typer.ml 43.2 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
let warning loc msg =
12
  Format.fprintf !Location.warning_ppf "Warning %a:@\n%a%s@." 
13
14
    Location.print_loc (loc,`Full)
    Location.html_hilight (loc,`Full)
15
16
    msg

17
18
type item =
  | Type of Types.t
19
  | Val of Types.t
20

21
type t = {
22
  ids : item Env.t;
23
24
  tenv_nspref: Ns.table;
}
25

26
27
28
29
30
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"
31
32

(* TODO: filter out builtin defs ? *)
33
34
35
36
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

37
let serialize s env =
38
  Serialize.Put.env Id.serialize serialize_item Env.iter s env.ids;
39
40
  Ns.serialize_table s env.tenv_nspref

41
42
43
44
45
let deserialize_item s = match Serialize.Get.bits 1 s with
  | 0 -> Type (Types.deserialize s)
  | 1 -> Val (Types.deserialize s)
  | _ -> assert false

46
let deserialize s =
47
48
  let ids = 
    Serialize.Get.env Id.deserialize deserialize_item Env.add Env.empty s in
49
50
51
52
  let ns = Ns.deserialize_table s in
  { ids = ids; tenv_nspref = ns }


53
54
55
56
57
58
59
60
61
62
63
64
65
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
66
    | Val _ -> raise Not_found
67
68

let enter_value id t env = 
69
  { env with ids = Env.add id (Val t) env.ids }
70
71
let enter_values l env =
  { env with ids = 
72
      List.fold_left (fun accu (id,t) -> Env.add id (Val t) accu) env.ids l }
73
74
let find_value id env =
  match Env.find id env.ids with
75
    | Val t -> t
76
77
    | _ -> raise Not_found
	
78
79
80
81
82
83
let value_name_ok id env =
  try match Env.find id env.ids with
    | Val t -> true
    | _ -> false
  with Not_found -> true

84
85
86
87
let iter_values env f =
  Env.iter (fun x ->
	      function Val t -> f x t;
		| _ -> ()) env.ids
88

89
(* Namespaces *)
90

91
92
93
let set_ns_table_for_printer env = 
  Ns.InternalPrinter.set_table env.tenv_nspref

94
let get_ns_table tenv = tenv.tenv_nspref
95

96
97
98
let enter_ns p ns env =
  { env with tenv_nspref = Ns.add_prefix p ns env.tenv_nspref }

99
100
101
102
103
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))
104

105
106
107
108
109
110
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
111

112
113
114
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)
115

116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
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 *)
133

134
exception NonExhaustive of Types.descr
135
exception Constraint of Types.descr * Types.descr
136
exception ShouldHave of Types.descr * string
137
exception ShouldHave2 of Types.descr * string * Types.descr
138
exception WrongLabel of Types.descr * label
139
exception UnboundId of id * bool
140
exception Error of string
141

142
143
let raise_loc loc exn = raise (Location (loc,`Full,exn))
let raise_loc_str loc ofs exn = raise (Location (loc,`Char ofs,exn))
144
let error loc msg = raise_loc loc (Error msg)
145

146
147
148
  (* Schema datastructures *)

module StringSet = Set.Make (String)
149
150
151

  (* just to remember imported schemas *)
let schemas = State.ref "Typer.schemas" StringSet.empty
152
153
154

let schema_types = State.ref "Typer.schema_types" (Hashtbl.create 51)
let schema_elements = State.ref "Typer.schema_elements" (Hashtbl.create 51)
155
let schema_attributes = State.ref "Typer.schema_attributes" (Hashtbl.create 51)
156

157
158
159
160
161
162
163
164
(* 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

165
166
167
168
169
170
171
172
173
174
175
176
177
178
(* 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
*)     

179
180
181
182
type derecurs_slot = {
  ploc : Location.loc;
  pid  : int;
  mutable ploop : bool;
183
  mutable pdescr : derecurs;
184
} and derecurs =
185
  | PDummy
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
  | 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

207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226

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;
227
}
228
229
230
231
232
233
234
235
236
237
238
239
240
    

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 = {
241
  penv_tenv : t;
242
243
244
245
  penv_derec : derecurs_slot Env.t;
}

let penv tenv = { penv_tenv = tenv; penv_derec = Env.empty }
246

247
let rec hash_derecurs = function
248
  | PDummy -> assert false
249
250
251
  | PAlias s -> 
      s.pid
  | PType t -> 
252
      1 + 17 * (Types.hash t)
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
  | 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) -> 
272
      12 + 17 * (Id.hash x) + 257 * (Types.Const.hash c)
273
274
  | PRegexp (p,q) -> 
      13 + 17 * (hash_derecurs_regexp p) + 257 * (hash_derecurs q)
275
and hash_derecurs_regexp = function
276
277
278
279
280
281
282
283
284
285
286
287
  | 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)
288
289

let rec equal_derecurs p1 p2 = (p1 == p2) || match p1,p2 with
290
291
292
  | PAlias s1, PAlias s2 -> 
      s1 == s2
  | PType t1, PType t2 -> 
293
      Types.equal t1 t2
294
295
296
297
298
  | 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)
299
300
301
302
303
304
305
306
307
  | 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) -> 
308
      (Id.equal x1 x2) && (Types.Const.equal c1 c2)
309
310
  | PRegexp (p1,q1), PRegexp (p2,q2) -> 
      (equal_derecurs_regexp p1 p2) && (equal_derecurs q1 q2)
311
312
  | _ -> false
and equal_derecurs_regexp r1 r2 = match r1,r2 with
313
314
315
316
  | PEpsilon, PEpsilon -> 
      true
  | PElem p1, PElem p2 -> 
      equal_derecurs p1 p2
317
  | PSeq (p1,q1), PSeq (p2,q2) 
318
319
  | PAlt (p1,q1), PAlt (p2,q2) -> 
      (equal_derecurs_regexp p1 p2) && (equal_derecurs_regexp q1 q2)
320
  | PStar p1, PStar p2
321
322
  | PWeakStar p1, PWeakStar p2 -> 
      equal_derecurs_regexp p1 p2
323
  | _ -> false
324

325
326
327
328
329
330
331
332
333
334
335
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 
336
337
338
339
    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)
340
341
  end
)
342

343
344
345
346
let gen = ref 0
let rank = ref 0
	     
let rec hash_descr = function
347
  | IDummy -> assert false
348
  | IType x -> Types.hash x
349
350
351
352
353
354
355
356
357
  | 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)
358
  | IConstant (x,y) -> 11 + 17 * (Id.hash x) + 257 * (Types.Const.hash y)
359
360
361
362
363
and hash_slot s =
  if s.gen1 = !gen then 13 * s.rank1
  else (
    incr rank;
    s.rank1 <- !rank; s.gen1 <- !gen;
364
    hash_descr s.d
365
366
367
368
  )
    
let rec equal_descr d1 d2 = 
  match (d1,d2) with
369
  | IType x1, IType x2 -> Types.equal x1 x2
370
371
372
373
374
375
376
  | 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)
377
378
  | IRecord (o1,r1), IRecord (o2,r2) -> 
      (o1 = o2) && (LabelMap.equal equal_slot r1 r2)
379
  | ICapture x1, ICapture x2 -> Id.equal x1 x2
380
  | IConstant (x1,y1), IConstant (x2,y2) -> 
381
      (Id.equal x1 x2) && (Types.Const.equal y1 y2)
382
383
384
385
386
387
388
389
  | _ -> 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;
390
     equal_descr s1.d s2.d
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
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 -> 
419
	 try PType (find_type v env.penv_tenv)
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
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
	 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 ]

    
485
486
487
488
489
let rec fv_slot s =
  match s.fv with
    | Some x -> x
    | None ->
	if s.gen1 = !gen then IdSet.empty 
490
	else (s.gen1 <- !gen; fv_descr s.d)
491
and fv_descr = function
492
  | IDummy -> assert false
493
  | IType _ -> IdSet.empty
494
495
496
497
498
499
500
  | 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)
501
502
  | IRecord (o,r) -> 
      List.fold_left IdSet.cup IdSet.empty (LabelMap.map_to_list fv_slot r)
503
  | ICapture x | IConstant (x,_) -> IdSet.singleton x
504

505
506
507
508
509
510
511
512
let compute_fv s =
  match s.fv with
    | Some x -> ()
    | None ->
	incr gen;
	let x = fv_slot s in
	s.fv <- Some x
	  
513
514
515
let check_no_capture loc s =
  match IdSet.pick s with
    | Some x ->  
516
	raise_loc_generic loc ("Capture variable not allowed: " ^ (Ident.to_string x))
517
    | None -> ()
518
    
519
520
521
let compile_slot_hash = DerecursTable.create 67
let compile_hash = DerecursTable.create 67

522
523
let todo_defs = ref []
let todo_fv = ref []
524
525
526
527
528
529
530
531

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
532
  | PDummy -> assert false
533
534
535
536
  | PAlias v ->
      if v.ploop then
	raise_loc_generic v.ploc ("Unguarded recursion on type/pattern");
      v.ploop <- true;
537
      let r = compile v.pdescr in
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
      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
554
555
556
  let add accu i = 
    match accu with None -> Some i | Some j -> Some (IOr (j,i)) in
  let get = function Some x -> x | None -> assert false in
557
  let rec queue accu = function
558
559
560
    | PRegexp (r,q) -> aux accu r q 
    | _ -> add accu (compile q)
  and aux accu r q =
561
562
563
564
    if RE.mem memo (r,q) then accu
    else (
      RE.add memo (r,q) ();
      match r with
565
	| PEpsilon -> queue accu q
566
567
568
569
570
571
572
573
574
575
576
	| PElem p ->
(* Be careful not to create pairs with same second component *)
	    let rec extract = function
	      | PConstant (x,v) -> `Const (x,v)
	      | POr (x,y) ->
		  (match extract x, extract y with
		    | `Pat x, `Pat y -> `Pat (POr (x,y))
		    | x, y -> `Or (x,y))
	      | p -> `Pat p
	    in
	    let rec mk accu = function
577
578
579
580
	      | `Const (x,v) -> 
		  (match queue None q with 
		    | Some q -> add accu (IAnd (IConstant (x,v), q))
		    | None -> accu)
581
	      | `Or (x,y) -> mk (mk accu x) y
582
583
	      | `Pat p -> 
		  add accu (ITimes (compile_slot p, compile_slot q))
584
585
	    in
	    mk accu (extract p)
586
587
588
589
590
591
	| 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
592
  get (aux None r q)
593
594
and compile_slot p =
  try DerecursTable.find compile_slot_hash p
595
  with Not_found ->
596
597
598
    let s = mk_slot () in
    todo_defs := (s,p) :: !todo_defs;
    todo_fv := s :: !todo_fv;
599
    DerecursTable.add compile_slot_hash p s;
600
    s
601

602
      
603
let timer_fv = Stats.Timer.create "Typer.fv"
604
let rec flush_defs () = 
605
606
607
608
  match !todo_defs with
    | [] -> 
	Stats.Timer.start timer_fv;
	List.iter compute_fv !todo_fv;
609
610
	todo_fv := [];
	Stats.Timer.stop timer_fv ()
611
612
613
614
    | (s,p)::t -> 
	todo_defs := t; 
	s.d <- compile p; 
	flush_defs ()
615
616
617
618
619
620
621
622
623
624
625
626
627
628
	
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)
629
  | IDummy | ICapture _ | IConstant (_,_) -> assert false
630
      
631
and typ_node s : Types.Node.t =
632
633
634
635
  try SlotTable.find typ_nodes s
  with Not_found ->
    let x = Types.make () in
    SlotTable.add typ_nodes s x;
636
    Types.define x (typ s.d);
637
638
639
640
641
642
643
644
    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
645
  | IDummy -> assert false
646
647
648
649
650
651
  | 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 _ ->
652
      raise (Patterns.Error "Differences are not allowed in patterns")
653
654
655
  | ITimes (s1,s2) -> Patterns.times (pat_node s1) (pat_node s2)
  | IXml (s1,s2) -> Patterns.xml (pat_node s1) (pat_node s2)
  | IOptional _ -> 
656
      raise (Patterns.Error "Optional fields are not allowed in record patterns")
657
658
659
660
661
662
663
664
665
666
667
668
669
670
  | 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 _ ->
671
      raise (Patterns.Error "Arrows are not allowed in patterns")
672
673
674
675
676
677
  | 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
678
679
    try
      SlotTable.add pat_nodes s x;
680
      Patterns.define x (pat s.d);
681
682
683
      x
    with exn -> SlotTable.remove pat_nodes s; raise exn
      (* For the toplevel ... *)
684

685

686
let type_defs env b =
687
688
  List.iter 
    (fun (v,p) ->
689
690
       if Env.mem v env.ids
       then raise_loc_generic p.loc ("Identifier " ^ (Ident.to_string v) ^ " is already bound")
691
    ) b;
692
693
  let penv = derecurs_def (penv env) b in
  let b = List.map (fun (v,p) -> (v,p,compile (derecurs penv p))) b in
694
695
696
697
  flush_defs ();
  let b = 
    List.map 
      (fun (v,p,s) -> 
698
	 check_no_capture p.loc (fv_descr s);
699
700
701
	 let t = typ s in
	 if (p.loc <> noloc) && (Types.is_empty t) then
	   warning p.loc 
702
	     ("This definition yields an empty type for " ^ (Ident.to_string v));
703
	 (v,t)) b in
704
  List.iter (fun (v,t) -> Types.Print.register_global (Id.value v) t) b;
705
  b
706
707


708
709
710
711
712
let dump_types ppf env =
  Env.iter (fun v -> 
	      function 
		  (Type _) -> Format.fprintf ppf " %a" Ident.print v
		| _ -> ()) env.ids
713

714
715
let dump_ns ppf env =
  Ns.dump_table ppf env.tenv_nspref
716

717

718
719
let do_typ loc r = 
  let s = compile_slot r in
720
  flush_defs ();
721
722
  check_no_capture loc (fv_slot s);
  typ_node s
723
   
724
725
let typ env p =
  do_typ p.loc (derecurs (penv env) p)
726
    
727
728
let pat env p = 
  let s = compile_slot (derecurs (penv env) p) in
729
730
731
  flush_defs ();
  try pat_node s
  with Patterns.Error e -> raise_loc_generic p.loc e
732
    | Location (loc,_,exn) when loc = noloc -> raise (Location (p.loc, `Full, exn))
733
734


735
736
(* II. Build skeleton *)

737

738
739
740
741
742
type type_fun = Types.t -> bool -> Types.t
let mk_unary_op = ref (fun _ _ -> assert false)
let typ_unary_op = ref (fun _ _ _ -> assert false)
let mk_binary_op = ref (fun _ _ -> assert false)
let typ_binary_op = ref (fun _ _ _ _ -> assert false)
743
744


745
module Fv = IdSet
746

747
748
749
type branch = Branch of Typed.branch * branch list

let cur_branch : branch list ref = ref []
750

751
let exp loc fv e =
752
753
  fv,
  { Typed.exp_loc = loc;
754
    Typed.exp_typ = Types.empty;
755
    Typed.exp_descr = e;
756
  }
757
758


759
760
let rec expr env loc = function
  | LocatedExpr (loc,e) -> expr env loc e
761
  | Forget (e,t) ->
762
      let (fv,e) = expr env loc e and t = typ env t in
763
764
765
766
      exp loc fv (Typed.Forget (e,t))
  | Var s -> 
      exp loc (Fv.singleton s) (Typed.Var s)
  | Apply (e1,e2) -> 
767
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
768
769
      exp loc (Fv.cup fv1 fv2) (Typed.Apply (e1,e2))
  | Abstraction a ->
770
      let iface = List.map (fun (t1,t2) -> (typ env t1, typ env t2)) 
771
772
773
774
775
776
777
		    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
778
      let (fv0,body) = branches env a.fun_body in
779
780
781
782
783
784
785
786
787
788
789
      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
790
  | (Integer _ | Char _ | Atom _) as c -> 
791
      exp loc Fv.empty (Typed.Cst (const env loc c))
792
  | Pair (e1,e2) ->
793
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
794
795
      exp loc (Fv.cup fv1 fv2) (Typed.Pair (e1,e2))
  | Xml (e1,e2) ->
796
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
797
798
      exp loc (Fv.cup fv1 fv2) (Typed.Xml (e1,e2))
  | Dot (e,l) ->
799
800
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.Dot (e,parse_label env loc l))
801
  | RemoveField (e,l) ->
802
803
      let (fv,e) = expr env loc e in
      exp loc fv (Typed.RemoveField (e,parse_label env loc l))
804
805
  | RecordLitt r -> 
      let fv = ref Fv.empty in
806
      let r = parse_record env loc
807
		(fun e -> 
808
		   let (fv2,e) = expr env loc e 
809
810
811
		   in fv := Fv.cup !fv fv2; e)
		r in
      exp loc !fv (Typed.RecordLitt r)
812
  | String (i,j,s,e) ->
813
      let (fv,e) = expr env loc e in
814
      exp loc fv (Typed.String (i,j,s,e))
815
  | Op (op,le) ->
816
      let (fvs,ltes) = List.split (List.map (expr env loc) le) in
817
      let fv = List.fold_left Fv.cup Fv.empty fvs in
818
      (try
819
820
821
	 (match ltes with
	    | [e] -> exp loc fv (Typed.UnaryOp (!mk_unary_op op env, e))
	    | [e1;e2] -> exp loc fv (Typed.BinaryOp (!mk_binary_op op env, e1,e2))
822
823
824
	    | _ -> assert false)
       with Not_found -> assert false)

825
  | Match (e,b) -> 
826
827
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
828
      exp loc (Fv.cup fv1 fv2) (Typed.Match (e, b))
829
  | Map (e,b) ->
830
831
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
832
833
      exp loc (Fv.cup fv1 fv2) (Typed.Map (e, b))
  | Transform (e,b) ->
834
835
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
836
      exp loc (Fv.cup fv1 fv2) (Typed.Transform (e, b))
837
  | Xtrans (e,b) ->
838
839
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
840
      exp loc (Fv.cup fv1 fv2) (Typed.Xtrans (e, b))
841
  | Validate (e,schema,elt) ->
842
      let (fv,e) = expr env loc e in
843
      exp loc fv (Typed.Validate (e, schema, elt))
844
  | Try (e,b) ->
845
846
      let (fv1,e) = expr env loc e
      and (fv2,b) = branches env b in
847
      exp loc (Fv.cup fv1 fv2) (Typed.Try (e, b))
848
  | NamespaceIn (pr,ns,e) ->
849
850
      let env = enter_ns pr ns env in
      expr env loc e
851
  | Ref (e,t) ->
852
      let (fv,e) = expr env loc e and t = typ env t in
853
      exp loc fv (Typed.Ref (e,t))
854
	      
855
  and branches env b = 
856
    let fv = ref Fv.empty in
857
    let accept = ref Types.empty in
858
    let branch (p,e) = 
859
860
      let cur_br = !cur_branch in
      cur_branch := [];
861
      let (fv2,e) = expr env noloc e in
862
      let br_loc = merge_loc p.loc e.Typed.exp_loc in
863
      let p = pat env p in
864
865
866
867
868
869
      (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 ^ 
870
	       " 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)."));
871
872
873
874
875
876
877
878
879
      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
880
      cur_branch := Branch (br, !cur_branch) :: cur_br;
881
882
      br in
    let b = List.map branch b in
883
884
885
886
    (!fv, 
     { 
       Typed.br_typ = Types.empty; 
       Typed.br_branches = b; 
887
888
       Typed.br_accept = !accept;
       Typed.br_compiled = None;
889
890
     } 
    )
891

892
let expr env e = snd (expr env noloc e)
893

894
895
let let_decl env p e =
  { Typed.let_pat = pat env p;
896
    Typed.let_body = expr env e;
897
898
    Typed.let_compiled = None }

899
900
901

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

902

903
904
(* III. Type-checks *)

905
906
open Typed

907
908
let require loc t s = 
  if not (Types.subtype t s) then raise_loc loc (Constraint (t, s))
909

910
let verify loc t s = 
911
912
  require loc t s; t

913
914
915
916
917
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 = 
918
919
  raise_loc loc (ShouldHave (constr,s))

920
921
922
let should_have_str loc ofs constr s = 
  raise_loc_str loc ofs (ShouldHave (constr,s))

923
924
925
926
927
928
929
930
931
932
933
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
934

935
936
let rec type_check env e constr precise = 
  let d = type_check' e.exp_loc env e.exp_descr constr precise in
937
  let d = if precise then d else constr in
938
939
940
  e.exp_typ <- Types.cup e.exp_typ d;
  d

941
and type_check' loc env e constr precise = match e with
942
943
944
  | Forget (e,t) ->
      let t = Types.descr t in
      ignore (type_check env e t false);
945
      verify loc t constr
946

947
  | Abstraction a ->
948
949
950
      let t =
	try Types.Arrow.check_strenghten a.fun_typ constr 
	with Not_found -> 
951
952
	  should_have loc constr
	    "but the interface of the abstraction is not compatible"
953
      in
954
955
      let env = match a.fun_name with
	| None -> env
956
	| Some f -> enter_value f a.fun_typ env in
957
958
      List.iter 
	(fun (t1,t2) ->
959
960
961
	   let acc = a.fun_body.br_accept in 
	   if not (Types.subtype t1 acc) then
	     raise_loc loc (NonExhaustive (Types.diff t1 acc));
962
	   ignore (type_check_branches loc env t1 a.fun_body t2 false)
963
964
	) a.fun_iface;
      t
965

966
967
  | Match (e,b) ->
      let t = type_check env e b.br_accept true in
968
      type_check_branches loc env t b constr precise
969
970
971

  | Try (e,b) ->
      let te = type_check env e constr precise in
972
      let tb = type_check_branches loc env Types.any b constr precise in
973
      Types.cup te tb
974

975
976
  | Pair (e1,e2) ->
      type_check_pair loc env e1 e2 constr precise
977

978
979
  | Xml (e1,e2) ->
      type_check_pair ~kind:`XML loc env e1 e2 constr precise
980

981
  | RecordLitt r ->
982
983
984
985
986
987
988
989
      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

990
991
992
993
  | 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
994
995
996
997
998
999
1000
      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
1001
      verify loc res constr
1002
1003

  | UnaryOp (o,e) ->
1004
1005
      let t = !typ_unary_op o loc (type_check env e) constr precise in
      verify loc t constr
1006
1007

  | BinaryOp (o,e1,e2) ->
1008
1009
1010
      let t = !typ_binary_op o loc