typer.ml 33.4 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
10
11
12
13
let warning loc msg =
  Format.fprintf !Location.warning_ppf "Warning %a:@\n%a%s@\n" 
    Location.print_loc loc
    Location.html_hilight loc
    msg

14
15
(* I. Transform the abstract syntax of types and patterns into
      the internal form *)
16
17
18

open Location
open Ast
19
open Ident
20

21
module S = struct type t = string let compare = compare end
22
module StringSet = Set.Make(S)
23
module TypeEnv = Map.Make(S)
24
module Env = Map.Make(Id)
25

26

27
28
exception NonExhaustive of Types.descr
exception Constraint of Types.descr * Types.descr * string
29
exception ShouldHave of Types.descr * string
30
exception ShouldHave2 of Types.descr * string * Types.descr
31
exception WrongLabel of Types.descr * label
32
exception UnboundId of string
33
34

let raise_loc loc exn = raise (Location (loc,exn))
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
110
111
112
113
114
115
116
(* 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
117

118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
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
)
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
  
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)
168
  | WeakStar p -> PWeakStar (derecurs_regexp vars env p)
169
170
171
172
173
174
175
176
  | 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
177

178
(* Stratification and recursive hash-consing *)
179
180
181
182
183
184
185
186
187

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
188
  | IOptional of descr
189
190
191
192
193
194
195
196
  | 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;
197
  mutable d    : descr option
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
257
258
259
260
261
262
263
}
    
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
	    
264
265
266
267
  let equal s1 s2 = 
    (s1 == s2) || 
    (incr gen; rank := 0; 
     let e = equal_slot s1 s2 in
268
(*     if e then Printf.eprintf "Recursive hash-consig: Equal\n";  *)
269
     e)
270
end
271
272
273
274
275
276
277
278
279
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
280
  | IType _ -> IdSet.empty
281
282
283
284
285
286
287
288
289
  | 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
290

291
292
293
294
295
296
297
298
299
      
let compute_fv s =
  match s.fv with
    | Some x -> ()
    | None ->
	incr gen;
	let x = fv_slot s in
	s.fv <- Some x
	  
300
301

let todo_fv = ref []
302
303
304
305
306
307
308
309
	  
let mk () =   
  let s = 
    { d = None;
      fv = None;
      hash = None;
      rank1 = 0; rank2 = 0;
      gen1 = 0; gen2 = 0 } in
310
  todo_fv := s :: !todo_fv;
311
  s
312
313
314
315

let flush_fv () =
  List.iter compute_fv !todo_fv;
  todo_fv := []
316
    
317
318
319
let compile_slot_hash = DerecursTable.create 67
let compile_hash = DerecursTable.create 67

320
let defs = ref []
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
362
363
364
365
366
367
368

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
369
370
  with Not_found ->
    let s = mk () in
371
372
    defs := (s,p) :: !defs;
    DerecursTable.add compile_slot_hash p s;
373
    s
374

375
376
377
378
      
let rec flush_defs () = 
  match !defs with
    | [] -> ()
379
    | (s,p)::t -> defs := t; s.d <- Some (compile p); flush_defs ()
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
444
445
446
447
448
449
450
	
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 =
451
452
453
454
455
456
457
  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
458
459
  flush_defs ();
  flush_fv ();
460
461
462
463
464
465
  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";
       let t = typ s in
       if (p.loc <> noloc) && (Types.is_empty t) then
466
	 warning p.loc ("This definition yields an empty type for " ^ v);
467
       Types.Print.register_global v t) b
468
469
470

let dump_global_types ppf =
  TypeEnv.iter (fun v _ -> Format.fprintf ppf " %s" v) !glb
471
472
473
    
    
let typ p =
474
  let s = compile_slot (derecurs !glb p) in
475
476
477
478
479
480
  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 = 
481
  let s = compile_slot (derecurs !glb p) in
482
483
484
485
486
  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))
487
488


489
490
(* II. Build skeleton *)

491
module Fv = IdSet
492

493
494
let all_branches = ref []

495
496
497
(* IDEA: introduce a node Loc in the AST to override nolocs
   in sub-expressions *)
   
498
let exp loc fv e =
499
500
  fv,
  { Typed.exp_loc = loc;
501
    Typed.exp_typ = Types.empty;
502
    Typed.exp_descr = e;
503
  }
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570


let rec expr loc = function
  | LocatedExpr (loc,e) -> expr loc e
  | Forget (e,t) ->
      let (fv,e) = expr loc e and t = typ t in
      exp loc fv (Typed.Forget (e,t))
  | Var s -> 
      exp loc (Fv.singleton s) (Typed.Var s)
  | Apply (e1,e2) -> 
      let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
      exp loc (Fv.cup fv1 fv2) (Typed.Apply (e1,e2))
  | Abstraction a ->
      let iface = List.map (fun (t1,t2) -> (typ t1, typ t2)) 
		    a.fun_iface in
      let t = List.fold_left 
		(fun accu (t1,t2) -> Types.cap accu (Types.arrow t1 t2)) 
		Types.any iface in
      let iface = List.map 
		    (fun (t1,t2) -> (Types.descr t1, Types.descr t2)) 
		    iface in
      let (fv0,body) = branches a.fun_body in
      let fv = match a.fun_name with
	| None -> fv0
	| Some f -> Fv.remove f fv0 in
      let e = Typed.Abstraction 
		{ Typed.fun_name = a.fun_name;
		  Typed.fun_iface = iface;
		  Typed.fun_body = body;
		  Typed.fun_typ = t;
		  Typed.fun_fv = fv
		} in
      exp loc fv e
  | Cst c -> 
      exp loc Fv.empty (Typed.Cst c)
  | Pair (e1,e2) ->
      let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
      exp loc (Fv.cup fv1 fv2) (Typed.Pair (e1,e2))
  | Xml (e1,e2) ->
      let (fv1,e1) = expr loc e1 and (fv2,e2) = expr loc e2 in
      exp loc (Fv.cup fv1 fv2) (Typed.Xml (e1,e2))
  | Dot (e,l) ->
      let (fv,e) = expr loc e in
      exp loc fv (Typed.Dot (e,l))
  | RemoveField (e,l) ->
      let (fv,e) = expr loc e in
      exp loc fv (Typed.RemoveField (e,l))
  | RecordLitt r -> 
      let fv = ref Fv.empty in
      let r = LabelMap.map 
		(fun e -> 
		   let (fv2,e) = expr loc e 
		   in fv := Fv.cup !fv fv2; e)
		r in
      exp loc !fv (Typed.RecordLitt r)
  | Op (op,le) ->
      let (fvs,ltes) = List.split (List.map (expr loc) le) in
      let fv = List.fold_left Fv.cup Fv.empty fvs in
      exp loc fv (Typed.Op (op,ltes))
  | Match (e,b) -> 
      let (fv1,e) = expr loc e
      and (fv2,b) = branches b in
      exp loc (Fv.cup fv1 fv2) (Typed.Match (e, b))
  | Map (e,b) ->
      let (fv1,e) = expr loc e
      and (fv2,b) = branches b in
      exp loc (Fv.cup fv1 fv2) (Typed.Map (e, b))
571
  | Xtrans (e,b) ->
572
573
574
      let b = b @ [ mknoloc (Internal Types.any), MatchFail ] in
      let (fv1,e) = expr loc e
      and (fv2,b) = branches b in
575
      exp loc (Fv.cup fv1 fv2) (Typed.Xtrans (e, b))
576
577
578
579
580
581
582
  | MatchFail -> 
      exp loc (Fv.empty) Typed.MatchFail
  | Try (e,b) ->
      let (fv1,e) = expr loc e
      and (fv2,b) = branches b in
      exp loc (Fv.cup fv1 fv2) (Typed.Try (e, b))

583
	      
584
  and branches b = 
585
    let fv = ref Fv.empty in
586
    let accept = ref Types.empty in
587
    let branch (p,e) = 
588
589
      let (fv2,e) = expr noloc e in
      let br_loc = merge_loc p.loc e.Typed.exp_loc in
590
591
592
593
594
595
596
597
598
599
600
601
602
      let p = pat p in
      let fv2 = Fv.diff fv2 (Patterns.fv p) in
      fv := Fv.cup !fv fv2;
      accept := Types.cup !accept (Types.descr (Patterns.accept p));
      let br = 
	{ 
	  Typed.br_loc = br_loc;
	  Typed.br_used = br_loc = noloc;
	  Typed.br_pat = p;
	  Typed.br_body = e } in
      all_branches := br :: !all_branches;
      br in
    let b = List.map branch b in
603
604
605
606
    (!fv, 
     { 
       Typed.br_typ = Types.empty; 
       Typed.br_branches = b; 
607
608
       Typed.br_accept = !accept;
       Typed.br_compiled = None;
609
610
     } 
    )
611

612
613
let expr = expr noloc

614
615
616
let let_decl p e =
  let (_,e) = expr e in
  { Typed.let_pat = pat p;
617
618
619
620
621
    Typed.let_body = e;
    Typed.let_compiled = None }

(* III. Type-checks *)

622
623
624
let int_cup_record = Types.cup Types.Int.any Types.Record.any


625
type env = Types.descr Env.t
626

627
628
let match_fail = ref Types.empty

629
630
open Typed

631
632
633
634

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

635
let rec type_check env e constr precise = 
636
(*  Format.fprintf Format.std_formatter "constr=%a precise=%b@\n"
637
638
    Types.Print.print_descr constr precise; 
*)
639
  let d = type_check' e.exp_loc env e.exp_descr constr precise in
640
641
642
  e.exp_typ <- Types.cup e.exp_typ d;
  d

643
and type_check' loc env e constr precise = match e with
644
645
646
647
  | Forget (e,t) ->
      let t = Types.descr t in
      ignore (type_check env e t false);
      t
648
  | Abstraction a ->
649
650
651
652
653
654
655
      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
656
657
658
      let env = match a.fun_name with
	| None -> env
	| Some f -> Env.add f a.fun_typ env in
659
660
      List.iter 
	(fun (t1,t2) ->
661
	   ignore (type_check_branches loc env t1 a.fun_body t2 false)
662
663
	) a.fun_iface;
      t
664

665
666
  | Match (e,b) ->
      let t = type_check env e b.br_accept true in
667
      type_check_branches loc env t b constr precise
668
669
670

  | Try (e,b) ->
      let te = type_check env e constr precise in
671
      let tb = type_check_branches loc env Types.any b constr precise in
672
      Types.cup te tb
673

674
675
676
677
  | 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
678

679
  | RecordLitt r ->
680
(* try to get rid of precise = true for values of fields *)
681
682
(* also: the use equivalent of need_second to optimize... *)
      let precise = true in
683
      if not (Types.Record.has_record constr) then
684
685
	raise_loc loc (ShouldHave (constr,"but it is a record."));
      let (rconstr,res) = 
686
	List.fold_left
687
	  (fun (rconstr,res) (l,e) ->
688
689
690
	     (* could compute (split l e) once... *)
	     let pi = Types.Record.project_opt rconstr l in
	     if Types.is_empty pi then 
691
692
693
	       raise_loc loc 
		 (ShouldHave (constr,(Printf.sprintf 
					"Field %s is not allowed here."
694
					(LabelPool.value l)
695
696
697
				     )
			     ));
	     let t = type_check env e pi true in
698
699
	     let rconstr = Types.Record.condition rconstr l t in
	     let res = if precise then (l,Types.cons t) :: res else res in
700
	     (rconstr,res)
701
	  ) (constr, []) (LabelMap.get r)
702
      in
703
704
705
      if not (Types.Record.has_empty_record rconstr) then
	raise_loc loc 
	  (ShouldHave (constr,"More field should be present"));
706
707
708
709
710
711
      let t = 
	Types.record' (false, LabelMap.from_list (fun _ _ -> assert false) res)
      in
      check loc t constr "";
      t
(*
712
713
714
      if precise then
	Types.record' (false, LabelMap.from_list (fun _ _ -> assert false) res)
      else constr
715
*)
716
717
718
719
720
  | 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
721
722
723
724
725
726
727
      (* Note: 
	 - could be more precise by integrating the decomposition
	 of constr inside Sequence.map.
      *)
      let res = 
	Sequence.map 
	  (fun t -> 
728
	     type_check_branches loc env t b constr' (precise || (not exact)))
729
730
731
	  t in
      if not exact then check loc res constr "";
      if precise then res else constr
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
  | 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
749
  | Apply (e1,e2) ->
750
(*
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
      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
777
*)
778
779
780
      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
781
782
783
784
785
786
787
788
789
      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
790
791
792
793
794
795
796
797
798
799
800
801
802
  | 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
803
804
805
  | Op ("atom_of", [e]) ->
      let t = type_check env e Sequence.string false in
      Types.atom Atoms.any
806
807
808
809
810
  | _ -> 
      let t : Types.descr = compute_type' loc env e in
      check loc t constr "";
      t

811
and type_check_pair ?(kind=`Normal) loc env e1 e2 constr precise =
812
  let rects = Types.Product.normal ~kind constr in
813
814
815
816
  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.")));
817
  let need_s = Types.Product.need_second rects in
818
819
820
821
822
  let t1 = type_check env e1 (Types.Product.pi1 rects) (precise || need_s) in
  let c2 = Types.Product.constraint_on_2 rects t1 in
  if Types.is_empty c2 then 
    raise_loc loc (ShouldHave2 (constr,"but the first component has type",t1));
  let t2 = type_check env e2 c2 precise in
823

824
  if precise then 
825
826
827
    match kind with
      | `Normal -> Types.times (Types.cons t1) (Types.cons t2)
      | `XML -> Types.xml (Types.cons t1) (Types.cons t2)
828
829
830
831
  else
    constr


832
833
834
835
and compute_type env e =
  type_check env e Types.any true

and compute_type' loc env = function
836
837
  | Var s -> 
      (try Env.find s env 
838
       with Not_found -> raise_loc loc (UnboundId (Id.value s))
839
      )
840
  | Cst c -> Types.constant c
841
842
843
844
  | 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)))
845
846
847
  | RemoveField (e,l) ->
      let t = type_check env e Types.Record.any true in
      Types.Record.remove_field t l
848
849
850
  | Op (op, el) ->
      let args = List.map (fun e -> (e.exp_loc, compute_type env e)) el in
      type_op loc op args
851
  | Xtrans (e,b) ->
852
853
854
855
856
857
858
859
860
861
862
      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
863
864
865
866
867

(* We keep these cases here to allow comparison and benchmarking ...
   Just comment the corresponding cases in type_check' to
   activate these ones.
*)
868
869
870
  | 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
871
872
873
874
875
  | 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 ->
876
      let r = LabelMap.map (fun e -> Types.cons (compute_type env e)) r in
877
      Types.record' (false,r)
878
  | _ -> assert false
879

880
and type_check_branches loc env targ brs constr precise =
881
  if Types.is_empty targ then Types.empty 
882
883
  else (
    brs.br_typ <- Types.cup brs.br_typ targ;
884
    branches_aux loc env targ 
885
886
      (if precise then Types.empty else constr) 
      constr precise brs.br_branches
887
  )
888
    
889
890
and branches_aux loc env targ tres constr precise = function
  | [] -> raise_loc loc (NonExhaustive targ)
891
892
893
  | { br_body = { exp_descr = MatchFail } } :: _ ->
      match_fail := Types.cup !match_fail targ;
      tres
894
895
896
897
898
899
  | 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' 
900
      then branches_aux loc env targ tres constr precise rem
901
902
903
904
905
906
      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
907
908
	  let t = type_check env' b.br_body constr precise in
	  let tres = if precise then Types.cup t tres else tres in
909
910
	  let targ'' = Types.diff targ acc in
	  if (Types.non_empty targ'') then 
911
	    branches_aux loc env targ'' tres constr precise rem 
912
913
	  else
	    tres
914
	)
915

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
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


942
943
and type_op loc op args =
  match (op,args) with
944
    | "+", [loc1,t1; loc2,t2] ->
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
	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)
	)
972
973
    | "-", [loc1,t1; loc2,t2] ->
	type_int_binop Intervals.sub loc1 t1 loc2 t2
974
    | ("*" | "/" | "mod"), [loc1,t1; loc2,t2] ->
975
	type_int_binop (fun i1 i2 -> Intervals.any) loc1 t1 loc2 t2
976
    | "@", [loc1,t1; loc2,t2] ->
977
978
979
	check loc1 t1 Sequence.any
	  "The first argument of @ must be a sequence";
	Sequence.concat t1 t2
980
    | "flatten", [loc1,t1] ->
981
982
983
	check loc1 t1 Sequence.seqseq 
	  "The argument of flatten must be a sequence of sequences";
	Sequence.flatten t1
984
985
986
987
    | "load_xml", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_xml must be a string (filename)";
	Types.any
988
989
990
991
    | "load_file", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_file must be a string (filename)";
	Sequence.string
992
993
994
995
    | "load_html", [loc1,t1] ->
	check loc1 t1 Sequence.string
	  "The argument of load_html must be a string (filename)";
	Types.any
996
997
    | "raise", [loc1,t1] ->
	Types.empty
998
999
    | "print_xml", [loc1,t1] ->
	Sequence.string
1000
1001
    | "print", [loc1,t1] ->
	check loc1 t1 Sequence.string
1002
1003
1004
1005
1006
1007
1008
1009
	  "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
1010
1011
    | "int_of", [loc1,t1] ->
	check loc1 t1 Sequence.string
1012
	  "The argument of int_of must be a string";
1013
1014
1015
	if not (Types.subtype t1 Builtin.intstr) then
	  warning loc "This application of int_of may fail";
	Types.interval Intervals.any
1016
1017
    | "string_of", [loc1,t1] ->
	Sequence.string
1018
    | "=", [loc1,t1; loc2,t2] ->
1019
1020
1021
1022
1023
1024
1025
	(* 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
1026
1027
1028
    | ("<=" | "<" | ">" | ">=" ), [loc1,t1; loc2,t2] ->
	(* could prevent comparision of functional value here... *)
	Builtin.bool
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
    | _ -> 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 
1040
	       (t2,Types.Int.any,
1041
1042
		"The second argument must be an integer"));
  Types.Int.put
1043
    (f (Types.Int.get t1) (Types.Int.get t2))
1044
1045
  

1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056


let report_unused_branches () =
  List.iter
    (fun b ->
       if not b.br_used then
	 warning b.br_loc "This branch is not used"
    )
    !all_branches;
  all_branches := []