two variables. I think I’ve heard this called difference logic. This paper also mentions galois connections. Galois connections are studied in the context of abstract interpetation?

https://arxiv.org/pdf/1807.08711.pdf

Using Agda and galois connections. Interestign example of sets of even and odd numbers.

This other paper does an interesting thing of defining division in terms of a glaois connection. Might we also get a good definition of fractions that way? a fraction t ~ m/n is a “number” such that forall z. x <= y -> x t n <= y * m? There is an notion of negative numbers, fractions, etc as reified weirdo operations that can’t be solved yet. Hmm. Yeah, This jives with galois theory doesn’t it. That book I was reading was talking about field extensions. Or number systems being invented to make more equations solvable. The reals make ? solvable. fractions make division solvable. Complex makes roots solvable. finite field extensions make simple algerbaic equations solvable.

https://www.sciencedirect.com/science/article/pii/S1567832612000525

% —————————————————————– % leanseq.pl - A sequent calculus prover implemented in Prolog % —————————————————————–

% operator definitions (TPTP syntax)

:- op( 500, fy, ~). % negation :- op(1000, xfy, &). % conjunction :- op(1100, xfy, ‘|’). % disjunction :- op(1110, xfy, =>). % implication :- op( 500, fy, !). % universal quantifier: ![X]: :- op( 500, fy, ?). % existential quantifier: ?[X]: :- op( 500,xfy, :).

% —————————————————————– prove0(F) :- prove([] > [F]). % —————————————————————–

Oliveira, that same guy as the Search monad and some other thingds

I suppose perhaps there is something similar happening in functional programming. to make recrsuively defined functions solvable, you need to extemnd the language with a Y-combinator or some of fixed point operator.

Interval airthemteic in a theorem prover. That is a way of getting sets. Min and Max. Interesting

This is also a pleasant by Backhouse

http://www.cs.nott.ac.uk/~psarb2/G53PAL/FPandGC.pdf

indirect equality

m=n === forall k. k <= m <=> k <= n

Galois connection between convex hulls of integer points? There is a sequence of abstractions for integer programming. You can turn dimension by dimension into integer so there are 2^D different domains to talk about. And there is this grid of connections that forms a lattice itself? Like the top is the completely R^D, and the bottom is Z^D. Using these connections is the branch and bound procedure.

Floor and Ceil are also galois connections. Maybe also round? I had been thinking in terms of ibjects being individual numbers, not convex sets of numbers. Convex sets does tie in much better to categorical thinking

http://www.cs.tau.ac.il/~msagiv/courses/asv/absint-1.pdf

An interesting paper tutorial on galois connections. V cool.

monotone functions between are like natural tranfromations?

One place used dup as the adjunction to max.

There may be more to galois connections than adjunctions, since they are assuming a meet and join operation. Some interesting doncturctions like the porduct of galois connections.

- Edward Kmett - Logic Programming à la Carte

minikanren is strongly unification based. it is using search for the unification map. In simplest form [UniMap]

https://github.com/JeffreyBenjaminBrown/learn-logict

https://www.msully.net/blog/2015/02/26/microkanren-%CE%BCkanren-in-haskell/

https://gup.ub.gu.se/file/207634 logic typed variable claesson

http://dev.stephendiehl.com/fun/006_hindley_milner.html#unification unification.

Pure, Declarative, and Constructive Arithmetic Relations (Declarative Pearl)

https://ro-che.info/articles/2017-06-17-generic-unification unification-fd .

https://winterkoninkje.dreamwidth.org/100478.html

unification-fd inserts itself in the same position Fix does.

Lattices. Subsumption lattice for unufication. More and less general partial order. meet and join. top and bottom

Notes from 2017 -Resolution and unification

So I was learning about about minikanren. There are some videos online. Minikanren is a logic programming language (like Prolog) that embeds easily into other languages because it has a small core.

Logic programming is weird mainly (partially) because you define relations rather than functions. Relations do not have a input output relationship like functions do. In a sense they are like functions where you get to choose which thing is output and which things are input. Pretty crazy.

The most obvious way to do this to me is to make every function a bag of functions. Just write one function for every possible choice of output variable. These functions may need to be non-deterministic, outputting multiple possibilities, for example for y = x^2 gives x = +sqrt(y) or -sqrt(y). However, this isn’t really how logic programs are written. Instead they deduce how to use a relation as a function.

I find most intro to prolog stuff off putting, talking about socrates(man), when that is not a problem I have ever given a crap about.

Resolution is the classical logic version of function composition.

a-> b and b->c can be combined easily into a function a-> c.

Implication in classical logic is weird. It translates to

a->b ====> (not a) or b

When a is true, b has to be true. When a is false, b can be true or false.

For the statement a implies b to be true then it needs to evaluate to false only when a is true and b is false

not (a and (not b))

using De Morgan’s law you can distribute nots turning ands into ors.

Then that becomes

(not a) or b.

If you have

((not a) or b) and ((not b) or c)

it does not matter what b is really because either b or not b will be false

so this is equivalent to

(not a) or c

since at least one has to be true to make the whole expression true

Unification

Predicate logic allows propositions to depend on variables. Think Prolog.

A simple question given two expressions is whether there are values of the variables that make the expressions equal.

I do this in my head sometimes when I’m looking a a parametric haskell type.

The type variables need to be matched up with two different expressions.

Sometimes this can be too hard to do in my head, especially when the type variables end up being functions. So there is need for an algorithmic procedure. I mean, also because you want a computer to do it.

First off, let’s suppose we have the two expressions in tree form.

Nodes can be Or, And, Predicates, Not.

We’ll want them in some canonical form too so that we don’t get tripped up by the commutativity or distributivity of Boolean operators

We need to collect up a bucket of deduced equivalences as we go down the trees of the two expressions. Whenever we hit a variable, we check if we have a substitution in our bucket. If so, replace it. If not, we put into our equivalences that variable is equal to whatever the other expression has in that place. When we hit something that trips us up, we announce failure and the expressions couldn’t be unified.

The prolog algorithm is roughly guess a pair of terms in the goal (the executing state) and knowledge base (the code base) that will unify. Try to unify them. If they do, then use resolution to get rid of those terms.

Like what if we reflected into Eisenberg’s Stitch? Or what was Weirich’s thing? People have been talking about intrinsically typed system-F lately.