typer.ml 41 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
open Location
open Ast
open Ident
10

11
12
let warning loc msg =
  Format.fprintf !Location.warning_ppf "Warning %a:@\n%a%s@\n" 
13
14
    Location.print_loc (loc,`Full)
    Location.html_hilight (loc,`Full)
15
16
    msg

17
18
19
20
21
22
type item =
  | Type of Types.t
  | Value of Types.t

type env = {
  ids : item Env.t;
23
24
  tenv_nspref: Ns.table;
}
25

26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
let empty_env = {
  ids = Env.empty;
  tenv_nspref = Ns.empty_table;
}

let enter_type id t env =
  { env with ids = Env.add id (Type t) env.ids }
let enter_types l env =
  { env with ids = 
      List.fold_left (fun accu (id,t) -> Env.add id (Type t) accu) env.ids l }
let find_type id env =
  match Env.find id env.ids with
    | Type t -> t
    | Value _ -> raise Not_found

let enter_value id t env = 
  { env with ids = Env.add id (Value t) env.ids }
let enter_values l env =
  { env with ids = 
      List.fold_left (fun accu (id,t) -> Env.add id (Value t) accu) env.ids l }
let find_value id env =
  match Env.find id env.ids with
    | Value t -> t
    | _ -> raise Not_found
	

52
(* Namespaces *)
53

54
55
56
let set_ns_table_for_printer env = 
  Ns.InternalPrinter.set_table env.tenv_nspref

57
let get_ns_table tenv = tenv.tenv_nspref
58

59
60
61
let enter_ns p ns env =
  { env with tenv_nspref = Ns.add_prefix p ns env.tenv_nspref }

62
63
64
65
66
let protect_error_ns loc f x =
  try f x
  with Ns.UnknownPrefix ns ->
    raise_loc_generic loc 
    ("Undefined namespace prefix " ^ (U.to_string ns))
67

68
69
70
71
72
73
let parse_atom env loc t =
  let (ns,l) = protect_error_ns loc (Ns.map_tag env.tenv_nspref) t in
  Atoms.V.mk ns l
 
let parse_ns env loc ns =
  protect_error_ns loc (Ns.map_prefix env.tenv_nspref) ns
74

75
76
77
let parse_label env loc t =
  let (ns,l) = protect_error_ns loc (Ns.map_attr env.tenv_nspref) t in
  LabelPool.mk (ns,l)
78

79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
let parse_record env loc f r =
  let r = List.map (fun (l,x) -> (parse_label env loc l, f x)) r in
  LabelMap.from_list (fun _ _ -> raise_loc_generic loc "Duplicated record field") r

let rec const env loc = function
  | LocatedExpr (loc,e) -> const env loc e
  | Pair (x,y) -> Types.Pair (const env loc x, const env loc y)
  | Xml (x,y) -> Types.Xml (const env loc x, const env loc y)
  | RecordLitt x -> Types.Record (parse_record env loc (const env loc) x)
  | String (i,j,s,c) -> Types.String (i,j,s,const env loc c)
  | Atom t -> Types.Atom (parse_atom env loc t)
  | Integer i -> Types.Integer i
  | Char c -> Types.Char c
  | _ -> raise_loc_generic loc "This should be a scalar or structured constant"

(* I. Transform the abstract syntax of types and patterns into
      the internal form *)
96

97
exception NonExhaustive of Types.descr
98
exception Constraint of Types.descr * Types.descr
99
exception ShouldHave of Types.descr * string
100
exception ShouldHave2 of Types.descr * string * Types.descr
101
exception WrongLabel of Types.descr * label
102
exception UnboundId of id * bool
103
exception Error of string
104

105
106
let raise_loc loc exn = raise (Location (loc,`Full,exn))
let raise_loc_str loc ofs exn = raise (Location (loc,`Char ofs,exn))
107
let error loc msg = raise_loc loc (Error msg)
108

109
110
111
  (* Schema datastructures *)

module StringSet = Set.Make (String)
112
113
114

  (* just to remember imported schemas *)
let schemas = State.ref "Typer.schemas" StringSet.empty
115
116
117

let schema_types = State.ref "Typer.schema_types" (Hashtbl.create 51)
let schema_elements = State.ref "Typer.schema_elements" (Hashtbl.create 51)
118
let schema_attributes = State.ref "Typer.schema_attributes" (Hashtbl.create 51)
119

120
121
122
123
124
125
126
127
(* 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

128
129
130
131
132
133
134
135
136
137
138
139
140
141
(* We use two intermediate representation from AST types/patterns
   to internal ones:

      AST -(1)-> derecurs -(2)-> slot -(3)-> internal

   (1) eliminate recursion, schema, 
       propagate sequence capture variables, keep regexps

   (2) stratify, detect ill-formed recursion, compile regexps

   (3) check additional constraints on types / patterns;
       deep (recursive) hash-consing
*)     

142
143
144
145
type derecurs_slot = {
  ploc : Location.loc;
  pid  : int;
  mutable ploop : bool;
146
  mutable pdescr : derecurs;
147
} and derecurs =
148
  | PDummy
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
  | 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

170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189

type descr = 
  | IDummy
  | 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
  | IOptional of descr
  | 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;
  mutable d    : descr;
190
}
191
192
193
194
195
196
197
198
199
200
201
202
203
    

let counter = ref 0
let mk_derecurs_slot loc = 
  incr counter; 
  { ploop = false; ploc = loc; pid = !counter; pdescr = PDummy }
	  
let mk_slot () = 
  { d=IDummy; fv=None; hash=None; rank1=0; rank2=0; gen1=0; gen2=0 } 


(* This environment is used in phase (1) to eliminate recursion *)
type penv = {
204
  penv_tenv : env;
205
206
207
208
  penv_derec : derecurs_slot Env.t;
}

let penv tenv = { penv_tenv = tenv; penv_derec = Env.empty }
209

210
let rec hash_derecurs = function
211
  | PDummy -> assert false
212
213
214
  | PAlias s -> 
      s.pid
  | PType t -> 
215
      1 + 17 * (Types.hash t)
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
  | 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)
238
and hash_derecurs_regexp = function
239
240
241
242
243
244
245
246
247
248
249
250
  | 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)
251
252

let rec equal_derecurs p1 p2 = (p1 == p2) || match p1,p2 with
253
254
255
  | PAlias s1, PAlias s2 -> 
      s1 == s2
  | PType t1, PType t2 -> 
256
      Types.equal t1 t2
257
258
259
260
261
  | 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)
262
263
264
265
266
267
268
269
270
271
272
273
  | 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)
274
275
  | _ -> false
and equal_derecurs_regexp r1 r2 = match r1,r2 with
276
277
278
279
  | PEpsilon, PEpsilon -> 
      true
  | PElem p1, PElem p2 -> 
      equal_derecurs p1 p2
280
  | PSeq (p1,q1), PSeq (p2,q2) 
281
282
  | PAlt (p1,q1), PAlt (p2,q2) -> 
      (equal_derecurs_regexp p1 p2) && (equal_derecurs_regexp q1 q2)
283
  | PStar p1, PStar p2
284
285
  | PWeakStar p1, PWeakStar p2 -> 
      equal_derecurs_regexp p1 p2
286
  | _ -> false
287

288
289
290
291
292
293
294
295
296
297
298
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 
299
300
301
302
    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)
303
304
  end
)
305

306
307
308
309
let gen = ref 0
let rank = ref 0
	     
let rec hash_descr = function
310
  | IDummy -> assert false
311
  | IType x -> Types.hash x
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
  | 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;
327
    hash_descr s.d
328
329
330
331
  )
    
let rec equal_descr d1 d2 = 
  match (d1,d2) with
332
  | IType x1, IType x2 -> Types.equal x1 x2
333
334
335
336
337
338
339
  | 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)
340
341
  | IRecord (o1,r1), IRecord (o2,r2) -> 
      (o1 = o2) && (LabelMap.equal equal_slot r1 r2)
342
  | ICapture x1, ICapture x2 -> Id.equal x1 x2
343
344
  | IConstant (x1,y1), IConstant (x2,y2) -> 
      (Id.equal x1 x2) && (Types.equal_const y1 y2)
345
346
347
348
349
350
351
352
  | _ -> 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;
353
     equal_descr s1.d s2.d
354
355
   ))
  
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
module SlotTable = Hashtbl.Make(
  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
	      
    let equal s1 s2 = 
      (s1 == s2) || 
      (incr gen; rank := 0; 
       let e = equal_slot s1 s2 in
       (*     if e then Printf.eprintf "Recursive hash-consing: Equal\n";  *)
       e)
  end)


let rec derecurs env p = match p.descr with
  | PatVar v ->
      (try PAlias (Env.find v env.penv_derec)
       with Not_found -> 
382
	 try PType (find_type v env.penv_tenv)
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
	 with Not_found -> PCapture v)
  | SchemaVar (kind, schema, item) ->
      PType (derecurs_schema env kind schema item)
  | Recurs (p,b) -> derecurs (derecurs_def env b) p
  | Internal t -> PType t
  | NsT ns -> PType (Types.atom (Atoms.any_in_ns (parse_ns env.penv_tenv p.loc ns)))
  | 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, parse_record env.penv_tenv p.loc (derecurs env) r)
  | Constant (x,c) -> PConstant (x,const env.penv_tenv p.loc c)
  | Cst c -> PType (Types.constant (const env.penv_tenv p.loc 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 -> 
      PWeakStar (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_derecurs_slot p.loc)) b in
  let n = 
    List.fold_left (fun env (v,p,s) -> Env.add v s env) env.penv_derec b in
  let env = { env with penv_derec = n } in
  List.iter (fun (v,p,s) -> s.pdescr <- derecurs env p) b;
  env

and derecurs_schema env kind schema item =
  let elt () = fst (Hashtbl.find !schema_elements (schema, item)) in
  let typ () = Hashtbl.find !schema_types (schema, item) in
  let att () = Hashtbl.find !schema_attributes (schema, item) in
  let rec do_try n = function
    | [] -> 
	let s = Printf.sprintf 
		  "No %s named '%s' found in schema '%s'" n item schema in
	failwith s
    | f :: rem -> (try f () with Not_found -> do_try n rem)  in
  match kind with
    | `Element -> do_try "element" [ elt ]
    | `Type -> do_try "type" [ typ ]
    | `Attribute -> do_try "atttribute" [ att ]
    | `Any -> do_try "item" [ elt; typ; att ]

    
448
449
450
451
452
let rec fv_slot s =
  match s.fv with
    | Some x -> x
    | None ->
	if s.gen1 = !gen then IdSet.empty 
453
	else (s.gen1 <- !gen; fv_descr s.d)
454
and fv_descr = function
455
  | IDummy -> assert false
456
  | IType _ -> IdSet.empty
457
458
459
460
461
462
463
  | 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)
464
465
  | IRecord (o,r) -> 
      List.fold_left IdSet.cup IdSet.empty (LabelMap.map_to_list fv_slot r)
466
  | ICapture x | IConstant (x,_) -> IdSet.singleton x
467

468
469
470
471
472
473
474
475
let compute_fv s =
  match s.fv with
    | Some x -> ()
    | None ->
	incr gen;
	let x = fv_slot s in
	s.fv <- Some x
	  
476
477
478
let check_no_capture loc s =
  match IdSet.pick s with
    | Some x ->  
479
	raise_loc_generic loc ("Unbound type name " ^ (Ident.to_string x))
480
    | None -> ()
481
    
482
483
484
let compile_slot_hash = DerecursTable.create 67
let compile_hash = DerecursTable.create 67

485
486
let todo_defs = ref []
let todo_fv = ref []
487
488
489
490
491
492
493
494

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
495
  | PDummy -> assert false
496
497
498
499
  | PAlias v ->
      if v.ploop then
	raise_loc_generic v.ploc ("Unguarded recursion on type/pattern");
      v.ploop <- true;
500
      let r = compile v.pdescr in
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
      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
522
	| PEpsilon ->
523
524
525
	    (match q with 
	       | PRegexp (r,q) -> aux accu r q 
	       | _ -> (compile q) :: accu)
526
527
528
529
530
531
532
533
534
535
536
537
538
	| 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
539
  with Not_found ->
540
541
542
    let s = mk_slot () in
    todo_defs := (s,p) :: !todo_defs;
    todo_fv := s :: !todo_fv;
543
    DerecursTable.add compile_slot_hash p s;
544
    s
545

546
      
547
let timer_fv = Stats.Timer.create "Typer.fv"
548
let rec flush_defs () = 
549
550
551
552
553
554
555
556
557
558
  match !todo_defs with
    | [] -> 
	Stats.Timer.start timer_fv;
	List.iter compute_fv !todo_fv;
	Stats.Timer.stop timer_fv;
	todo_fv := []
    | (s,p)::t -> 
	todo_defs := t; 
	s.d <- compile p; 
	flush_defs ()
559
560
561
562
563
564
565
566
567
568
569
570
571
572
	
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)
573
  | IDummy | ICapture _ | IConstant (_,_) -> assert false
574
      
575
and typ_node s : Types.Node.t =
576
577
578
579
  try SlotTable.find typ_nodes s
  with Not_found ->
    let x = Types.make () in
    SlotTable.add typ_nodes s x;
580
    Types.define x (typ s.d);
581
582
583
584
585
586
587
588
    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
589
  | IDummy -> assert false
590
591
592
593
594
595
  | 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 _ ->
596
      raise (Patterns.Error "Differences are not allowed in patterns")
597
598
599
  | ITimes (s1,s2) -> Patterns.times (pat_node s1) (pat_node s2)
  | IXml (s1,s2) -> Patterns.xml (pat_node s1) (pat_node s2)
  | IOptional _ -> 
600
      raise (Patterns.Error "Optional fields are not allowed in record patterns")
601
602
603
604
605
606
607
608
609
610
611
612
613
614
  | 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 _ ->
615
      raise (Patterns.Error "Arrows are not allowed in patterns")
616
617
618
619
620
621
  | 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
622
623
    try
      SlotTable.add pat_nodes s x;
624
      Patterns.define x (pat s.d);
625
626
627
      x
    with exn -> SlotTable.remove pat_nodes s; raise exn
      (* For the toplevel ... *)
628

629

630
let type_defs env b =
631
632
  List.iter 
    (fun (v,p) ->
633
634
       if Env.mem v env.ids
       then raise_loc_generic p.loc ("Identifier " ^ (Ident.to_string v) ^ " is already bound")
635
    ) b;
636
637
  let penv = derecurs_def (penv env) b in
  let b = List.map (fun (v,p) -> (v,p,compile (derecurs penv p))) b in
638
639
640
641
  flush_defs ();
  let b = 
    List.map 
      (fun (v,p,s) -> 
642
	 check_no_capture p.loc (fv_descr s);
643
644
645
	 let t = typ s in
	 if (p.loc <> noloc) && (Types.is_empty t) then
	   warning p.loc 
646
	     ("This definition yields an empty type for " ^ (Ident.to_string v));
647
	 (v,t)) b in
648
  List.iter (fun (v,t) -> Types.Print.register_global (Id.value v) t) b;
649
  b
650
651


652
653
654
655
656
let dump_types ppf env =
  Env.iter (fun v -> 
	      function 
		  (Type _) -> Format.fprintf ppf " %a" Ident.print v
		| _ -> ()) env.ids
657

658
659
let dump_ns ppf env =
  Ns.dump_table ppf env.tenv_nspref
660

661

662
663
let do_typ loc r = 
  let s = compile_slot r in
664
  flush_defs ();
665
666
  check_no_capture loc (fv_slot s);
  typ_node s
667
   
668
669
let typ env p =
  do_typ p.loc (derecurs (penv env) p)
670
    
671
672
let pat env p = 
  let s = compile_slot (derecurs (penv env) p) in
673
674
675
  flush_defs ();
  try pat_node s
  with Patterns.Error e -> raise_loc_generic p.loc e
676
    | Location (loc,_,exn) when loc = noloc -> raise (Location (p.loc, `Full, exn))
677
678


679
680
(* II. Build skeleton *)

681

682
type op = [ `Unary of env -> Typed.unary_op | `Binary of env -> Typed.binary_op ]
683
684
685
686
687
688
let op_table : (string,op) Hashtbl.t = Hashtbl.create 31
let register_unary_op s f = Hashtbl.add op_table s (`Unary f)
let register_binary_op s f = Hashtbl.add op_table s (`Binary f)
let find_op s = Hashtbl.find op_table s


689
module Fv = IdSet
690

691
692
693
type branch = Branch of Typed.branch * branch list

let cur_branch : branch list ref = ref []
694

695
let exp loc fv e =
696
697
  fv,
  { Typed.exp_loc = loc;
698
    Typed.exp_typ = Types.empty;
699
    Typed.exp_descr = e;
700
  }
701
702


703
704
let rec expr env loc = function
  | LocatedExpr (loc,e) -> expr env loc e
705
  | Forget (e,t) ->
706
      let (fv,e) = expr env loc e and t = typ env t in
707
708
709
710
      exp loc fv (Typed.Forget (e,t))
  | Var s -> 
      exp loc (Fv.singleton s) (Typed.Var s)
  | Apply (e1,e2) -> 
711
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
712
713
      exp loc (Fv.cup fv1 fv2) (Typed.Apply (e1,e2))
  | Abstraction a ->
714
      let iface = List.map (fun (t1,t2) -> (typ env t1, typ env t2)) 
715
716
717
718
719
720
721
		    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
722
      let (fv0,body) = branches env a.fun_body in
723
724
725
726
727
728
729
730
731
732
733
      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
734
  | (Integer _ | Char _ | Atom _) as c -> 
735
      exp loc Fv.empty (Typed.Cst (const env loc c))
736
  | Pair (e1,e2) ->
737
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
738
739
      exp loc (Fv.cup fv1 fv2) (Typed.Pair (e1,e2))
  | Xml (e1,e2) ->
740
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
741
742
      exp loc (Fv.cup fv1 fv2) (Typed.Xml (e1,e2))
  | Dot (e,l) ->
743
744
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.Dot (e,parse_label env loc l))
745
  | RemoveField (e,l) ->
746
747
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.RemoveField (e,parse_label env loc l))
748
749
  | RecordLitt r -> 
      let fv = ref Fv.empty in
750
      let r = parse_record env loc
751
		(fun e -> 
752
		   let (fv2,e) = expr env loc e 
753
754
755
		   in fv := Fv.cup !fv fv2; e)
		r in
      exp loc !fv (Typed.RecordLitt r)
756
  | String (i,j,s,e) ->
757
      let (fv,e) = expr env loc e in
758
      exp loc fv (Typed.String (i,j,s,e))
759
  | Op (op,le) ->
760
      let (fvs,ltes) = List.split (List.map (expr env loc) le) in
761
      let fv = List.fold_left Fv.cup Fv.empty fvs in
762
      (try
763
	 (match (ltes,find_op op) with
764
765
	    | [e], `Unary op -> exp loc fv (Typed.UnaryOp (op env, e))
	    | [e1;e2], `Binary op -> exp loc fv (Typed.BinaryOp (op env, e1,e2))
766
767
768
	    | _ -> assert false)
       with Not_found -> assert false)

769
  | Match (e,b) -> 
770
771
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
772
      exp loc (Fv.cup fv1 fv2) (Typed.Match (e, b))
773
  | Map (e,b) ->
774
775
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
776
777
      exp loc (Fv.cup fv1 fv2) (Typed.Map (e, b))
  | Transform (e,b) ->
778
779
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
780
      exp loc (Fv.cup fv1 fv2) (Typed.Transform (e, b))
781
  | Xtrans (e,b) ->
782
783
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
784
      exp loc (Fv.cup fv1 fv2) (Typed.Xtrans (e, b))
785
  | Validate (e,schema,elt) ->
786
      let (fv,e) = expr env loc e in
787
      exp loc fv (Typed.Validate (e, schema, elt))
788
  | Try (e,b) ->
789
790
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
791
      exp loc (Fv.cup fv1 fv2) (Typed.Try (e, b))
792
  | NamespaceIn (pr,ns,e) ->
793
794
      let env = enter_ns pr ns env in
      expr env loc e
795
  | Ref (e,t) ->
796
      let (fv,e) = expr env loc e and t = typ env t in
797
      exp loc fv (Typed.Ref (e,t))
798
	      
799
  and branches env b = 
800
    let fv = ref Fv.empty in
801
    let accept = ref Types.empty in
802
    let branch (p,e) = 
803
804
      let cur_br = !cur_branch in
      cur_branch := [];
805
      let (fv2,e) = expr env noloc e in
806
      let br_loc = merge_loc p.loc e.Typed.exp_loc in
807
      let p = pat env p in
808
809
810
811
812
813
      (match Fv.pick (Fv.diff (Patterns.fv p) fv2) with
	| None -> ()
	| Some x ->
	    let x = U.to_string (Id.value x) in
	    warning br_loc 
	      ("The capture variable " ^ x ^ 
814
	       " is declared in the pattern but not used in the body of this branch. It might be a misspelled type or name (if not use _ instead)."));
815
816
817
818
819
820
821
822
823
      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
824
      cur_branch := Branch (br, !cur_branch) :: cur_br;
825
826
      br in
    let b = List.map branch b in
827
828
829
830
    (!fv, 
     { 
       Typed.br_typ = Types.empty; 
       Typed.br_branches = b; 
831
832
       Typed.br_accept = !accept;
       Typed.br_compiled = None;
833
834
     } 
    )
835

836
let expr env = expr env noloc
837

838
839
840
let let_decl env p e =
  let (_,e) = expr env e in
  { Typed.let_pat = pat env p;
841
842
843
    Typed.let_body = e;
    Typed.let_compiled = None }

844
845
846

(* Hide global "typing/parsing" environment *)

847

848
849
(* III. Type-checks *)

850
851
open Typed

852
853
let require loc t s = 
  if not (Types.subtype t s) then raise_loc loc (Constraint (t, s))
854

855
856
857
let check loc t s = 
  require loc t s; t

858
859
860
861
862
let check_str loc ofs t s = 
  if not (Types.subtype t s) then raise_loc_str loc ofs (Constraint (t, s));
  t

let should_have loc constr s = 
863
864
  raise_loc loc (ShouldHave (constr,s))

865
866
867
let should_have_str loc ofs constr s = 
  raise_loc_str loc ofs (ShouldHave (constr,s))

868
869
870
871
872
873
874
875
876
877
878
let flatten loc arg constr precise =
  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 = arg sconstr' precise in
    if precise then Sequence.flatten t else constr
  else
    let t = arg sconstr' true in
    Sequence.flatten t
879

880
881
let rec type_check env e constr precise = 
  let d = type_check' e.exp_loc env e.exp_descr constr precise in
882
  let d = if precise then d else constr in
883
884
885
  e.exp_typ <- Types.cup e.exp_typ d;
  d

886
and type_check' loc env e constr precise = match e with
887
888
889
  | Forget (e,t) ->
      let t = Types.descr t in
      ignore (type_check env e t false);
890
891
      check loc t constr

892
  | Abstraction a ->
893
894
895
      let t =
	try Types.Arrow.check_strenghten a.fun_typ constr 
	with Not_found -> 
896
897
	  should_have loc constr
	    "but the interface of the abstraction is not compatible"
898
      in
899
900
      let env = match a.fun_name with
	| None -> env
901
	| Some f -> enter_value f a.fun_typ env in
902
903
      List.iter 
	(fun (t1,t2) ->
904
905
906
	   let acc = a.fun_body.br_accept in 
	   if not (Types.subtype t1 acc) then
	     raise_loc loc (NonExhaustive (Types.diff t1 acc));
907
	   ignore (type_check_branches loc env t1 a.fun_body t2 false)
908
909
	) a.fun_iface;
      t
910

911
912
  | Match (e,b) ->
      let t = type_check env e b.br_accept true in
913
      type_check_branches loc env t b constr precise
914
915
916

  | Try (e,b) ->
      let te = type_check env e constr precise in
917
      let tb = type_check_branches loc env Types.any b constr precise in
918
      Types.cup te tb
919

920
921
  | Pair (e1,e2) ->
      type_check_pair loc env e1 e2 constr precise
922

923
924
  | Xml (e1,e2) ->
      type_check_pair ~kind:`XML loc env e1 e2 constr precise
925

926
  | RecordLitt r ->
927
928
929
930
931
932
933
934
      type_record loc env r constr precise

  | Map (e,b) ->
      type_map loc env false e b constr precise

  | Transform (e,b) ->
      flatten loc (type_map loc env true e b) constr precise

935
936
937
938
  | Apply (e1,e2) ->
      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
939
940
941
942
943
944
945
      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
946
947
948
      check loc res constr

  | UnaryOp (o,e) ->
949
950
      let t = o.un_op_typer loc 
		(type_check env e) constr precise in
951
952
953
      check loc t constr

  | BinaryOp (o,e1,e2) ->
954
955
956
      let t = o.bin_op_typer loc 
		(type_check env e1) 
		(type_check env e2) constr precise in
957
958
959
960
      check loc t constr

  | Var s -> 
      let t = 
961
	try find_value s env
962
	with Not_found -> 
963
	  raise_loc loc (UnboundId (s, Env.mem s env.ids) ) in
964
965
966
967
968
      check loc t constr
      
  | Cst c -> 
      check loc (Types.constant c) constr

969
970
971
  | String (i,j,s,e) ->
      type_check_string loc env 0 s i j e constr precise

972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
  | Dot (e,l) ->
      let t = type_check env e Types.Record.any true in
      let t = 
        try (Types.Record.project t l) 
        with Not_found -> raise_loc loc (WrongLabel(t,l))
      in
      check loc t constr

  | RemoveField (e,l) ->
      let t = type_check env e Types.Record.any true in
      let t = Types.Record.remove_field t l in
      check loc t constr

  | Xtrans (e,b) ->
      let t = type_check env e Sequence.any true in
      let t = 
	Sequence.map_tree 
	  (fun t ->
	     let resid = Types.diff t b.br_accept in
	     let res = type_check_branches loc env t b Sequence.any true in
	     (res,resid)
	  ) t in
      check loc t constr

996
997
998
999
  | Validate (e, schema_name, elt_name) ->
      ignore (type_check env e Types.any false);
      let t = fst (Hashtbl.find !schema_elements (schema_name, elt_name)) in
      check loc t constr
1000

1001
1002
1003
1004
  | Ref (e,t) ->
      ignore (type_check env e (Types.descr t) false);
      check loc (Builtin_defs.ref_type t) constr

1005
and type_check_pair ?(kind=`Normal) loc env e1 e2 constr precise =
1006
  let rects = Types.Product.normal ~kind constr in
1007
1008
  if Types.Product.is_empty rects then 
    (match kind with
1009
1010
      | `Normal -> should_have loc constr "but it is a pair"
      | `XML -> should_have loc constr "but it is an XML element");
1011
  let need_s = Types.Product.need_second rects in
1012
1013
1014
1015
1016
  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
1017

1018
  if precise then 
1019
1020
1021
    match kind with
      | `Normal -> Types.times (Types.cons t1) (Types.cons t2)
      | `XML -> Types.xml (Types.cons t1) (Types.cons t2)
1022
1023
1024
  else
    constr

1025
1026
1027
1028
1029
1030
1031
1032
1033
and type_check_string loc env ofs s i j e constr precise =
  if U.equal_index i j then type_check env e constr precise
  else
    let rects = Types.Product.normal constr in
    if Types.Product.is_empty rects 
    then should_have_str loc ofs constr "but it is a string"
    else
      let need_s = Types.Product.need_second rects in
      let (ch,i') = U.next s i in
1034
      let ch = Chars.V.mk_int ch in
1035
1036
1037
1038
1039
1040
1041
      let tch = Types.constant (Types.Char ch) in
      let t1 = check_str loc ofs tch (Types.Product.pi1 rects) in
      let c2 = Types.Product.constraint_on_2 rects t1 in
      let t2 = type_check_string loc env (ofs + 1) s i' j e c2 precise in
      if precise then Types.times (Types.cons t1) (Types.cons t2)
      else constr

1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
and type_record loc env r constr precise =
(* try to get rid of precise = true for values of fields *)
(* also: the use equivalent of need_second to optimize... *)
  if not (Types.Record.has_record constr) then
    should_have loc constr "but it is a record";
  let (rconstr,res) = 
    List.fold_left
      (fun (rconstr,res) (l,e) ->
	 (* could compute (split l e) once... *)
	 let pi = Types.Record.project_opt rconstr l in
	 if Types.is_empty pi then 
1053
	   (let l = Label.to_string (LabelPool.value l) in
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
	    should_have loc constr
	      (Printf.sprintf "Field %s is not allowed here." l));
	 let t = type_check env e pi true in
	 let rconstr = Types.Record.condition rconstr l t in
	 let res = (l,Types.cons t) :: res in
	 (rconstr,res)
      ) (constr, []) (LabelMap.get r)
  in
  if not (Types.Record.has_empty_record rconstr) then
    should_have loc constr "More fields should be present";
  let t = 
    Types.record' (false, LabelMap.from_list (fun _ _ -> assert false) res)
  in
  check loc t constr
1068

1069

1070
and type_check_branches loc env targ brs constr precise =
1071
  if Types.is_empty targ then Types.empty
1072
1073
  else (
    brs.br_typ <- Types.cup brs.br_typ targ;