typer.ml 56 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 debug_schema = false

13
let warning loc msg =
14
  Format.fprintf !Location.warning_ppf "Warning %a:@\n%a%s@." 
15
16
    Location.print_loc (loc,`Full)
    Location.html_hilight (loc,`Full)
17
18
    msg

19
20
21
22
23
24
25
26
27
exception NonExhaustive of Types.descr
exception Constraint of Types.descr * Types.descr
exception ShouldHave of Types.descr * string
exception ShouldHave2 of Types.descr * string * Types.descr
exception WrongLabel of Types.descr * label
exception UnboundId of id * bool
exception UnboundExtId of Types.CompUnit.t * id
exception Error of string

28
29
30

exception Warning of string * Types.t

31
32
33
34
let raise_loc loc exn = raise (Location (loc,`Full,exn))
let raise_loc_str loc ofs exn = raise (Location (loc,`Char ofs,exn))
let error loc msg = raise_loc loc (Error msg)

35
36
type item =
  | Type of Types.t
37
  | Val of Types.t
38

39
40
module UEnv = Map.Make(U)

41
type t = {
42
  ids : item Env.t;
43
  ns: Ns.table;
44
  cu: Types.CompUnit.t UEnv.t;
45
  schemas: string UEnv.t
46
}
47

48
49
50
51
52
let hash _ = failwith "Typer.hash"
let compare _ _ = failwith "Typer.compare"
let dump ppf _ = failwith "Typer.dump"
let equal _ _ = failwith "Typer.equal"
let check _ = failwith "Typer.check"
53
54

(* TODO: filter out builtin defs ? *)
55
56
57
58
let serialize_item s = function
  | Type t -> Serialize.Put.bits 1 s 0; Types.serialize s t
  | Val t -> Serialize.Put.bits 1 s 1; Types.serialize s t

59
let serialize s env =
60
  Serialize.Put.env Id.serialize serialize_item Env.iter s env.ids;
61
  Ns.serialize_table s env.ns
62

63
64
65
66
67
let deserialize_item s = match Serialize.Get.bits 1 s with
  | 0 -> Type (Types.deserialize s)
  | 1 -> Val (Types.deserialize s)
  | _ -> assert false

68
let deserialize s =
69
  let ids = Serialize.Get.env Id.deserialize deserialize_item Env.add Env.empty s in
70
  let ns = Ns.deserialize_table s in
71
  { ids = ids; ns = ns; cu = UEnv.empty; schemas = UEnv.empty }
72
73


74
75
let empty_env = {
  ids = Env.empty;
76
  ns = Ns.empty_table;
77
  cu = UEnv.empty;
78
  schemas = UEnv.empty
79
80
}

81
82
let from_comp_unit = ref (fun cu -> assert false)

83
let enter_cu x cu env =
84
  { env with cu = UEnv.add x cu env.cu }
85

86
87
88
let find_cu x env =
  try UEnv.find x env.cu
  with Not_found -> Types.CompUnit.mk x
89
90


91
92
93
94
95
96
let enter_schema x uri env =
  { env with schemas = UEnv.add x uri env.schemas }
let find_schema x env =
  try UEnv.find x env.schemas
  with Not_found -> raise (Error (Printf.sprintf "%s: no such schema" (U.get_str x)))

97
98
99
100
101
102
103
104
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
105
    | Val _ -> raise Not_found
106

107
let find_type_global loc cu id env =
108
  let cu = find_cu cu env in
109
110
111
  let env = !from_comp_unit cu in
  find_type id env

112
let enter_value id t env = 
113
  { env with ids = Env.add id (Val t) env.ids }
114
115
let enter_values l env =
  { env with ids = 
116
      List.fold_left (fun accu (id,t) -> Env.add id (Val t) accu) env.ids l }
117
118
119
let enter_values_dummy l env =
  { env with ids = 
      List.fold_left (fun accu id -> Env.add id (Val Types.empty) accu) env.ids l }
120
121
let find_value id env =
  match Env.find id env.ids with
122
    | Val t -> t
123
    | _ -> raise Not_found
124
125
126
let find_value_global cu id env =
  let env = !from_comp_unit cu in
  find_value id env
127
	
128
129
130
131
132
133
let value_name_ok id env =
  try match Env.find id env.ids with
    | Val t -> true
    | _ -> false
  with Not_found -> true

134
135
136
137
let iter_values env f =
  Env.iter (fun x ->
	      function Val t -> f x t;
		| _ -> ()) env.ids
138

139

140
141
142
143
144
145
146
147
148
let register_types cu env =
  let prefix = U.concat (Types.CompUnit.value cu) (U.mk ":") in
  Env.iter (fun x ->
	      function 
		| Type t ->
		    let n = U.concat prefix (Id.value x) in
		    Types.Print.register_global n t
		| _ -> ()) env.ids

149

150
(* Namespaces *)
151

152
let set_ns_table_for_printer env = 
153
  Ns.InternalPrinter.set_table env.ns
154

155
let get_ns_table tenv = tenv.ns
156

157
let enter_ns p ns env =
158
  { env with ns = Ns.add_prefix p ns env.ns }
159

160
161
162
163
164
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))
165

166
let parse_atom env loc t =
167
  let (ns,l) = protect_error_ns loc (Ns.map_tag env.ns) t in
168
169
170
  Atoms.V.mk ns l
 
let parse_ns env loc ns =
171
  protect_error_ns loc (Ns.map_prefix env.ns) ns
172

173
let parse_label env loc t =
174
  let (ns,l) = protect_error_ns loc (Ns.map_attr env.ns) t in
175
  LabelPool.mk (ns,l)
176

177
178
179
180
181
182
183
184
185
186
187
188
189
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
190
  | Const c -> c
191
192
193
194
  | _ -> 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 *)
195

196

197
(* Schema *)
198

199
200
201
let is_registered_schema env s = UEnv.mem s env.schemas

(* uri -> schema binding *)
202
let schemas = State.ref "Typer.schemas" (Hashtbl.create 3)
203
204
205

let schema_types = State.ref "Typer.schema_types" (Hashtbl.create 51)
let schema_elements = State.ref "Typer.schema_elements" (Hashtbl.create 51)
206
let schema_attributes = State.ref "Typer.schema_attributes" (Hashtbl.create 51)
207
208
209
210
let schema_attribute_groups =
  State.ref "Typer.schema_attribute_groups" (Hashtbl.create 51)
let schema_model_groups =
  State.ref "Typer.schema_model_groups" (Hashtbl.create 51)
211

212
213


214
215
  (* raise Not_found *)

216
217
218
219

let get_schema_fwd = ref (fun _ -> assert false)

let find_schema_descr_uri kind uri name =
220
  try
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
    ignore (!get_schema_fwd uri);
    let elt () = Hashtbl.find !schema_elements (uri, name) in
    let typ () = Hashtbl.find !schema_types (uri, name) in
    let att () = Hashtbl.find !schema_attributes (uri, name) in
    let att_group () = Hashtbl.find !schema_attribute_groups (uri, name) in
    let mod_group () = Hashtbl.find !schema_model_groups (uri, name) in
    let rec do_try n = function
      | [] -> raise Not_found
      | f :: rem -> (try f () with Not_found -> do_try n rem)
    in
    match kind with
      | Some `Element -> do_try "element" [ elt ]
      | Some `Type -> do_try "type" [ typ ]
      | Some `Attribute -> do_try "atttribute" [ att ]
      | Some `Attribute_group -> do_try "attribute group" [ att_group ]
      | Some `Model_group -> do_try "model group" [ mod_group ]
      | None ->
          (* policy for unqualified schema component resolution. This order should
           * be consistent with Schema_component.get_component *)
          do_try "component" [ elt; typ; att; att_group; mod_group ]
    with Not_found ->    
242
      raise (Error (Printf.sprintf "No %s named '%s' found in schema '%s'"
243
244
245
246
247
248
		      (Schema_common.string_of_component_kind kind) (U.get_str name) uri))

let find_schema_descr env kind schema name =
  let uri = find_schema schema env in
  find_schema_descr_uri kind uri name

249

250
251
(* Eliminate Recursion, propagate Sequence Capture Variables *)

252
253
254
255
256
257
258
259
260
261
262
263
264
265
(* 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
*)     

266
267
268
269
type derecurs_slot = {
  ploc : Location.loc;
  pid  : int;
  mutable ploop : bool;
270
  mutable pdescr : derecurs;
271
} and derecurs =
272
  | PDummy
273
274
275
276
277
278
279
280
281
  | 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
282
  | PRecord of bool * (derecurs * derecurs option) label_map
283
284
  | PCapture of id
  | PConstant of id * Types.const
285
  | PRegexp of derecurs_regexp
286
287
288
and derecurs_regexp =
  | PEpsilon
  | PElem of derecurs
289
  | PGuard of derecurs
290
291
292
293
294
  | PSeq of derecurs_regexp * derecurs_regexp
  | PAlt of derecurs_regexp * derecurs_regexp
  | PStar of derecurs_regexp
  | PWeakStar of derecurs_regexp

295
296
297
298
299
300
301
302
303
304
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
305
  | IRecord of bool * (slot * descr option) label_map
306
307
308
309
310
311
312
313
  | 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;
314
}
315
316
317
318
319
320
321
322
323
324
325
326
327
    

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 = {
328
  penv_tenv : t;
329
330
331
332
  penv_derec : derecurs_slot Env.t;
}

let penv tenv = { penv_tenv = tenv; penv_derec = Env.empty }
333

334
let rec hash_derecurs = function
335
  | PDummy -> assert false
336
337
338
  | PAlias s -> 
      s.pid
  | PType t -> 
339
      1 + 17 * (Types.hash t)
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
  | 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) -> 
355
      (if o then 9 else 10) + 17 * (LabelMap.hash hash_derecurs_field r)
356
357
358
  | PCapture x -> 
      11 + 17 * (Id.hash x)
  | PConstant (x,c) -> 
359
      12 + 17 * (Id.hash x) + 257 * (Types.Const.hash c)
360
361
  | PRegexp p -> 
      13 + 17 * (hash_derecurs_regexp p)
362
363
364
and hash_derecurs_field = function
  | (p, Some e) -> 1 + 17 * hash_derecurs p + 257 * hash_derecurs e
  | (p, None) -> 2 + 17 * hash_derecurs p
365
and hash_derecurs_regexp = function
366
367
368
369
370
371
372
373
374
375
376
377
  | 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)
378
379
  | PGuard p ->
      7 + 17 * (hash_derecurs p)
380
381

let rec equal_derecurs p1 p2 = (p1 == p2) || match p1,p2 with
382
383
384
  | PAlias s1, PAlias s2 -> 
      s1 == s2
  | PType t1, PType t2 -> 
385
      Types.equal t1 t2
386
387
388
389
390
  | 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)
391
392
393
394
395
  | 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) -> 
396
      (o1 == o2) && (LabelMap.equal equal_derecurs_field r1 r2)
397
398
399
  | PCapture x1, PCapture x2 -> 
      Id.equal x1 x2
  | PConstant (x1,c1), PConstant (x2,c2) -> 
400
      (Id.equal x1 x2) && (Types.Const.equal c1 c2)
401
402
  | PRegexp p1, PRegexp p2 -> 
      equal_derecurs_regexp p1 p2
403
  | _ -> false
404
405
406
407
and equal_derecurs_field r1 r2 = match (r1,r2) with
  | (p1,None),(p2,None) -> equal_derecurs p1 p2
  | (p1, Some e1), (p2, Some e2) -> equal_derecurs p1 p2 && equal_derecurs e1 e2
  | _ -> false
408
and equal_derecurs_regexp r1 r2 = match r1,r2 with
409
410
411
412
  | PEpsilon, PEpsilon -> 
      true
  | PElem p1, PElem p2 -> 
      equal_derecurs p1 p2
413
414
  | PGuard p1, PGuard p2 ->
      equal_derecurs p1 p2
415
  | PSeq (p1,q1), PSeq (p2,q2) 
416
417
  | PAlt (p1,q1), PAlt (p2,q2) -> 
      (equal_derecurs_regexp p1 p2) && (equal_derecurs_regexp q1 q2)
418
  | PStar p1, PStar p2
419
420
  | PWeakStar p1, PWeakStar p2 -> 
      equal_derecurs_regexp p1 p2
421
  | _ -> false
422

423
424
425
426
427
428
429
430
module DerecursTable = Hashtbl.Make(
  struct 
    type t = derecurs 
    let hash = hash_derecurs
    let equal = equal_derecurs
  end
)

431
432
433
434
let gen = ref 0
let rank = ref 0
	     
let rec hash_descr = function
435
  | IDummy -> assert false
436
  | IType x -> Types.hash x
437
438
439
440
441
442
443
  | 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)
444
  | IRecord (o,r) -> (if o then 8 else 9) + 17 * (LabelMap.hash hash_descr_field r)
445
  | ICapture x -> 10 + 17 * (Id.hash x)
446
  | IConstant (x,y) -> 11 + 17 * (Id.hash x) + 257 * (Types.Const.hash y)
447
448
449
and hash_descr_field = function
  | (d, Some e) -> 1 + 17 * hash_slot d + 257 * hash_descr e
  | (d, None) -> 2 + 17 * hash_slot d
450
451
452
453
454
and hash_slot s =
  if s.gen1 = !gen then 13 * s.rank1
  else (
    incr rank;
    s.rank1 <- !rank; s.gen1 <- !gen;
455
    hash_descr s.d
456
457
458
459
  )
    
let rec equal_descr d1 d2 = 
  match (d1,d2) with
460
  | IType x1, IType x2 -> Types.equal x1 x2
461
462
463
464
465
466
467
  | 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)
468
  | IRecord (o1,r1), IRecord (o2,r2) -> 
469
      (o1 = o2) && (LabelMap.equal equal_descr_field r1 r2)
470
  | ICapture x1, ICapture x2 -> Id.equal x1 x2
471
  | IConstant (x1,y1), IConstant (x2,y2) -> 
472
      (Id.equal x1 x2) && (Types.Const.equal y1 y2)
473
  | _ -> false
474
475
476
477
and equal_descr_field d1 d2 = match (d1,d2) with
  | (d1,None),(d2,None) -> equal_slot d1 d2
  | (d1, Some e1), (d2, Some e2) -> equal_slot d1 d2 && equal_descr e1 e2
  | _ -> false
478
479
480
481
482
483
484
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;
485
     equal_descr s1.d s2.d
486
487
   ))
  
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
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)


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
let pempty = PType Types.empty

let por p1 p2 =
  if p1 == pempty then p2 else
    if p2 == pempty then p1 else
      POr (p1,p2)

let pand p1 p2 =
  if (p1 == pempty) || (p2 == pempty) then pempty else PAnd (p1,p2)

let rec remove_regexp r q = match r with
  | PEpsilon ->
      q
  | PElem p ->
      PTimes (p, q)
  | PGuard p ->
      pand p q
  | PSeq (r1,r2) ->
      remove_regexp r1 (remove_regexp r2 q)
  | PAlt (r1,r2) ->
      por (remove_regexp r1 q) (remove_regexp r2 q)
  | PStar r ->
      let x = mk_derecurs_slot noloc in
      let res = POr (PAlias x, q) in
      x.pdescr <- remove_regexp2 r res pempty;
      res
  | PWeakStar r ->
      let x = mk_derecurs_slot noloc in
      let res = POr (q, PAlias x) in
      x.pdescr <- remove_regexp2 r res pempty;
      res

and remove_regexp2 r q_nonempty q_empty =
  if q_nonempty == q_empty then remove_regexp r q_empty
  else match r with
    | PEpsilon ->
        q_empty
    | PElem p ->
        PTimes (p, q_nonempty)
    | PGuard p ->
	pand p q_empty
    | PSeq (r1,r2) ->
        remove_regexp2 r1
        (remove_regexp2 r2 q_nonempty q_nonempty)
        (remove_regexp2 r2 q_nonempty q_empty)
    | PAlt (r1,r2) ->
        por
        (remove_regexp2 r1 q_nonempty q_empty)
        (remove_regexp2 r2 q_nonempty q_empty)
    | PStar r ->
        let x = mk_derecurs_slot noloc in
        x.pdescr <- remove_regexp2 r (POr (PAlias x, q_nonempty)) pempty;
        por (PAlias x) q_empty
    | PWeakStar r ->
        let x = mk_derecurs_slot noloc in
        x.pdescr <- remove_regexp2 r (POr (q_nonempty, PAlias x)) pempty;
        por q_empty (PAlias x)

568
569
570
571
572
573
let cst_nil = Types.Atom Sequence.nil_atom
let capture_all vars p = IdSet.fold (fun p x -> PAnd (p, PCapture x)) p vars
let termin b vars p = 
  if b then p 
  else IdSet.fold (fun p x -> PSeq (p, PGuard (PConstant (x,cst_nil)))) p vars

574
let rec derecurs env p = match p.descr with
575
  | PatVar v -> derecurs_var env p.loc v
576
  | SchemaVar (kind, schema_name, component_name) ->
577
      PType (find_schema_descr env.penv_tenv kind schema_name component_name)
578
579
580
581
582
583
584
585
586
587
  | 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)
588
589
590
591
592
  | Record (o,r) -> 
      let aux = function
	| (p,Some e) -> (derecurs env p, Some (derecurs env e))
	| (p,None) -> derecurs env p, None in
      PRecord (o, parse_record env.penv_tenv p.loc aux r)
593
594
  | Constant (x,c) -> PConstant (x,const env.penv_tenv p.loc c)
  | Cst c -> PType (Types.constant (const env.penv_tenv p.loc c))
595
596
597
  | Regexp r ->
      let r,_ = derecurs_regexp IdSet.empty false IdSet.empty true env r in
      PRegexp r
598
599
600
	(* Note: computing remove_regexp here is slower (because
	   of caching ?) *)

601
and derecurs_regexp vars b rvars f env = function
602
603
604
605
(* - vars: seq variables to be propagated top-down and added
     to each captured element
   - b: below a star ?
   - rvars: seq variables that appear on the right of the regexp
606
   - f: tail position
607
608

  returns the set of seq variable of the regexp minus rvars
609
  (they have already been terminated if not below a star)
610
*)
611
  | Epsilon -> 
612
      PEpsilon, IdSet.empty
613
  | Elem p -> 
614
      PElem (capture_all vars (derecurs env p)), IdSet.empty
615
  | Guard p ->
616
      PGuard (derecurs env p), IdSet.empty
617
  | Seq (p1,p2) -> 
618
619
      let (p2,v2) = derecurs_regexp vars b rvars f env p2 in
      let (p1,v1) = derecurs_regexp vars b (IdSet.cup rvars v2) false env p1 in
620
      PSeq (p1,p2), IdSet.cup v1 v2
621
  | Alt (p1,p2) -> 
622
623
      let (p1,v1) = derecurs_regexp vars b rvars f env p1
      and (p2,v2) = derecurs_regexp vars b rvars f env p2 in
624
625
      PAlt (termin b (IdSet.diff v2 v1) p1, termin b (IdSet.diff v1 v2) p2),
      IdSet.cup v1 v2
626
  | Star p -> 
627
      let (p,v) = derecurs_regexp vars true rvars false env p in
628
      termin b v (PStar p), v
629
  | WeakStar p -> 
630
      let (p,v) = derecurs_regexp vars true rvars false env p in
631
      termin b v (PWeakStar p), v
632
  | SeqCapture (x,p) -> 
633
      let vars = if f then vars else IdSet.add x vars in
634
635
      let after = IdSet.mem rvars x in
      let rvars = IdSet.add x rvars in
636
637
638
639
      let (p,v) = derecurs_regexp vars b rvars false env p in
      (if f 
       then PSeq (PGuard (PCapture x), p) 
       else termin (after || b) (IdSet.singleton x) p), 
640
641
      (if after then v else IdSet.add x v)

642

643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
and derecurs_var env loc v =
  match Ns.split_qname v with
    | "", v ->
	let v = ident v in
	(try PAlias (Env.find v env.penv_derec)
	 with Not_found -> 
	   try PType (find_type v env.penv_tenv)
	   with Not_found -> PCapture v)
    | cu, v -> 
	try 
	  let cu = U.mk cu in
	  PType (find_type_global loc cu (ident v) env.penv_tenv)
	with Not_found ->
	  raise_loc_generic loc 
	  ("Unbound external type " ^ cu ^ ":" ^ (U.to_string v))

659
660
661
662
663
664
665
666
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

667

668
669
670
671
672
let rec fv_slot s =
  match s.fv with
    | Some x -> x
    | None ->
	if s.gen1 = !gen then IdSet.empty 
673
	else (s.gen1 <- !gen; fv_descr s.d)
674
and fv_descr = function
675
  | IDummy -> assert false
676
  | IType _ -> IdSet.empty
677
678
679
680
681
682
683
  | 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)
684
  | IRecord (o,r) -> 
685
      List.fold_left IdSet.cup IdSet.empty (LabelMap.map_to_list fv_field r)
686
  | ICapture x | IConstant (x,_) -> IdSet.singleton x
687
688
689
690
and fv_field = function
  | (d,Some e) -> IdSet.cup (fv_slot d) (fv_descr e)
  | (d,None) -> fv_slot d

691

692
693
694
695
696
697
698
699
let compute_fv s =
  match s.fv with
    | Some x -> ()
    | None ->
	incr gen;
	let x = fv_slot s in
	s.fv <- Some x
	  
700
701
702
let check_no_capture loc s =
  match IdSet.pick s with
    | Some x ->  
703
	raise_loc_generic loc ("Capture variable not allowed: " ^ (Ident.to_string x))
704
    | None -> ()
705
    
706
707
708
let compile_slot_hash = DerecursTable.create 67
let compile_hash = DerecursTable.create 67

709
710
let todo_defs = ref []
let todo_fv = ref []
711
712
713
714
715
716
717
718

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
719
  | PDummy -> assert false
720
721
722
723
  | PAlias v ->
      if v.ploop then
	raise_loc_generic v.ploc ("Unguarded recursion on type/pattern");
      v.ploop <- true;
724
      let r = compile v.pdescr in
725
726
727
728
729
730
731
732
733
734
      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)
735
  | PRecord (o,r) ->  IRecord (o, LabelMap.map compile_field r)
736
737
  | PConstant (x,v) -> IConstant (x,v)
  | PCapture x -> ICapture x
738
  | PRegexp r -> compile (remove_regexp r (PType Sequence.nil_type))
739

740
741
742
743
and compile_field = function
  | (p, Some e) -> (compile_slot p, Some (compile e))
  | (p, None) -> (compile_slot p, None)

744
745
and compile_slot p =
  try DerecursTable.find compile_slot_hash p
746
  with Not_found ->
747
748
749
    let s = mk_slot () in
    todo_defs := (s,p) :: !todo_defs;
    todo_fv := s :: !todo_fv;
750
    DerecursTable.add compile_slot_hash p s;
751
    s
752

753
      
754
let timer_fv = Stats.Timer.create "Typer.fv"
755
let rec flush_defs () = 
756
757
758
759
  match !todo_defs with
    | [] -> 
	Stats.Timer.start timer_fv;
	List.iter compute_fv !todo_fv;
760
761
	todo_fv := [];
	Stats.Timer.stop timer_fv ()
762
763
764
765
    | (s,p)::t -> 
	todo_defs := t; 
	s.d <- compile p; 
	flush_defs ()
766
767
768
769
770
771
772
773
774
775
776
777
778
	
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)
779
  | IRecord (o,r) ->  Types.record' (o, LabelMap.map typ_field r)
780
  | IDummy | ICapture _ | IConstant (_,_) -> assert false
781
      
782
783
784
785
786
and typ_field = function
  | (s, None) -> typ_node s
  | (s, Some _) -> 
      raise (Patterns.Error "Or-else clauses are not allowed in types")

787
and typ_node s : Types.Node.t =
788
789
790
791
  try SlotTable.find typ_nodes s
  with Not_found ->
    let x = Types.make () in
    SlotTable.add typ_nodes s x;
792
    Types.define x (typ s.d);
793
794
795
796
797
798
799
800
    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
801
  | IDummy -> assert false
802
803
804
805
806
807
  | 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 _ ->
808
      raise (Patterns.Error "Differences are not allowed in patterns")
809
810
811
  | ITimes (s1,s2) -> Patterns.times (pat_node s1) (pat_node s2)
  | IXml (s1,s2) -> Patterns.xml (pat_node s1) (pat_node s2)
  | IOptional _ -> 
812
      raise (Patterns.Error "Optional fields are not allowed in record patterns")
813
814
  | IRecord (o,r) ->
      let pats = ref [] in
815
816
817
818
819
820
821
822
823
824
825
826
827
828
      let aux l = function
	| (s,None) ->
	    if IdSet.is_empty (fv_slot s) then typ_node s
	    else
	      ( pats := Patterns.record l (pat_node s) :: !pats;
		Types.any_node )
	| (s,Some e) ->
	    if IdSet.is_empty (fv_slot s) then
	      raise (Patterns.Error "Or-else clauses are not allowed in types")
	    else
	      ( pats := Patterns.cup 
		  (Patterns.record l (pat_node s))
		  (pat e) :: !pats;
		Types.Record.any_or_absent_node )
829
830
831
832
833
834
835
      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 _ ->
836
      raise (Patterns.Error "Arrows are not allowed in patterns")
837
838
839
840
841
842
  | 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
843
844
    try
      SlotTable.add pat_nodes s x;
845
      Patterns.define x (pat s.d);
846
847
848
      x
    with exn -> SlotTable.remove pat_nodes s; raise exn
      (* For the toplevel ... *)
849

850

851
module Ids = Set.Make(Id)
852
let type_defs env b =
853
854
855
856
857
858
859
860
861
862
  ignore 
    (List.fold_left 
       (fun seen (v,p) ->
	  if Ids.mem v seen then 
	    raise_loc_generic p.loc 
	      ("Multiple definitions for the type identifer " ^ 
	       (Ident.to_string v));
	  Ids.add v seen
       ) Ids.empty b);

863
864
  let penv = derecurs_def (penv env) b in
  let b = List.map (fun (v,p) -> (v,p,compile (derecurs penv p))) b in
865
866
867
868
  flush_defs ();
  let b = 
    List.map 
      (fun (v,p,s) -> 
869
	 check_no_capture p.loc (fv_descr s);
870
871
872
	 let t = typ s in
	 if (p.loc <> noloc) && (Types.is_empty t) then
	   warning p.loc 
873
	     ("This definition yields an empty type for " ^ (Ident.to_string v));
874
	 (v,t)) b in
875
  List.iter (fun (v,t) -> Types.Print.register_global (Id.value v) t) b;
876
  b
877
878


879
880
881
882
883
let dump_types ppf env =
  Env.iter (fun v -> 
	      function 
		  (Type _) -> Format.fprintf ppf " %a" Ident.print v
		| _ -> ()) env.ids
884
885
let dump_type ppf env name =
  try
886
    (match Env.find (Ident.ident name) env.ids with
887
888
    | Type t -> Types.Print.print ppf t
    | _ -> raise Not_found)
889
890
  with Not_found ->
    raise (Error (Printf.sprintf "Type %s not found" (U.get_str name)))
891

892
893
894
let dump_schema_type ppf env (k, s, n) =
  let uri = find_schema s env in
  let descr = find_schema_descr_uri k uri n in
895
  Types.Print.print ppf descr
896

897
let dump_ns ppf env =
898
  Ns.dump_table ppf env.ns
899

900

901
902
let do_typ loc r = 
  let s = compile_slot r in
903
  flush_defs ();
904
905
  check_no_capture loc (fv_slot s);
  typ_node s
906
   
907
908
let typ env p =
  do_typ p.loc (derecurs (penv env) p)
909
    
910
911
let pat env p = 
  let s = compile_slot (derecurs (penv env) p) in
912
913
914
  flush_defs ();
  try pat_node s
  with Patterns.Error e -> raise_loc_generic p.loc e
915
    | Location (loc,_,exn) when loc = noloc -> raise (Location (p.loc, `Full, exn))
916
917


918
919
(* II. Build skeleton *)

920

921
type type_fun = Types.t -> bool -> Types.t
922

923
module Fv = IdSet
924

925
926
927
type branch = Branch of Typed.branch * branch list

let cur_branch : branch list ref = ref []
928

929
let exp loc fv e =
930
931
  fv,
  { Typed.exp_loc = loc;
932
    Typed.exp_typ = Types.empty;
933
    Typed.exp_descr = e;
934
  }
935

936
let ops = Hashtbl.create 13
937
938
let register_op op arity f = Hashtbl.add ops op (arity,f)
let typ_op op = snd (Hashtbl.find ops op)
939

940
941
942
943
944
let is_op env s = 
  if (Env.mem (ident s) env.ids) then None
  else 
    try let s = U.get_str s in Some (s, fst (Hashtbl.find ops s))
    with Not_found -> None
945

946
947
let rec expr env loc = function
  | LocatedExpr (loc,e) -> expr env loc e
948
  | Forget (e,t) ->
949
      let (fv,e) = expr env loc e and t = typ env t in
950
      exp loc fv (Typed.Forget (e,t))
951
  | Var s -> var env loc s
952
  | Apply (e1,e2) -> 
953
954
955
956
957
958
959
960
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
      let fv = Fv.cup fv1 fv2 in
      (match e1.Typed.exp_descr with
	 | Typed.Op (op,arity,args) when arity > 0 -> 
	     exp loc fv (Typed.Op (op,arity - 1,args @ [e2]))
	 | _ ->
	     exp loc fv (Typed.Apply (e1,e2)))
  | Abstraction a -> abstraction env loc a
961
  | (Integer _ | Char _ | Atom _ | Const _) as c -> 
962
      exp loc Fv.empty (Typed.Cst (const env loc c))
963
  | Pair (e1,e2) ->
964
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
965
966
      exp loc (Fv.cup fv1 fv2) (Typed.Pair (e1,e2))
  | Xml (e1,e2) ->
967
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
968
969
      exp loc (Fv.cup fv1 fv2) (Typed.Xml (e1,e2))
  | Dot (e,l) ->
970
971
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.Dot (e,parse_label env loc l))
972
  | RemoveField (e,l) ->
973
974
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.RemoveField (e,parse_label env loc l))
975
976
  | RecordLitt r -> 
      let fv = ref Fv.empty in
977
      let r = parse_record env loc
978
		(fun e -> 
979
		   let (fv2,e) = expr env loc e