typer.ml 32.3 KB
Newer Older
1
(* TODO:
2
 - rewrite type-checking of operators to propagate constraint
3
4
 - optimize computation of pattern free variables
 - check whether it is worth using recursive hash-consing internally
5
6
*)

7

8
9
(* I. Transform the abstract syntax of types and patterns into
      the internal form *)
10
11
12

open Location
open Ast
13
open Ident
14

15
module S = struct type t = string let compare = compare end
16
module StringSet = Set.Make(S)
17
module TypeEnv = Map.Make(S)
18
module Env = Map.Make(Id)
19

20

21
22
exception NonExhaustive of Types.descr
exception Constraint of Types.descr * Types.descr * string
23
exception ShouldHave of Types.descr * string
24
exception WrongLabel of Types.descr * label
25
exception UnboundId of string
26
27

let raise_loc loc exn = raise (Location (loc,exn))
28
29


30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
(* Eliminate Recursion, propagate Sequence Capture Variables *)

let rec seq_vars accu = function
  | Epsilon | Elem _ -> accu
  | Seq (r1,r2) | Alt (r1,r2) -> seq_vars (seq_vars accu r1) r2
  | Star r | WeakStar r -> seq_vars accu r
  | SeqCapture (v,r) -> seq_vars (IdSet.add v accu) r

type derecurs_slot = {
  ploc : Location.loc;
  pid  : int;
  mutable ploop : bool;
  mutable pdescr : derecurs option
} and derecurs =
  | PAlias of derecurs_slot
  | PType of Types.descr
  | POr of derecurs * derecurs
  | PAnd of derecurs * derecurs
  | PDiff of derecurs * derecurs
  | PTimes of derecurs * derecurs
  | PXml of derecurs * derecurs
  | PArrow of derecurs * derecurs
  | POptional of derecurs
  | PRecord of bool * derecurs label_map
  | PCapture of id
  | PConstant of id * Types.const
  | PRegexp of derecurs_regexp * derecurs
and derecurs_regexp =
  | PEpsilon
  | PElem of derecurs
  | PSeq of derecurs_regexp * derecurs_regexp
  | PAlt of derecurs_regexp * derecurs_regexp
  | PStar of derecurs_regexp
  | PWeakStar of derecurs_regexp

let rec hash_derecurs = function
  | PAlias s -> s.pid
  | PType t -> 1 + 17 * (Types.hash_descr t)
  | POr (p1,p2) -> 2 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | PAnd (p1,p2) -> 3 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | PDiff (p1,p2) -> 4 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | PTimes (p1,p2) -> 5 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | PXml (p1,p2) -> 6 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | PArrow (p1,p2) -> 7 + 17 * (hash_derecurs p1) + 257 * (hash_derecurs p2)
  | POptional p -> 8 + 17 * (hash_derecurs p)
  | PRecord (o,r) -> (if o then 9 else 10) + 17 * (LabelMap.hash hash_derecurs r)
  | PCapture x -> 11 + 17 * (Id.hash x)
  | PConstant (x,c) -> 12 + 17 * (Id.hash x) + 257 * (Types.hash_const c)
  | PRegexp (p,q) -> 13 + 17 * (hash_derecurs_regexp p) + 257 * (hash_derecurs q)
and hash_derecurs_regexp = function
  | PEpsilon -> 1
  | PElem p -> 2 + 17 * (hash_derecurs p)
  | PSeq (p1,p2) -> 3 + 17 * (hash_derecurs_regexp p1) + 257 * (hash_derecurs_regexp p2)
  | PAlt (p1,p2) -> 4 + 17 * (hash_derecurs_regexp p1) + 257 * (hash_derecurs_regexp p2)
  | PStar p -> 5 + 17 * (hash_derecurs_regexp p)
  | PWeakStar p -> 6 + 17 * (hash_derecurs_regexp p)

let rec equal_derecurs p1 p2 = (p1 == p2) || match p1,p2 with
  | PAlias s1, PAlias s2 -> s1 == s2
  | PType t1, PType t2 -> Types.equal_descr t1 t2
  | POr (p1,q1), POr (p2,q2)
  | PAnd (p1,q1), PAnd (p2,q2)
  | PDiff (p1,q1), PDiff (p2,q2)
  | PTimes (p1,q1), PTimes (p2,q2)
  | PXml (p1,q1), PXml (p2,q2)
  | PArrow (p1,q1), PArrow (p2,q2) -> (equal_derecurs p1 p2) && (equal_derecurs q1 q2)
  | POptional p1, POptional p2 -> equal_derecurs p1 p2
  | PRecord (o1,r1), PRecord (o2,r2) -> (o1 == o2) && (LabelMap.equal equal_derecurs r1 r2)
  | PCapture x1, PCapture x2 -> Id.equal x1 x2
  | PConstant (x1,c1), PConstant (x2,c2) -> (Id.equal x1 x2) && (Types.equal_const c1 c2)
  | PRegexp (p1,q1), PRegexp (p2,q2) -> (equal_derecurs_regexp p1 p2) && (equal_derecurs q1 q2)
  | _ -> false
and equal_derecurs_regexp r1 r2 = match r1,r2 with
  | PEpsilon, PEpsilon -> true
  | PElem p1, PElem p2 -> equal_derecurs p1 p2
  | PSeq (p1,q1), PSeq (p2,q2) 
  | PAlt (p1,q1), PAlt (p2,q2) -> (equal_derecurs_regexp p1 p2) && (equal_derecurs_regexp q1 q2)
  | PStar p1, PStar p2
  | PWeakStar p1, PWeakStar p2 -> equal_derecurs_regexp p1 p2
  | _ -> false
110

111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
module DerecursTable = Hashtbl.Make(
  struct 
    type t = derecurs 
    let hash = hash_derecurs
    let equal = equal_derecurs
  end
)

module RE = Hashtbl.Make(
  struct 
    type t = derecurs_regexp * derecurs 
    let hash (p,q) = (hash_derecurs_regexp p) + 17 * (hash_derecurs q)
    let equal (p1,q1) (p2,q2) = (equal_derecurs_regexp p1 p2) && (equal_derecurs q1 q2)
  end
)
126

127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
  
let counter = State.ref "Typer.counter - derecurs" 0
let mk_slot loc = 
  incr counter; 
  { ploop = false; ploc = loc; pid = !counter; pdescr = None }
  
let rec derecurs env p = match p.descr with
  | PatVar v ->
      (try PAlias (TypeEnv.find v env)
       with Not_found -> raise_loc_generic p.loc ("Undefined type/pattern " ^ v))
  | Recurs (p,b) -> derecurs (derecurs_def env b) p
  | Internal t -> PType t
  | Or (p1,p2) -> POr (derecurs env p1, derecurs env p2)
  | And (p1,p2) -> PAnd (derecurs env p1, derecurs env p2)
  | Diff (p1,p2) -> PDiff (derecurs env p1, derecurs env p2)
  | Prod (p1,p2) -> PTimes (derecurs env p1, derecurs env p2)
  | XmlT (p1,p2) -> PXml (derecurs env p1, derecurs env p2)
  | Arrow (p1,p2) -> PArrow (derecurs env p1, derecurs env p2)
  | Optional p -> POptional (derecurs env p)
  | Record (o,r) -> PRecord (o, LabelMap.map (derecurs env) r)
  | Capture x -> PCapture x
  | Constant (x,c) -> PConstant (x,c)
  | Regexp (r,q) -> 
      let constant_nil t v = PAnd (t, PConstant (v, Types.Atom Sequence.nil_atom)) in
      let vars = seq_vars IdSet.empty r in
      let q = IdSet.fold constant_nil (derecurs env q) vars in
      let r = derecurs_regexp (fun p -> p) env r in
      PRegexp (r, q)
and derecurs_regexp vars env = function
  | Epsilon -> PEpsilon
  | Elem p -> PElem (vars (derecurs env p))
  | Seq (p1,p2) -> PSeq (derecurs_regexp vars env p1, derecurs_regexp vars env p2)
  | Alt (p1,p2) -> PAlt (derecurs_regexp vars env p1, derecurs_regexp vars env p2)
  | Star p -> PStar (derecurs_regexp vars env p)
  | WeakStar p -> PStar (derecurs_regexp vars env p)
  | SeqCapture (x,p) -> derecurs_regexp (fun p -> PAnd (vars p, PCapture x)) env p


and derecurs_def env b =
  let b = List.map (fun (v,p) -> (v,p,mk_slot p.loc)) b in
  let env = List.fold_left (fun env (v,p,s) -> TypeEnv.add v s env) env b in
  List.iter (fun (v,p,s) -> s.pdescr <- Some (derecurs env p)) b;
  env
170

171
(* Stratification and recursive hash-consing *)
172
173
174
175
176
177
178
179
180

type descr = 
  | IType of Types.descr
  | IOr of descr * descr
  | IAnd of descr * descr
  | IDiff of descr * descr
  | ITimes of slot * slot
  | IXml of slot * slot
  | IArrow of slot * slot
181
  | IOptional of descr
182
183
184
185
186
187
188
189
  | IRecord of bool * slot label_map
  | ICapture of id
  | IConstant of id * Types.const
and slot = {
  mutable fv : fv option;
  mutable hash : int option;
  mutable rank1: int; mutable rank2: int;
  mutable gen1 : int; mutable gen2: int;
190
  mutable d    : descr option
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
}
    
let descr s = 
  match s.d with
    | Some d -> d
    | None -> assert false
	
let gen = ref 0
let rank = ref 0
	     
let rec hash_descr = function
  | IType x -> Types.hash_descr x
  | IOr (d1,d2) -> 1 + 17 * (hash_descr d1) + 257 * (hash_descr d2)
  | IAnd (d1,d2) -> 2 + 17 * (hash_descr d1) + 257 * (hash_descr d2)
  | IDiff (d1,d2) -> 3 + 17 * (hash_descr d1) + 257 * (hash_descr d2)
  | IOptional d -> 4 + 17 * (hash_descr d)
  | ITimes (s1,s2) -> 5 + 17 * (hash_slot s1) + 257 * (hash_slot s2)
  | IXml (s1,s2) -> 6 + 17 * (hash_slot s1) + 257 * (hash_slot s2)
  | IArrow (s1,s2) -> 7 + 17 * (hash_slot s1) + 257 * (hash_slot s2)
  | IRecord (o,r) -> (if o then 8 else 9) + 17 * (LabelMap.hash hash_slot r)
  | ICapture x -> 10 + 17 * (Id.hash x)
  | IConstant (x,y) -> 11 + 17 * (Id.hash x) + 257 * (Types.hash_const y)
and hash_slot s =
  if s.gen1 = !gen then 13 * s.rank1
  else (
    incr rank;
    s.rank1 <- !rank; s.gen1 <- !gen;
    hash_descr (descr s)
  )
    
let rec equal_descr d1 d2 = 
  match (d1,d2) with
  | IType x1, IType x2 -> Types.equal_descr x1 x2
  | IOr (x1,y1), IOr (x2,y2) 
  | IAnd (x1,y1), IAnd (x2,y2) 
  | IDiff (x1,y1), IDiff (x2,y2) -> (equal_descr x1 x2) && (equal_descr y1 y2)
  | IOptional x1, IOptional x2 -> equal_descr x1 x2
  | ITimes (x1,y1), ITimes (x2,y2) 
  | IXml (x1,y1), IXml (x2,y2) 
  | IArrow (x1,y1), IArrow (x2,y2) -> (equal_slot x1 x2) && (equal_slot y1 y2)
  | IRecord (o1,r1), IRecord (o2,r2) -> (o1 = o2) && (LabelMap.equal equal_slot r1 r2)
  | ICapture x1, ICapture x2 -> Id.equal x1 x2
  | IConstant (x1,y1), IConstant (x2,y2) -> (Id.equal x1 x2) && (Types.equal_const y1 y2)
  | _ -> false
and equal_slot s1 s2 =
  ((s1.gen1 = !gen) && (s2.gen2 = !gen) && (s1.rank1 = s2.rank2))
  ||
  ((s1.gen1 <> !gen) && (s2.gen2 <> !gen) && (
     incr rank;
     s1.rank1 <- !rank; s1.gen1 <- !gen;
     s2.rank2 <- !rank; s2.gen2 <- !gen;
     equal_descr (descr s1) (descr s2)
   ))
  
module Arg = struct
  type t = slot
      
  let hash s =
    match s.hash with
      | Some h -> h
      | None ->
	  incr gen; rank := 0; 
	  let h = hash_slot s in
	  s.hash <- Some h;
	  h
	    
257
258
259
260
261
262
  let equal s1 s2 = 
    (s1 == s2) || 
    (incr gen; rank := 0; 
     let e = equal_slot s1 s2 in
(*     if e then Printf.eprintf "Equal\n"; *)
     e)
263
end
264
265
266
267
268
269
270
271
272
module SlotTable = Hashtbl.Make(Arg)
  
let rec fv_slot s =
  match s.fv with
    | Some x -> x
    | None ->
	if s.gen1 = !gen then IdSet.empty 
	else (s.gen1 <- !gen; fv_descr (descr s))
and fv_descr = function
273
  | IType _ -> IdSet.empty
274
275
276
277
278
279
280
281
282
  | IOr (d1,d2)
  | IAnd (d1,d2)  
  | IDiff (d1,d2) -> IdSet.cup (fv_descr d1) (fv_descr d2)
  | IOptional d -> fv_descr d
  | ITimes (s1,s2)  
  | IXml (s1,s2)  
  | IArrow (s1,s2) -> IdSet.cup (fv_slot s1) (fv_slot s2)
  | IRecord (o,r) -> List.fold_left IdSet.cup IdSet.empty (LabelMap.map_to_list fv_slot r)
  | ICapture x | IConstant (x,_) -> IdSet.singleton x
283

284
285
286
287
288
289
290
291
292
      
let compute_fv s =
  match s.fv with
    | Some x -> ()
    | None ->
	incr gen;
	let x = fv_slot s in
	s.fv <- Some x
	  
293
294

let todo_fv = ref []
295
296
297
298
299
300
301
302
	  
let mk () =   
  let s = 
    { d = None;
      fv = None;
      hash = None;
      rank1 = 0; rank2 = 0;
      gen1 = 0; gen2 = 0 } in
303
  todo_fv := s :: !todo_fv;
304
  s
305
306
307
308

let flush_fv () =
  List.iter compute_fv !todo_fv;
  todo_fv := []
309
    
310
311
312
let compile_slot_hash = DerecursTable.create 67
let compile_hash = DerecursTable.create 67

313
let defs = ref []
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361

let rec compile p =
  try DerecursTable.find compile_hash p
  with Not_found ->
    let c = real_compile p in
    DerecursTable.replace compile_hash p c;
    c
and real_compile = function
  | PAlias v ->
      if v.ploop then
	raise_loc_generic v.ploc ("Unguarded recursion on type/pattern");
      v.ploop <- true;
      let r = match v.pdescr with Some x -> compile x | _ -> assert false in
      v.ploop <- false;
      r
  | PType t -> IType t
  | POr (t1,t2) -> IOr (compile t1, compile t2)
  | PAnd (t1,t2) -> IAnd (compile t1, compile t2)
  | PDiff (t1,t2) -> IDiff (compile t1, compile t2)
  | PTimes (t1,t2) -> ITimes (compile_slot t1, compile_slot t2)
  | PXml (t1,t2) -> IXml (compile_slot t1, compile_slot t2)
  | PArrow (t1,t2) -> IArrow (compile_slot t1, compile_slot t2)
  | POptional t -> IOptional (compile t)
  | PRecord (o,r) ->  IRecord (o, LabelMap.map compile_slot r)
  | PConstant (x,v) -> IConstant (x,v)
  | PCapture x -> ICapture x
  | PRegexp (r,q) -> compile_regexp r q
and compile_regexp r q =
  let memo = RE.create 17 in
  let rec aux accu r q =
    if RE.mem memo (r,q) then accu
    else (
      RE.add memo (r,q) ();
      match r with
	| PEpsilon -> (match q with PRegexp (r,q) -> aux accu r q | _ -> (compile q) :: accu)
	| PElem p -> ITimes (compile_slot p, compile_slot q) :: accu
	| PSeq (r1,r2) -> aux accu r1 (PRegexp (r2,q))
	| PAlt (r1,r2) -> aux (aux accu r1 q) r2 q
	| PStar r1 -> aux (aux accu r1 (PRegexp (r,q))) PEpsilon q
	| PWeakStar r1 -> aux (aux accu PEpsilon q) r1 (PRegexp (r,q))
    )
  in
  let accu = aux [] r q in
  match accu with
    | [] -> assert false
    | p::l -> List.fold_left (fun acc p -> IOr (p,acc)) p l
and compile_slot p =
  try DerecursTable.find compile_slot_hash p
362
363
  with Not_found ->
    let s = mk () in
364
365
    defs := (s,p) :: !defs;
    DerecursTable.add compile_slot_hash p s;
366
    s
367

368
369
370
371
      
let rec flush_defs () = 
  match !defs with
    | [] -> ()
372
    | (s,p)::t -> defs := t; s.d <- Some (compile p); flush_defs ()
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
	
let typ_nodes = SlotTable.create 67
let pat_nodes = SlotTable.create 67
		  
let rec typ = function
  | IType t -> t
  | IOr (s1,s2) -> Types.cup (typ s1) (typ s2)
  | IAnd (s1,s2) ->  Types.cap (typ s1) (typ s2)
  | IDiff (s1,s2) -> Types.diff (typ s1) (typ s2)
  | ITimes (s1,s2) -> Types.times (typ_node s1) (typ_node s2)
  | IXml (s1,s2) -> Types.xml (typ_node s1) (typ_node s2)
  | IArrow (s1,s2) -> Types.arrow (typ_node s1) (typ_node s2)
  | IOptional s -> Types.Record.or_absent (typ s)
  | IRecord (o,r) -> Types.record' (o, LabelMap.map typ_node r)
  | ICapture x | IConstant (x,_) -> assert false
      
and typ_node s : Types.node =
  try SlotTable.find typ_nodes s
  with Not_found ->
    let x = Types.make () in
    SlotTable.add typ_nodes s x;
    Types.define x (typ (descr s));
    x
      
let rec pat d : Patterns.descr =
  if IdSet.is_empty (fv_descr d)
  then Patterns.constr (typ d)
  else pat_aux d
    
    
and pat_aux = function
  | IOr (s1,s2) -> Patterns.cup (pat s1) (pat s2)
  | IAnd (s1,s2) -> Patterns.cap (pat s1) (pat s2)
  | IDiff (s1,s2) when IdSet.is_empty (fv_descr s2) ->
      let s2 = Types.neg (typ s2) in
      Patterns.cap (pat s1) (Patterns.constr s2)
  | IDiff _ ->
      raise (Patterns.Error "Difference not allowed in patterns")
  | ITimes (s1,s2) -> Patterns.times (pat_node s1) (pat_node s2)
  | IXml (s1,s2) -> Patterns.xml (pat_node s1) (pat_node s2)
  | IOptional _ -> 
      raise (Patterns.Error "Optional field not allowed in record patterns")
  | IRecord (o,r) ->
      let pats = ref [] in
      let aux l s = 
	if IdSet.is_empty (fv_slot s) then typ_node s
	else
	  ( pats := Patterns.record l (pat_node s) :: !pats;
	    Types.any_node )
      in
      let constr = Types.record' (o,LabelMap.mapi aux r) in
      List.fold_left Patterns.cap (Patterns.constr constr) !pats
	(* TODO: can avoid constr when o=true, and all fields have fv *)
  | ICapture x -> Patterns.capture x
  | IConstant (x,c) -> Patterns.constant x c
  | IArrow _ ->
      raise (Patterns.Error "Arrow not allowed in patterns")
  | IType _ -> assert false
      
and pat_node s : Patterns.node =
  try SlotTable.find pat_nodes s
  with Not_found ->
    let x = Patterns.make (fv_slot s) in
    SlotTable.add pat_nodes s x;
    Patterns.define x (pat (descr s));
    x
      
let glb = State.ref "Typer.glb_env" TypeEnv.empty
	    
	    
let register_global_types b =
444
445
446
447
448
449
450
  List.iter 
    (fun (v,p) ->
       if TypeEnv.mem v !glb
       then raise_loc_generic p.loc ("Multiple definition for type " ^ v)
    ) b;
  glb := derecurs_def !glb b;
  let b = List.map (fun (v,p) -> (v,p,compile (derecurs !glb p))) b in
451
452
453
454
455
456
  flush_defs ();
  flush_fv ();
  List.iter (fun (v,p,s) -> 
	       if not (IdSet.is_empty (fv_descr s)) then
		 raise_loc_generic p.loc "Capture variables are not allowed in types";
	       Types.Print.register_global v (typ s)) b
457
458
459

let dump_global_types ppf =
  TypeEnv.iter (fun v _ -> Format.fprintf ppf " %s" v) !glb
460
461
462
    
    
let typ p =
463
  let s = compile_slot (derecurs !glb p) in
464
465
466
467
468
469
  flush_defs ();
  flush_fv ();
  if IdSet.is_empty (fv_slot s) then typ_node s
  else raise_loc_generic p.loc "Capture variables are not allowed in types"
    
let pat p = 
470
  let s = compile_slot (derecurs !glb p) in
471
472
473
474
475
  flush_defs ();
  flush_fv ();
  try pat_node s
  with Patterns.Error e -> raise_loc_generic p.loc e
    | Location (loc,exn) when loc = noloc -> raise (Location (p.loc, exn))
476
477


478
479
(* II. Build skeleton *)

480
module Fv = IdSet
481

482
483
484
(* IDEA: introduce a node Loc in the AST to override nolocs
   in sub-expressions *)
   
485
let rec expr loc' { loc = loc; descr = d } = 
486
  let loc =  if loc = noloc then loc' else loc in
487
  let (fv,td) = 
488
    match d with
489
      | Forget (e,t) ->
490
	  let (fv,e) = expr loc e and t = typ t in
491
	  (fv, Typed.Forget (e,t))
492
493
      | Var s -> (Fv.singleton s, Typed.Var s)
      | Apply (e1,e2) -> 
494
	  let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
495
	  (Fv.cup fv1 fv2, Typed.Apply (e1,e2))
496
      | Abstraction a ->
497
	  let iface = List.map (fun (t1,t2) -> (typ t1, typ t2)) 
498
			a.fun_iface in
499
500
501
	  let t = List.fold_left 
		    (fun accu (t1,t2) -> Types.cap accu (Types.arrow t1 t2)) 
		    Types.any iface in
502
503
504
	  let iface = List.map 
			(fun (t1,t2) -> (Types.descr t1, Types.descr t2)) 
			iface in
505
	  let (fv0,body) = branches loc a.fun_body in
506
507
508
509
510
511
512
513
514
	  let fv = match a.fun_name with
	    | None -> fv0
	    | Some f -> Fv.remove f fv0 in
	  (fv,
	   Typed.Abstraction 
	     { Typed.fun_name = a.fun_name;
	       Typed.fun_iface = iface;
	       Typed.fun_body = body;
	       Typed.fun_typ = t;
515
	       Typed.fun_fv = fv
516
517
518
519
	     }
	  )
      | Cst c -> (Fv.empty, Typed.Cst c)
      | Pair (e1,e2) ->
520
	  let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
521
	  (Fv.cup fv1 fv2, Typed.Pair (e1,e2))
522
      | Xml (e1,e2) ->
523
	  let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
524
	  (Fv.cup fv1 fv2, Typed.Xml (e1,e2))
525
      | Dot (e,l) ->
526
	  let (fv,e) = expr loc e in
527
	  (fv,  Typed.Dot (e,l))
528
      | RemoveField (e,l) ->
529
	  let (fv,e) = expr loc e in
530
	  (fv,  Typed.RemoveField (e,l))
531
532
      | RecordLitt r -> 
	  let fv = ref Fv.empty in
533
534
	  let r = LabelMap.map 
		    (fun e -> 
535
		       let (fv2,e) = expr loc e 
536
		       in fv := Fv.cup !fv fv2; e)
537
		    r in
538
	  (!fv, Typed.RecordLitt r)
539
      | Op (op,le) ->
540
	  let (fvs,ltes) = List.split (List.map (expr loc) le) in
541
	  let fv = List.fold_left Fv.cup Fv.empty fvs in
542
	  (fv, Typed.Op (op,ltes))
543
      | Match (e,b) -> 
544
545
	  let (fv1,e) = expr loc e
	  and (fv2,b) = branches loc b in
546
	  (Fv.cup fv1 fv2, Typed.Match (e, b))
547
      | Map (e,b) ->
548
549
	  let (fv1,e) = expr loc e
	  and (fv2,b) = branches loc b in
550
	  (Fv.cup fv1 fv2, Typed.Map (e, b))
551
552
      | Ttree (e,b) ->
	  let b = b @ [ (mknoloc (Internal Types.any)), mknoloc MatchFail ] in
553
554
	  let (fv1,e) = expr loc e
	  and (fv2,b) = branches loc b in
555
556
	  (Fv.cup fv1 fv2, Typed.Ttree (e, b))
      | MatchFail -> (Fv.empty, Typed.MatchFail)
557
      | Try (e,b) ->
558
559
	  let (fv1,e) = expr loc e
	  and (fv2,b) = branches loc b in
560
	  (Fv.cup fv1 fv2, Typed.Try (e, b))
561
  in
562
563
  fv,
  { Typed.exp_loc = loc;
564
565
566
567
    Typed.exp_typ = Types.empty;
    Typed.exp_descr = td;
  }
	      
568
  and branches loc b = 
569
    let fv = ref Fv.empty in
570
    let accept = ref Types.empty in
571
572
    let b = List.map 
	      (fun (p,e) ->
573
574
		 let (fv2,e) = expr loc e in
		 let p = pat p in
575
576
		 let fv2 = Fv.diff fv2 (Patterns.fv p) in
		 fv := Fv.cup !fv fv2;
577
		 accept := Types.cup !accept (Types.descr (Patterns.accept p));
578
		 { Typed.br_used = false;
579
		   Typed.br_pat = p;
580
581
		   Typed.br_body = e }
	      ) b in
582
583
584
585
    (!fv, 
     { 
       Typed.br_typ = Types.empty; 
       Typed.br_branches = b; 
586
587
       Typed.br_accept = !accept;
       Typed.br_compiled = None;
588
589
     } 
    )
590

591
592
let expr = expr noloc

593
594
595
let let_decl p e =
  let (_,e) = expr e in
  { Typed.let_pat = pat p;
596
597
598
599
600
    Typed.let_body = e;
    Typed.let_compiled = None }

(* III. Type-checks *)

601
602
603
let int_cup_record = Types.cup Types.Int.any Types.Record.any


604
type env = Types.descr Env.t
605

606
607
let match_fail = ref Types.empty

608
609
open Typed

610
let warning loc msg =
611
612
613
614
  Format.fprintf !Location.warning_ppf "Warning %a:@\n%a%s@\n" 
    Location.print_loc loc
    Location.html_hilight loc
    msg
615
616
617
618

let check loc t s msg =
  if not (Types.subtype t s) then raise_loc loc (Constraint (t, s, msg))

619
let rec type_check env e constr precise = 
620
(*  Format.fprintf Format.std_formatter "constr=%a precise=%b@\n"
621
622
    Types.Print.print_descr constr precise; 
*)
623
  let d = type_check' e.exp_loc env e.exp_descr constr precise in
624
625
626
  e.exp_typ <- Types.cup e.exp_typ d;
  d

627
and type_check' loc env e constr precise = match e with
628
629
630
631
  | Forget (e,t) ->
      let t = Types.descr t in
      ignore (type_check env e t false);
      t
632
  | Abstraction a ->
633
634
635
636
637
638
639
      let t =
	try Types.Arrow.check_strenghten a.fun_typ constr 
	with Not_found -> 
	  raise_loc loc 
	  (ShouldHave
	     (constr, "but the interface of the abstraction is not compatible"))
      in
640
641
642
      let env = match a.fun_name with
	| None -> env
	| Some f -> Env.add f a.fun_typ env in
643
644
      List.iter 
	(fun (t1,t2) ->
645
	   ignore (type_check_branches loc env t1 a.fun_body t2 false)
646
647
	) a.fun_iface;
      t
648

649
650
  | Match (e,b) ->
      let t = type_check env e b.br_accept true in
651
      type_check_branches loc env t b constr precise
652
653
654

  | Try (e,b) ->
      let te = type_check env e constr precise in
655
      let tb = type_check_branches loc env Types.any b constr precise in
656
      Types.cup te tb
657

658
659
660
661
  | Pair (e1,e2) ->
      type_check_pair loc env e1 e2 constr precise
  | Xml (e1,e2) ->
      type_check_pair ~kind:`XML loc env e1 e2 constr precise
662

663
  | RecordLitt r ->
664
(* try to get rid of precise = true for values of fields *)
665
      if not (Types.Record.has_record constr) then
666
667
	raise_loc loc (ShouldHave (constr,"but it is a record."));
      let (rconstr,res) = 
668
	List.fold_left
669
	  (fun (rconstr,res) (l,e) ->
670
671
672
	     (* could compute (split l e) once... *)
	     let pi = Types.Record.project_opt rconstr l in
	     if Types.is_empty pi then 
673
674
675
	       raise_loc loc 
		 (ShouldHave (constr,(Printf.sprintf 
					"Field %s is not allowed here."
676
					(LabelPool.value l)
677
678
679
				     )
			     ));
	     let t = type_check env e pi true in
680
681
	     let rconstr = Types.Record.condition rconstr l t in
	     let res = if precise then (l,Types.cons t) :: res else res in
682
	     (rconstr,res)
683
	  ) (constr, []) (LabelMap.get r)
684
      in
685
686
687
688
689
690
      if not (Types.Record.has_empty_record rconstr) then
	raise_loc loc 
	  (ShouldHave (constr,"More field should be present"));
      if precise then
	Types.record' (false, LabelMap.from_list (fun _ _ -> assert false) res)
      else constr
691
692
693
694
695
  | Map (e,b) ->
      let t = type_check env e (Sequence.star b.br_accept) true in

      let constr' = Sequence.approx (Types.cap Sequence.any constr) in
      let exact = Types.subtype (Sequence.star constr') constr in
696
697
698
699
700
701
702
      (* Note: 
	 - could be more precise by integrating the decomposition
	 of constr inside Sequence.map.
      *)
      let res = 
	Sequence.map 
	  (fun t -> 
703
	     type_check_branches loc env t b constr' (precise || (not exact)))
704
705
706
	  t in
      if not exact then check loc res constr "";
      if precise then res else constr
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
  | Op ("@", [e1;e2]) ->
      let constr' = Sequence.star 
		      (Sequence.approx (Types.cap Sequence.any constr)) in
      let exact = Types.subtype constr' constr in
      if exact then
	let t1 = type_check env e1 constr' precise
	and t2 = type_check env e2 constr' precise in
	if precise then Sequence.concat t1 t2 else constr
      else
	(* Note:
	   the knownledge of t1 may makes it useless to
	   check t2 with 'precise' ... *)
	let t1 = type_check env e1 constr' true
	and t2 = type_check env e2 constr' true in
	let res = Sequence.concat t1 t2 in
	check loc res constr "";
	if precise then res else constr
724
  | Apply (e1,e2) ->
725
(*
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
      let constr' = Sequence.star 
		      (Sequence.approx (Types.cap Sequence.any constr)) in
      let t1 = type_check env e1 (Types.cup Types.Arrow.any constr') true in
      let t1_fun = Types.Arrow.get t1 in

      let has_fun = not (Types.Arrow.is_empty t1_fun)
      and has_seq = not (Types.subtype t1 Types.Arrow.any) in

      let constr' =
	Types.cap 
	  (if has_fun then Types.Arrow.domain t1_fun else Types.any)
	  (if has_seq then constr' else Types.any)
      in
      let need_arg = has_fun && Types.Arrow.need_arg t1_fun in
      let precise  = need_arg || has_seq in
      let t2 = type_check env e2 constr' precise in
      let res = Types.cup 
		  (if has_fun then 
		     if need_arg then Types.Arrow.apply t1_fun t2
		     else Types.Arrow.apply_noarg t1_fun
		   else Types.empty)
		  (if has_seq then Sequence.concat t1 t2
		   else Types.empty)
      in
      check loc res constr "";
      res
752
*)
753
754
755
      let t1 = type_check env e1 Types.Arrow.any true in
      let t1 = Types.Arrow.get t1 in
      let dom = Types.Arrow.domain t1 in
756
757
758
759
760
761
762
763
764
      let res =
	if Types.Arrow.need_arg t1 then
	  let t2 = type_check env e2 dom true in
	  Types.Arrow.apply t1 t2
	else
	  (ignore (type_check env e2 dom false); Types.Arrow.apply_noarg t1)
      in
      check loc res constr "";
      res
765
766
767
768
769
770
771
772
773
774
775
776
777
  | Op ("flatten", [e]) ->
      let constr' = Sequence.star 
		      (Sequence.approx (Types.cap Sequence.any constr)) in
      let sconstr' = Sequence.star constr' in
      let exact = Types.subtype constr' constr in
      if exact then
	let t = type_check env e sconstr' precise in
	if precise then Sequence.flatten t else constr
      else
	let t = type_check env e sconstr' true in
	let res = Sequence.flatten t in
	check loc res constr "";
	if precise then res else constr
778
779
780
781
782
  | _ -> 
      let t : Types.descr = compute_type' loc env e in
      check loc t constr "";
      t

783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
and type_check_pair ?(kind=`Normal) loc env e1 e2 constr precise =
  let rects = Types.Product.get ~kind constr in
  if Types.Product.is_empty rects then 
    (match kind with
      | `Normal -> raise_loc loc (ShouldHave (constr,"but it is a pair."))
      | `XML -> raise_loc loc (ShouldHave (constr,"but it is an XML element.")));
  let pi1 = Types.Product.pi1 rects in
  
  let t1 = type_check env e1 (Types.Product.pi1 rects) 
	     (precise || (Types.Product.need_second rects))in
  let rects = Types.Product.restrict_1 rects t1 in
  let t2 = type_check env e2 (Types.Product.pi2 rects) precise in
  if precise then 
    match kind with
      | `Normal -> Types.times (Types.cons t1) (Types.cons t2)
      | `XML -> Types.xml (Types.cons t1) (Types.cons t2)
  else
    constr


803
804
805
806
and compute_type env e =
  type_check env e Types.any true

and compute_type' loc env = function
807
808
  | Var s -> 
      (try Env.find s env 
809
       with Not_found -> raise_loc loc (UnboundId (Id.value s))
810
      )
811
  | Cst c -> Types.constant c
812
813
814
815
  | Dot (e,l) ->
      let t = type_check env e Types.Record.any true in
         (try (Types.Record.project t l) 
          with Not_found -> raise_loc loc (WrongLabel(t,l)))
816
817
818
  | RemoveField (e,l) ->
      let t = type_check env e Types.Record.any true in
      Types.Record.remove_field t l
819
820
821
  | Op (op, el) ->
      let args = List.map (fun e -> (e.exp_loc, compute_type env e)) el in
      type_op loc op args
822
823
824
825
826
827
828
829
830
831
832
833
  | Ttree (e,b) ->
      let t = type_check env e Sequence.any true in
      let r = 
	Sequence.map_tree 
	  (fun t -> 
	     let res = type_check_branches loc env t b Sequence.any true in
	     let resid = !match_fail in
	     match_fail := Types.empty;
	     (res,resid)
	  ) t
      in
      r
834
835
836
837
838

(* We keep these cases here to allow comparison and benchmarking ...
   Just comment the corresponding cases in type_check' to
   activate these ones.
*)
839
840
841
  | Map (e,b) ->
      let t = compute_type env e in
      Sequence.map (fun t -> type_check_branches loc env t b Types.any true) t
842
843
844
845
846
  | Pair (e1,e2) -> 
      let t1 = compute_type env e1 
      and t2 = compute_type env e2 in
      Types.times (Types.cons t1) (Types.cons t2)
  | RecordLitt r ->
847
      let r = LabelMap.map (fun e -> Types.cons (compute_type env e)) r in
848
      Types.record' (false,r)
849
  | _ -> assert false
850

851
and type_check_branches loc env targ brs constr precise =
852
  if Types.is_empty targ then Types.empty 
853
854
  else (
    brs.br_typ <- Types.cup brs.br_typ targ;
855
    branches_aux loc env targ 
856
857
      (if precise then Types.empty else constr) 
      constr precise brs.br_branches
858
  )
859
    
860
861
and branches_aux loc env targ tres constr precise = function
  | [] -> raise_loc loc (NonExhaustive targ)
862
863
864
  | { br_body = { exp_descr = MatchFail } } :: _ ->
      match_fail := Types.cup !match_fail targ;
      tres
865
866
867
868
869
870
  | b :: rem ->
      let p = b.br_pat in
      let acc = Types.descr (Patterns.accept p) in

      let targ' = Types.cap targ acc in
      if Types.is_empty targ' 
871
      then branches_aux loc env targ tres constr precise rem
872
873
874
875
876
877
      else 
	( b.br_used <- true;
	  let res = Patterns.filter targ' p in
	  let env' = List.fold_left 
		       (fun env (x,t) -> Env.add x (Types.descr t) env) 
		       env res in
878
879
	  let t = type_check env' b.br_body constr precise in
	  let tres = if precise then Types.cup t tres else tres in
880
881
	  let targ'' = Types.diff targ acc in
	  if (Types.non_empty targ'') then 
882
	    branches_aux loc env targ'' tres constr precise rem 
883
884
	  else
	    tres
885
	)
886

887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
and type_let_decl env l =
  let acc = Types.descr (Patterns.accept l.let_pat) in
  let t = type_check env l.let_body acc true in
  let res = Patterns.filter t l.let_pat in
  List.map (fun (x,t) -> (x, Types.descr t)) res

and type_rec_funs env l =
  let types = 
    List.fold_left
      (fun accu -> function  {let_body={exp_descr=Abstraction a}} as l ->
	 let t = a.fun_typ in
	 let acc = Types.descr (Patterns.accept l.let_pat) in
	 if not (Types.subtype t acc) then
	   raise_loc l.let_body.exp_loc (NonExhaustive (Types.diff t acc));
	 let res = Patterns.filter t l.let_pat in
	 List.fold_left (fun accu (x,t) -> (x, Types.descr t)::accu) accu res
	 | _ -> assert false) [] l
  in
  let env' = List.fold_left (fun env (x,t) -> Env.add x t env) env types in
  List.iter 
    (function  { let_body = { exp_descr = Abstraction a } } as l ->
       ignore (type_check env' l.let_body Types.any false)
       | _ -> assert false) l;
  types


913
914
and type_op loc op args =
  match (op,args) with
915
    | "+", [loc1,t1; loc2,t2] ->
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
	check loc1 t1 int_cup_record
	"The first argument of + must be an integer or a record";
	let int = Types.Int.get t1 in
	let int = if Intervals.is_empty int then None else Some int in
	let r = if Types.Record.has_record t1 then Some t1 else None in
	(match (int,r) with
	   | Some t1, None ->
	       if not (Types.Int.is_int t2) then
		 raise_loc loc2
		   (Constraint 
		      (t2,Types.Int.any,
		       "The second argument of + must be an integer"));
	       Types.Int.put
		 (Intervals.add t1 (Types.Int.get t2));
	   | None, Some r1 ->
	       check loc2 t2 Types.Record.any 
	       "The second argument of + must be a record";
	       Types.Record.merge r1 t2
	   | None, None ->
	       Types.empty
	   | Some t1, Some r1 ->
	       check loc2 t2 int_cup_record
	       "The second argument of + must be an integer or a record";
	       Types.cup 
		 (Types.Int.put (Intervals.add t1 (Types.Int.get t2)))
		 (Types.Record.merge r1 t2)
	)
943
944
    | "-", [loc1,t1; loc2,t2] ->
	type_int_binop Intervals.sub loc1 t1 loc2 t2
945
    | ("*" | "/" | "mod"), [loc1,t1; loc2,t2] ->
946
	type_int_binop (fun i1 i2 -> Intervals.any) loc1 t1 loc2 t2
947
    | "@", [loc1,t1; loc2,t2] ->
948
949
950
	check loc1 t1 Sequence.any
	  "The first argument of @ must be a sequence";
	Sequence.concat t1 t2
951
    | "flatten", [loc1,t1] ->
952
953
954
	check loc1 t1 Sequence.seqseq 
	  "The argument of flatten must be a sequence of sequences";
	Sequence.flatten t1
955
956
957
958
    | "load_xml", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_xml must be a string (filename)";
	Types.any
959
960
961
962
    | "load_file", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_file must be a string (filename)";
	Sequence.string
963
964
965
966
    | "load_html", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_html must be a string (filename)";
	Types.any
967
968
    | "raise", [loc1,t1] ->
	Types.empty
969
970
    | "print_xml", [loc1,t1] ->
	Sequence.string
971
972
    | "print", [loc1,t1] ->
	check loc1 t1 Sequence.string
973
974
975
976
977
978
979
980
	  "The argument of print must be a string";
	Sequence.nil_type
    | "dump_to_file", [loc1,t1; loc2,t2] ->
	check loc1 t1 Sequence.string
	  "The argument of dump_to_file must be a string (filename)";
	check loc2 t2 Sequence.string
	  "The argument of dump_to_file must be a string (value to dump)";
	Sequence.nil_type
981
982
    | "int_of", [loc1,t1] ->
	check loc1 t1 Sequence.string
983
	  "The argument of int_of must be a string";
984
985
986
	if not (Types.subtype t1 Builtin.intstr) then
	  warning loc "This application of int_of may fail";
	Types.interval Intervals.any
987
988
    | "string_of", [loc1,t1] ->
	Sequence.string
989
    | "=", [loc1,t1; loc2,t2] ->
990
991
992
993
994
995
996
	(* could prevent comparision of functional value here... *)
	(* could also handle the case when t1 and t2 are the same 
	   singleton type *)
	if Types.is_empty (Types.cap t1 t2) then
	  Builtin.false_type
	else 
	  Builtin.bool
997
998
999
    | ("<=" | "<" | ">" | ">=" ), [loc1,t1; loc2,t2] ->
	(* could prevent comparision of functional value here... *)
	Builtin.bool
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
    | _ -> assert false

and type_int_binop f loc1 t1 loc2 t2 =
  if not (Types.Int.is_int t1) then
    raise_loc loc1 
      (Constraint 
	 (t1,Types.Int.any,
	  "The first argument must be an integer"));
  if not (Types.Int.is_int t2) then
    raise_loc loc2
      (Constraint 
1011
	       (t2,Types.Int.any,
1012
1013
1014
1015
1016
		"The second argument must be an integer"));
  Types.Int.put
    (f (Types.Int.get t1) (Types.Int.get t2));