Library RelationAlgebra.lsyntax


lsyntax: syntactic model for flat structures (lattice operations)


Require Export positives.
Require Import comparisons lattice lset sups.
Set Implicit Arguments.

Free syntactic model


Section s.
Variable A: Set.

Boolean lattice expressions over a set A of variables
Inductive expr :=
| e_bot
| e_top
| e_cup (e f: expr)
| e_cap (e f: expr)
| e_neg (e: expr)
| e_var (a: A).

level of an expression: the set of operations that appear in that expression
Fixpoint e_level e :=
  match e with
    | e_botBOT
    | e_topTOP
    | e_cup x yCUP + e_level x + e_level y
    | e_cap x yCAP + e_level x + e_level y
    | e_neg xBL + e_level x
      
    | e_var _MIN
  end%level.

Section e.
Context {X: ops}.
Variable f: A X.

interpretation of an expression into an arbitray Boolean lattice structure, given an assignation f of the variables
Fixpoint eval e: X :=
  match e with
    | e_botbot
    | e_toptop
    | e_cup x yeval x \cup eval y
    | e_cap x yeval x \cap eval y
    | e_neg x! eval x
    | e_var af a
  end.
End e.

Section l.
Variable l: level.

(In)equality of syntactic expressions.

Rather than defining (in)equality of syntactic expressions as inductive predicates, we exploit the standard impredicative encoding of such predicates: two expressions are equal (resp., lower or equal) iff they are equal (resp., lower or equal) under any interpretation.
These definitions are parametrised by the level l at which one wants to interpret the expressions: this allows us to capture once and for all the equational theories of each flat structure.

Definition e_leq (x y: expr) := X (L: laws l X) (f: A X), eval f x <== eval f y.
Definition e_weq (x y: expr) := X (L: laws l X) (f: A X), eval f x == eval f y.

by packing syntactic expressions and the above predicates into a canonical structure, we get all notations for free
Canonical Structure expr_ops := {|
  car := expr;
  leq := e_leq;
  weq := e_weq;
  cup := e_cup;
  cap := e_cap;
  neg := e_neg;
  bot := e_bot;
  top := e_top
|}.

we easily show that we get a model so that we immediately benefit from all lemmas about flat structures

Global Instance expr_laws: laws l expr_ops.
Proof.
  constructor; try right. constructor.
   intros x X L f. reflexivity.
   intros x y z H H' X L f. transitivity (eval f y); auto.
   intros x y. split.
    intro H. split; intros X L f. now apply weq_leq, H. now apply weq_geq, H.
    intros [H H'] X L f. apply antisym; auto.
   intros Hl x y z. split.
    intro H. split; intros X L f; specialize (H X L f); simpl in H; hlattice.
    intros [H H'] X L f. simpl. apply cup_spec; auto.
   intros Hl x y z. split.
    intro H. split; intros X L f; specialize (H X L f); simpl in H; hlattice.
    intros [H H'] X L f. simpl. apply cap_spec; auto.
   intros x X L f. apply leq_bx.
   intros x X L f. apply leq_xt.
   intros Hl x y z X L f. apply cupcap_.
   intros Hl x X L f. apply capneg.
   intros Hl x X L f. apply cupneg.
Qed.

the interpretation function is an homomorphism, so that it preserves all finite sups and infs
Lemma eval_sup I (J: list I) (f: I expr) (X: lattice.ops) (g: A X):
  eval g (sup (X:=expr_ops) f J) = \sup_(i\in J) eval g (f i).
Proof. apply f_sup_eq; now f_equal. Qed.

Lemma eval_inf I (J: list I) (f: I expr) (X: lattice.ops) (g: A X):
  eval g (sup (X:=dual expr_ops) f J) = \inf_(i\in J) eval g (f i).
Proof. apply (f_sup_eq (Y:=dual X)); now f_equal. Qed.

e_var is a unit for the underlying monad
Lemma eval_var (e: expr_ops): eval e_var e = e.
Proof. induction e; simpl; congruence. Qed.

End l.

End s.
Arguments e_var [A] a.
Arguments e_bot [A].
Arguments e_top [A].

Bind Scope last_scope with expr.
Delimit Scope last_scope with last.

additional notations, to specify explicitly at which level expressions are considered, or to work directly with the bare constructors (by opposition with the encapsulated ones, through lattice.ops)
Notation expr_ l := (car (expr_ops _ l)).
Notation "x <==_[ l ] y" := (@leq (expr_ops _ l) x%last y%last) (at level 79): ra_scope.
Notation "x ==_[ l ] y" := (@weq (expr_ops _ l) x%last y%last) (at level 79): ra_scope.

Infix "\cup" := e_cup: last_scope.
Infix "\cap" := e_cap: last_scope.
Notation "! x" := (e_neg x): last_scope.

Comparing expressions

we get a cmpType on expressions if the set of variable is such (currently used in ugregex_dec)

Section expr_cmp.
Context {A: cmpType}.
Fixpoint expr_compare (x y: expr A) :=
  match x,y with
    | e_bot, e_bot
    | e_top, e_topEq
    | e_var a, e_var bcmp a b
    | e_cup x x', e_cup y y'
    | e_cap x x', e_cap y y'lex (expr_compare x y) (expr_compare x' y')
    | e_neg x, e_neg yexpr_compare x y
    | e_bot, _Lt
    | _, e_botGt
    | e_top, _Lt
    | _, e_topGt
    | e_var _, _Lt
    | _, e_var _Gt
    | e_cup _ _, _Lt
    | _, e_cup _ _Gt
    | e_cap _ _, _Lt
    | _, e_cap _ _Gt
  end.

Lemma expr_compare_spec: x y, compare_spec (x=y) (expr_compare x y).
Proof.
  induction x; destruct y; try (constructor; congruence).
  - eapply lex_spec; eauto. intuition congruence.
  - eapply lex_spec; eauto. intuition congruence.
  - simpl; case IHx; constructor; congruence.
  - simpl; case cmp_spec; constructor; congruence.
Qed.

Canonical Structure cmp_expr := mk_simple_cmp _ expr_compare_spec.

variables appearing in an expression (A needs to be a cmpType so that the resulting list is without duplicates)
Fixpoint vars (e: expr A): list A :=
  match e with
    | e_bot
    | e_top[]
    | e_cup x y
    | e_cap x yunion (vars x) (vars y)
    | e_neg xvars x
    | e_var x[x]
  end.
End expr_cmp.