typer.ml 60.9 KB
Newer Older
1
(* TODO:
2
 - rewrite type-checking of operators to propagate constraint
3
 - check whether it is worth using recursive hash-consing internally
4
5
*)

6
7
8
open Location
open Ast
open Ident
9

10
11
12
13
14
15
let (=) (x:int) y = x = y
let (<=) (x:int) y = x <= y
let (<) (x:int) y = x < y
let (>=) (x:int) y = x >= y
let (>) (x:int) y = x > y

16
17
let debug_schema = false

18
let warning loc msg =
19
  let v = Location.get_viewport () in
20
  let ppf = if Html.is_html v then Html.ppf v else Format.err_formatter in
21
22
23
  Format.fprintf ppf "Warning %a:@\n" Location.print_loc (loc,`Full);
  Location.html_hilight (loc,`Full);
  Format.fprintf ppf "%s@." msg
24

25
26
27
28
29
30
31
32
33
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

34
35
36

exception Warning of string * Types.t

37
38
39
40
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)

41

42
43
type item =
  | Type of Types.t
44
  | Val of Types.t
45

46
47
48
49
50
type ext =
  | ECDuce of Types.CompUnit.t   (* CDuce unit *)
  | EOCaml of string             (* OCaml module *)
  | ESchema of string            (* XML Schema *)

51
52
module UEnv = Map.Make(U)

53
type t = {
54
  ids : item Env.t;
55
  ns: Ns.table;
56
  cu: ext UEnv.t;
57
}
58

59
60
61
62
63
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"
64

65
66
67

let load_schema_fwd = ref (fun x uri -> assert false)

68
let enter_schema x uri env =
69
  let sch = !load_schema_fwd x uri in
70
  { env with cu = UEnv.add x (ESchema uri) env.cu }
71

72
(* TODO: filter out builtin defs ? *)
73
74
75
76
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

77
let serialize s env =
78
  Serialize.Put.env Id.serialize serialize_item Env.iter s env.ids;
79
80
81
  Ns.serialize_table s env.ns;

  let schs =
82
83
84
    UEnv.fold (fun name cu accu -> 
		 match cu with ESchema uri -> (name,uri)::accu | _ -> accu) 
      env.cu [] in
85
  Serialize.Put.list (Serialize.Put.pair U.serialize Serialize.Put.string) s schs
86

87
88
89
90
91
let deserialize_item s = match Serialize.Get.bits 1 s with
  | 0 -> Type (Types.deserialize s)
  | 1 -> Val (Types.deserialize s)
  | _ -> assert false

92
let deserialize s =
93
  let ids = Serialize.Get.env Id.deserialize deserialize_item Env.add Env.empty s in
94
  let ns = Ns.deserialize_table s in
95
96
97
98
  let schs = 
    Serialize.Get.list 
      (Serialize.Get.pair U.deserialize Serialize.Get.string) s in
  let env = 
99
    { ids = ids; ns = ns; cu = UEnv.empty } in
100
  List.fold_left (fun env (name,uri) -> enter_schema name uri env) env schs
101
102


103
104
let empty_env = {
  ids = Env.empty;
105
  ns = Ns.empty_table;
106
  cu = UEnv.empty;
107
108
}

109
110
111
let from_comp_unit = ref (fun (cu : Types.CompUnit.t) -> assert false)
let has_comp_unit = ref (fun cu -> assert false)
let has_ocaml_unit = ref (fun cu -> false)
112
113
let has_static_external = ref (fun _ -> assert false)

114

115
let enter_cu x cu env =
116
  { env with cu = UEnv.add x (ECDuce cu) env.cu }
117

118
let find_cu loc x env =
119
  try UEnv.find x env.cu
120
121
122
123
  with Not_found ->
    if !has_comp_unit x then (ECDuce (Types.CompUnit.mk x))
    else if !has_ocaml_unit x then (EOCaml (U.get_str x))
    else error loc ("Cannot find external unit " ^ (U.to_string x))
124
125


126
let find_schema x env =
127
128
129
130
131
132
  try 
    (match UEnv.find x env.cu with
      | ESchema s -> s 
      | _ -> raise Not_found)
  with Not_found -> 
    raise (Error (Printf.sprintf "%s: no such schema" (U.to_string x)))
133

134
135
136
137
138
139
140
141
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
142
    | Val _ -> raise Not_found
143

144

145
let enter_value id t env = 
146
  { env with ids = Env.add id (Val t) env.ids }
147
148
let enter_values l env =
  { env with ids = 
149
      List.fold_left (fun accu (id,t) -> Env.add id (Val t) accu) env.ids l }
150
151
152
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 }
153
154
let find_value id env =
  match Env.find id env.ids with
155
    | Val t -> t
156
    | _ -> raise Not_found
157
158
159
let find_value_global loc cu id env =
  try find_value id (!from_comp_unit cu)
  with Not_found -> raise_loc loc (UnboundExtId (cu,id))
160
	
161
162
163
164
165
166
let value_name_ok id env =
  try match Env.find id env.ids with
    | Val t -> true
    | _ -> false
  with Not_found -> true

167
168
169
170
let iter_values env f =
  Env.iter (fun x ->
	      function Val t -> f x t;
		| _ -> ()) env.ids
171

172

173
let register_types cu env =
174
175
176
  Env.iter (fun x t -> match t with
	      | Type t -> Types.Print.register_global cu (Ident.value x) t
	      | _ -> ()) env.ids
177

178

179
(* Namespaces *)
180

181
let set_ns_table_for_printer env = 
182
  Ns.InternalPrinter.set_table env.ns
183

184
let get_ns_table tenv = tenv.ns
185

186
let enter_ns p ns env =
187
  { env with ns = Ns.add_prefix p ns env.ns }
188

189
190
191
192
193
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))
194

195
196
197
let qname env loc t = 
  protect_error_ns loc (Ns.map_tag env.ns) t
    
198
199
200
201
202
203
204
205
206
207
let ident env loc t =
  let q = protect_error_ns loc (Ns.map_attr env.ns) t in
  Ident.ident q

let has_value id env =
  try match Env.find (Ident.ident (Ns.map_attr env.ns id)) env.ids with
    | Val t -> true
    | _ -> false
  with Not_found | Ns.UnknownPrefix _ -> false

208
let parse_atom env loc t =
209
  Atoms.V.of_qname (qname env loc t)
210
211
 
let parse_ns env loc ns =
212
  protect_error_ns loc (Ns.map_prefix env.ns) ns
213

214
let parse_label env loc t =
215
  let (ns,l) = protect_error_ns loc (Ns.map_attr env.ns) t in
216
  LabelPool.mk (ns,l)
217

218
219
220
221
222
223
224
225
226
227
228
229
230
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
231
  | Const c -> c
232
233
234
235
  | _ -> 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 *)
236

237

238
(* Schema *)
239

240
(* uri -> schema binding *)
241
let schemas = Hashtbl.create 3
242

243
let find_schema_descr uri (name : Ns.qname) =
244
  try
245
246
247
248
249
    let sch = snd (Hashtbl.find schemas uri) in
    fst (Env.find (Ident.ident name) sch)
  with Not_found ->    
    raise (Error (Printf.sprintf "No component named '%s' found in schema '%s'"
		    (Ns.QName.to_string name) uri))
250
251


252
253
254
255
256
257
let find_type_global loc cu id env =
  match find_cu loc cu env with
    | ECDuce cu -> find_type id (!from_comp_unit cu)
    | EOCaml _ -> error loc "OCaml units don't export types" (* TODO *)
    | ESchema s -> find_schema_descr s (Ident.value id)
	
258

259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
module IType = struct
  type node = {
    mutable desc: desc;
    mutable smallhash: int;  (* Local hash *)
    mutable rechash: int;    (* Global (recursive) hash *)
    mutable sid: int;        (* Sequential id used to compute rechash *)
    mutable t: Types.t option;
    mutable tnode: Types.Node.t option;
    mutable p: Patterns.descr option;
    mutable pnode: Patterns.node option;
    mutable fv: fv option
  } 
  and desc =
    | ILink of node
    | IType of Types.descr * int
    | IOr of node * node
    | IAnd of node * node
    | IDiff of node * node
    | ITimes of node * node
    | IXml of node * node
    | IArrow of node * node
    | IOptional of node
    | IRecord of bool * (node * node option) label_map
    | ICapture of id
    | IConstant of id * Types.const
284
    | IConcat of node * node
285
    | IMerge of node * node
286
287
288
289
290
291
292
293

  let rec node_temp = { 
    desc = ILink node_temp;
    smallhash = 0; rechash = 0; sid = 0;
    t = None; tnode = None; p = None; pnode = None;
    fv = None
  }
			
294
(* Recursive hash-consing *)
295

296
297
298
299
300
301
  let hash_field f = function
    | (p, Some e) -> 1 + 17 * f p + 257 * f e
    | (p, None) -> 2 + 17 * f p

  let rec hash f n = match n.desc with
    | ILink n -> hash f n
302
    | IType (t,h) -> 1 + 17 * h
303
304
305
306
307
308
309
310
311
    | IOr (p1,p2) -> 2 + 17 * f p1 + 257 * f p2
    | IAnd (p1,p2) -> 3 + 17 * f p1 + 257 * f p2
    | IDiff (p1,p2) -> 4 + 17 * f p1 + 257 * f p2
    | ITimes (p1,p2) -> 5 + 17 * f p1 + 257 * f p2
    | IXml (p1,p2) -> 6 + 17 * f p1 + 257 * f p2
    | IArrow (p1,p2) -> 7 + 17 * f p1 + 257 * f p2
    | IOptional p -> 8 + 17 * f p
    | IRecord (o,r)->9+(if o then 17 else 0)+
	257*(LabelMap.hash (hash_field f) r)
312
313
    | ICapture x -> 10 + 17 * (Id.hash x)
    | IConstant (x,c) -> 11 + 17 * (Id.hash x) + 257*(Types.Const.hash c)
314
    | IConcat _ | IMerge _ -> assert false
315

316
317
318
319
320
  let hash0 = hash (fun n -> 1)
  let hash1 = hash hash0
  let hash2 = hash hash1
  let hash3 = hash hash2

321
322
  let smallhash n =
    if n.smallhash !=0 then n.smallhash
323
324
325
326
    else (
      let h = hash2 n in 
      n.smallhash <- h; h
    )
327
328

  let rec repr = function
329
    | { desc = ILink n } as m -> let z = repr n in m.desc <- ILink z; z
330
331
332
333
    | n -> n

  let back = ref []

334
335
336
337
  let rec prot_repr = function
    | { desc = ILink n } -> repr n
    | n -> n

338
339
340
341
342
343
344
345
346
  let link x y = match x,y with
    | { t = None } as x, y 
    | y, ({ t = None } as x) -> back := (x,x.desc) :: !back; x.desc <- ILink y
    | _ -> assert false

  exception Unify

  let rec unify x y =
    if x == y then ()
347
348
349
350
351
    else let x = prot_repr x and y = prot_repr y in if x == y then ()
    else if (smallhash x != smallhash y) then raise Unify 
    else if (x.t != None) && (y.t != None) then raise Unify
      (* x and y have been internalized; if they were equivalent,
	 they would be equal *)
352
    else match x.desc,y.desc with
353
      | IType (tx,_), IType (ty,_) when Types.equal tx ty -> link x y
354
355
356
357
358
      | IOr (x1,x2), IOr (y1,y2)
      | IAnd (x1,x2), IAnd (y1,y2)
      | IDiff (x1,x2), IDiff (y1,y2)
      | ITimes (x1,x2), ITimes (y1,y2)
      | IXml (x1,x2), IXml (y1,y2)
359
360
      | IArrow (x1,x2), IArrow (y1,y2) -> link x y; unify x1 y1; unify x2 y2
      | IOptional x1, IOptional y1 -> link x y; unify x1 y1
361
362
363
364
365
366
367
368
369
370
371
      | IRecord (xo,xr), IRecord (yo,yr) when xo == yo ->
	  link x y; LabelMap.may_collide unify_field Unify xr yr
      | ICapture xv, ICapture yv when Id.equal xv yv -> ()
      | IConstant (xv,xc), IConstant (yv,yc) when
	  Id.equal xv yv && Types.Const.equal xc yc -> ()
      | _ -> raise Unify
  and unify_field f1 f2 = match f1,f2 with
    | (p1, Some e1), (p2, Some e2) -> unify p1 p2; unify e1 e2
    | (p1, None), (p2, None) -> unify p1 p2
    | _ -> raise Unify

372

373
374
  let may_unify x y =
    try unify x y; back := []; true
375
    with Unify ->
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
      List.iter (fun (x,xd) -> x.desc <- xd) !back; back := []; false

  module SmallHash = Hashtbl.Make(
    struct 
      type t = node
      let equal = may_unify
      let hash = smallhash
    end
  )

  let iter_field f = function
    | (x, Some y) -> f x; f y
    | (x, None) -> f x
  let iter f = function
    | IOr (x,y) | IAnd (x,y) | IDiff (x,y)
    | ITimes (x,y) | IXml (x,y) | IArrow (x,y) -> f x; f y
    | IOptional x -> f x
    | IRecord (_,r) -> LabelMap.iter (iter_field f) r
    | _ -> ()

  let minimize ((mem,add) as h) =
    let rec aux n =
      let n = repr n in
399
400
401
402
      if mem n then () else (
	let n = repr n in add n (); 
	if n.t == None then iter aux n.desc
      )
403
404
405
406
407
408
409
    in aux

  let to_clear = ref []
  let sid = ref 0
  let rec rechash n =
    let n = repr n in
    if (n.sid != 0) then 17 * n.sid
410
    else (incr sid; n.sid <- !sid; to_clear := n :: !to_clear; hash rechash n)
411
412

  let clear () =
413
414
    sid := 0; List.iter (fun x -> x.sid <- 0) !to_clear;
    to_clear := []
415
416
417
418
419
420
421
422
423
424
425
426
427
428

  let rechash n =
    let n = repr n in
    if (n.rechash != 0) then n.rechash 
    else (let h = rechash n in clear (); n.rechash <- h; h)

  module RecHash = Hashtbl.Make(
    struct
      type t = node
      let equal = may_unify
      let hash = smallhash
    end
  )

429
430
431

(** Two-phases recursive hash-consing **)
(*
432
433
434
  let gtable = RecHash.create 17577

  let internalize n =
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
    let local = SmallHash.create 17 in
    minimize (SmallHash.mem local, SmallHash.add local) n; 
    minimize (RecHash.mem gtable, RecHash.add gtable) n;
    ()
*)

(** Single-phase hash-consing **)
  let gtable = SmallHash.create 17

  let internalize n =
    minimize (SmallHash.mem gtable, SmallHash.add gtable) n



(*  let internalize n = () *)
450
451
452
453
454
455
456
457
458

(* Compute free variables *)

  let fv n =
    let fv = ref IdSet.empty in
    let rec aux n =
      let n = repr n in
      if (n.sid = 0) then (
	n.sid <- 1;
459
	to_clear := n :: !to_clear; 
460
461
462
463
464
465
	match n.fv, n.desc with
	  | Some x, _ -> fv := IdSet.cup !fv x
	  | None, (ICapture x | IConstant (x,_)) -> fv := IdSet.add x !fv
	  | None, d -> iter aux d
      )
    in
466
    assert(!to_clear == []);
467
468
469
470
    match n.fv with
      | Some x -> x
      | None -> aux n; clear (); n.fv <- Some !fv; !fv

471
472
473
(* optimized version to check closedness *)

  let no_fv = Some IdSet.empty
474
475
476
  exception FoundFv of id
  let peek_fv n =
    let err x = raise (FoundFv x) in
477
478
479
480
481
482
    let rec aux n =
      let n = repr n in
      if (n.sid = 0) then (
	n.sid <- 1;
	to_clear := n :: !to_clear; 
	match n.fv, n.desc with
483
484
	  | Some x, _ when IdSet.is_empty x -> ()
	  | Some x, _ -> err (IdSet.choose x)
485
486
487
488
	  | None, (ICapture x | IConstant (x,_)) -> err x;
	  | None, d -> iter aux d
      )
    in
489
    assert(!to_clear == []);
490
491
    try
      match n.fv with
492
493
	| Some x when IdSet.is_empty x -> ()
	| Some x -> err (IdSet.choose x)
494
495
496
497
498
	| None -> aux n; 
	    List.iter (fun n -> n.sid <- 0; n.fv <- no_fv) !to_clear;
	    to_clear := []
    with exn -> clear (); raise exn

499
500
501
502
503
504
505
506
507
508
509
  let check_no_fv loc n =
    try peek_fv n 
    with FoundFv x ->
      raise_loc_generic loc 
	("Capture variable not allowed: " ^ (Ident.to_string x))

  let has_no_fv n =
    try peek_fv n; true
    with FoundFv _ -> false


510
(* From the intermediate representation to the internal one *)
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529


  let rec typ n =
    let n = repr n in
    match n.t with
      | Some t -> t
      | None -> let t = compute_typ n.desc in n.t <- Some t; t
  and compute_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 compute_typ_field r)
    | ILink _ -> assert false
    | ICapture _ | IConstant (_,_) -> assert false
530
    | IConcat _ | IMerge _ -> assert false
531
532
533
534
535
536
  and compute_typ_field = function
    | (s, None) -> typ_node s
    | (s, Some _) -> 
	raise (Patterns.Error "Or-else clauses are not allowed in types")

  and typ_node n =
537
    let n = repr n in
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
    match n.tnode with
      | Some t -> t
      | None ->
	  let x = Types.make () in
	  n.tnode <- Some x;
	  Types.define x (typ n);
	  x
      
  let rec pat n =
    let n = repr n in
    if IdSet.is_empty (fv n)
    then Patterns.constr (typ n)
    else match n.p with
      | Some p -> p
      | None -> let p = compute_pat n.desc in n.p <- Some p; p

  and compute_pat = function
    | IOr (s1,s2) -> Patterns.cup (pat s1) (pat s2)
    | IAnd (s1,s2) -> Patterns.cap (pat s1) (pat s2)
    | IDiff (s1,s2) when IdSet.is_empty (fv s2) ->
	let s2 = Types.neg (typ s2) in
	Patterns.cap (pat s1) (Patterns.constr s2)
    | IDiff _ ->
	raise (Patterns.Error "Differences are not allowed in patterns")
    | ITimes (s1,s2) -> Patterns.times (pat_node s1) (pat_node s2)
    | IXml (s1,s2) -> Patterns.xml (pat_node s1) (pat_node s2)
    | IOptional _ -> 
	raise (Patterns.Error "Optional fields are not allowed in record patterns")
    | IRecord (o,r) ->
	let pats = ref [] in
	let aux l = function
	  | (s,None) ->
	      if IdSet.is_empty (fv 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 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 )
	in
	let constr = Types.record' (o,LabelMap.mapi aux r) in
	List.fold_left Patterns.cap (Patterns.constr constr) !pats
	  (* TODO: can avoid constr when o=true, and all fields have fv *)
    | ICapture x -> Patterns.capture x
    | IConstant (x,c) -> Patterns.constant x c
    | IArrow _ ->
	raise (Patterns.Error "Arrows are not allowed in patterns")
590
    | IType _ | ILink _ | IConcat _ | IMerge _ -> assert false
591
592
      
  and pat_node n =
593
    let n = repr n in
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
    match n.pnode with
      | Some p -> p
      | None ->
	  let x = Patterns.make (fv n) in
	  try
	    n.pnode <- Some x;
	    Patterns.define x (pat n);
	    x
	  with exn -> n.pnode <- None; raise exn

(* From AST to the intermediate representation *)

  type penv = {
    penv_tenv : t;
    penv_derec : node Env.t;
  }

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

613
614
  let concats = ref []

615
616
617
618
619
620
  let mk d = { node_temp with desc = d }
  let mk_delayed () = { node_temp with desc = ILink node_temp }
  let itype t = mk (IType (t, Types.hash t))
  let iempty = itype Types.empty

  let ior p1 p2 =
621
622
    if p1.desc == iempty.desc then p2 
    else if p2.desc == iempty.desc then p1 
623
624
625
    else mk (IOr (p1,p2))

  let iand p1 p2 =
626
    if (p1.desc == iempty.desc) || (p2.desc == iempty.desc) then iempty 
627
628
629
630
631
    else mk (IAnd (p1,p2))

  type regexp =
    | PElem of node
    | PGuard of node
632
633
    | PSeq of regexp list
    | PAlt of regexp list
634
635
636
    | PStar of regexp
    | PWeakStar of regexp

637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
  let rec nullable = function
    | PElem _ -> false
    | PSeq rl -> List.for_all nullable rl
    | PAlt rl -> List.exists nullable rl
    | PStar _ | PWeakStar _ | PGuard _ -> true

  let eps = PSeq []
  let emp = PAlt []

  let seq r1 r2 =
    let r1 = match r1 with PSeq l -> l | x -> [ x ] in
    let r2 = match r2 with PSeq l -> l | x -> [ x ] in
    match r1 @ r2 with
      | [ x ] -> x
      | l -> PSeq l

  let alt r1 r2 =
    let r1 = match r1 with PAlt l -> l | x -> [ x ] in
    let r2 = match r2 with PAlt l -> l | x -> [ x ] in
    match r1 @ r2 with
      | [ x ] -> x
      | l -> PAlt l

  let rec merge_alt = function
661
    | PElem p::PElem q::l -> merge_alt (PElem (ior p q) :: l)
662
663
    | r::l -> r::(merge_alt l)
    | [] -> []
664
665
666
667
668
669
670
671
672

(* Works only for types, not patterns, because
   [ (x&Int|_) R' ] is possible *)
  let rec simplify_regexp = function
    | PSeq l -> PSeq (List.map simplify_regexp l)
    | PAlt l -> PAlt (merge_alt (List.map simplify_regexp l))
    | PStar r | PWeakStar r -> PStar (simplify_regexp r)
    | x -> x

673
674
675
676
677
678
679
680
681
  let rec print_regexp ppf = function
    | PElem _ -> Format.fprintf ppf "Elem"
    | PGuard _ -> Format.fprintf ppf "Guard"
    | PSeq l -> Format.fprintf ppf "Seq(%a)" print_regexp_list l
    | PAlt l -> Format.fprintf ppf "Alt(%a)" print_regexp_list l
    | PStar r -> Format.fprintf ppf "Star(%a)" print_regexp r
    | PWeakStar r -> Format.fprintf ppf "WStar(%a)" print_regexp r
  and print_regexp_list ppf l =
    List.iter (fun x -> Format.fprintf ppf "%a;" print_regexp x) l
682

683
684
  let rec remove_regexp r q = 
    match r with
685
686
687
688
    | PElem p ->
	mk (ITimes (p, q))
    | PGuard p ->
	iand p q
689
690
691
692
    | PSeq l ->
	List.fold_right (fun r a -> remove_regexp r a) l q
    | PAlt rl ->
	List.fold_left (fun a r -> ior a (remove_regexp r q)) iempty rl
693
694
695
    | PStar r ->
	let x = mk_delayed () in
	let res = ior x q in
696
	x.desc <- ILink (remove_regexp_nullable r res iempty);
697
698
699
700
	res
    | PWeakStar r ->
	let x = mk_delayed () in
	let res = ior q x in
701
	x.desc <- ILink (remove_regexp_nullable r res iempty);
702
	res
703
704
705
706
707

  and remove_regexp_nullable r q_nonempty q_empty =
    if nullable r then remove_regexp2 r q_nonempty q_empty
    else remove_regexp r q_nonempty

708
  and remove_regexp2 r q_nonempty q_empty =
709
710
    (* Assume r is nullable *)
    if q_nonempty == q_empty then remove_regexp r q_nonempty
711
    else match r with
712
      | PSeq [] ->
713
714
          q_empty
      | PElem p ->
715
	  assert false
716
717
      | PGuard p ->
	  iand p q_empty
718
719
720
721
722
723
724
725
      | PSeq (r::rl) ->
          remove_regexp2 r
            (remove_regexp (PSeq rl) q_nonempty)
            (remove_regexp2 (PSeq rl) q_nonempty q_empty)
      | PAlt rl ->
	  List.fold_left 
	    (fun a r -> ior a (remove_regexp_nullable r q_nonempty q_empty))
	    iempty rl
726
727
      | PStar r ->
 	  let x = mk_delayed () in
728
          x.desc <- ILink (remove_regexp_nullable r (ior x q_nonempty) iempty);
729
730
731
          ior x q_empty
      | PWeakStar r ->
 	  let x = mk_delayed () in
732
          x.desc <- ILink (remove_regexp_nullable r (ior q_nonempty x) iempty);
733
734
735
736
737
738
739
740
741
          ior q_empty x


  let cst_nil = Types.Atom Sequence.nil_atom
  let capture_all vars p = 
    IdSet.fold (fun p x -> iand p (mk (ICapture x))) p vars
  let termin b vars p = 
    if b then p 
    else IdSet.fold 
742
      (fun p x -> seq p (PGuard (mk (IConstant (x,cst_nil))))) p vars
743
744
745

  let rexp r = remove_regexp r (itype Sequence.nil_type)

746
747
  let all_delayed = ref []

748
749
  let clean_on_err () = all_delayed := []

750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
  let delayed loc =
    let s = mk_delayed () in
    all_delayed := (loc,s) :: !all_delayed;
    s

  let check_one_delayed (loc,p) =
    let rec aux q = if p == q then raise Exit; aux2 q.desc
    and aux2 = function
      | IOr (q1,q2) | IAnd (q1,q2) | IDiff (q1,q2) -> aux q1; aux q2
      | ILink q -> aux q
      | _ -> ()
    in
    try aux2 p.desc
    with Exit -> error loc "Ill-formed recursion"
    
  let check_delayed () =
    let l = !all_delayed in
    all_delayed := []; 
    List.iter check_one_delayed l
769

770
    
771
  let rec derecurs env p = match p.descr with
772
    | PatVar (cu,v) -> derecurs_var env p.loc cu v
773
    | Recurs (p,b) -> derecurs (fst (derecurs_def env b)) p
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
    | Internal t -> itype t
    | NsT ns -> 
	itype (Types.atom (Atoms.any_in_ns (parse_ns env.penv_tenv p.loc ns)))
    | Or (p1,p2) -> mk (IOr (derecurs env p1, derecurs env p2))
    | And (p1,p2) -> mk (IAnd (derecurs env p1, derecurs env p2))
    | Diff (p1,p2) -> mk (IDiff (derecurs env p1, derecurs env p2))
    | Prod (p1,p2) -> mk (ITimes (derecurs env p1, derecurs env p2))
    | XmlT (p1,p2) -> mk (IXml (derecurs env p1, derecurs env p2))
    | Arrow (p1,p2) -> mk (IArrow (derecurs env p1, derecurs env p2))
    | Optional p -> mk (IOptional (derecurs env p))
    | Record (o,r) -> 
	let aux = function
	  | (p,Some e) -> (derecurs env p, Some (derecurs env e))
	  | (p,None) -> derecurs env p, None in
	mk (IRecord (o, parse_record env.penv_tenv p.loc aux r))
789
790
    | Constant (x,c) -> mk (IConstant (ident env.penv_tenv p.loc x,
				       const env.penv_tenv p.loc c))
791
792
793
794
    | Cst c -> itype (Types.constant (const env.penv_tenv p.loc c))
    | Regexp r ->
	let r,_ = derecurs_regexp IdSet.empty false IdSet.empty true env r in
	rexp r
795
796
797
798
    | Concat (p1,p2) -> 
	let n = mk (IConcat (derecurs env p1, derecurs env p2)) in
	concats := n :: !concats;
	n
799
800
801
802
    | Merge (p1,p2) -> 
	let n = mk (IMerge (derecurs env p1, derecurs env p2)) in
	concats := n :: !concats;
	n
803
804
805
806
807
808
809
810
811
812
813
814
	  
  and derecurs_regexp vars b rvars f env = function
      (* - 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
	 - f: tail position
	 
	 returns the set of seq variable of the regexp minus rvars
	 (they have already been terminated if not below a star)
      *)
    | Epsilon -> 
815
	PSeq [], IdSet.empty
816
817
818
819
820
821
822
    | Elem p -> 
	PElem (capture_all vars (derecurs env p)), IdSet.empty
    | Guard p ->
	PGuard (derecurs env p), IdSet.empty
    | Seq (p1,p2) -> 
	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
823
	seq p1 p2, IdSet.cup v1 v2
824
825
826
    | Alt (p1,p2) -> 
	let (p1,v1) = derecurs_regexp vars b rvars f env p1
	and (p2,v2) = derecurs_regexp vars b rvars f env p2 in
827
	alt (termin b (IdSet.diff v2 v1) p1) (termin b (IdSet.diff v1 v2) p2),
828
829
830
831
832
833
834
	IdSet.cup v1 v2
    | Star p -> 
	let (p,v) = derecurs_regexp vars true rvars false env p in
	termin b v (PStar p), v
    | WeakStar p -> 
	let (p,v) = derecurs_regexp vars true rvars false env p in
	termin b v (PWeakStar p), v
835
836
    | SeqCapture (loc,x,p) -> 
	let x = ident env.penv_tenv loc x in
837
838
839
840
841
	let vars = if f then vars else IdSet.add x vars in
	let after = IdSet.mem rvars x in
	let rvars = IdSet.add x rvars in
	let (p,v) = derecurs_regexp vars b rvars false env p in
	(if f 
842
	 then seq (PGuard (mk (ICapture x))) p 
843
844
845
846
	 else termin (after || b) (IdSet.singleton x) p), 
	(if after then v else IdSet.add x v)
	  
	  
847
848
849
850
851
  and derecurs_var env loc cu v =
    let v = ident env.penv_tenv loc v in
    match cu with
      | None ->
	  (try Env.find v env.penv_derec 
852
853
854
	   with Not_found -> 
	     try itype (find_type v env.penv_tenv)
	     with Not_found -> mk (ICapture v))
855
856
857
858
859
860
      | Some cu ->
	  (try itype (find_type_global loc cu v env.penv_tenv)
	   with Not_found ->
	     raise_loc_generic loc 
	       ("Unbound external type " ^ (U.get_str cu) ^ "." ^ 
		  (Ident.to_string v)))
861
862
	      
  and derecurs_def env b =
863
864
865
866
867
868
869
870
871
872
873
874
875
    let seen = ref IdSet.empty in
    let b = 
      List.map 
	(fun (loc,v,p) -> 
	   let v = ident env.penv_tenv loc v in
	   if IdSet.mem !seen v then 
	     raise_loc_generic loc
	       ("Multiple definitions for the type identifer " ^ 
		  (Ident.to_string v));
	   seen := IdSet.add v !seen;
	   (v,p,delayed loc))
	b in

876
877
878
879
    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.desc <- ILink (derecurs env p)) b;
880
881
882
883
884
885
886
    (env, b)

  module H = Hashtbl.Make(Types)

  let rec elim_concat n =
    match n.desc with
      | IConcat (a,b) ->
887
888
	  if (n.sid > 0) 
	  then 	raise (Patterns.Error "Ill-formed concatenation loop");
889
890
	  n.sid <- 1;
	  n.desc <- ILink (elim_conc a b)
891
892
893
894
895
      | IMerge (a,b) ->
	  if (n.sid > 0) 
	  then 	raise (Patterns.Error "Ill-formed concatenation loop");
	  n.sid <- 1;
	  n.desc <- ILink (elim_merge a b)
896
      | _ -> ()
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
  and elim_merge a b =
    let get_rec t =
      let t = Types.Record.get t in
      List.map (fun (l,o,_) ->
		  o, 
		  LabelMap.map 
		    (fun (opt,x) ->
		       let x = itype x in 
		       (if opt then mk (IOptional x) else x),
		       None)
		    l) t in
    let merge (o1,l1) (o2,l2) =
      mk (IRecord (o1||o2, LabelMap.merge (fun _ x -> x) l1 l2)) in
    (* Problem: repr can loop with ill-formed recursion.
       type t = s + t where s = s | s;; *)
    match (repr a).desc, (repr b).desc with
      | IType (t1,_), IType (t2,_) -> 
	  if not (Types.subtype t1 Types.Record.any) then
	    raise 
	      (Patterns.Error 
		 "Left argument of record concatenation is not a record type");
	  if not (Types.subtype t2 Types.Record.any) then
	    raise 
	      (Patterns.Error 
		 "Right argument of record concatenation is not a record type");
	  itype (Types.Record.merge t1 t2)
      | IOr (a1,a2), _ -> ior (elim_merge a1 b) (elim_merge a2 b)
      | _, IOr (b1,b2) -> ior (elim_merge a b1) (elim_merge a b2)
      | IRecord (o1,l1), IRecord (o2,l2) -> merge (o1,l1) (o2,l2)
      | IType (t1,_), IRecord (o2,l2) ->
	  if not (Types.subtype t1 Types.Record.any) then
	    raise 
	      (Patterns.Error 
		 "Left argument of record concatenation is not a record type");
	  List.fold_left (fun accu (o1,l1) -> 
			    ior accu (merge (o1,l1) (o2,l2)))
	    iempty (get_rec t1)
      | IRecord (o1,l1), IType (t2,_) ->
	  if not (Types.subtype t2 Types.Record.any) then
	    raise 
	      (Patterns.Error 
		 "Right argument of record concatenation is not a record type");
	  List.fold_left (fun accu (o2,l2) -> 
			    ior accu (merge (o1,l1) (o2,l2)))
	    iempty (get_rec t2)
      | _ -> raise (Patterns.Error "Cannot compute record concatenation")
943
944
945
946
947
948
949
950
  and elim_conc n q =
    let mem = ref [] in
    let rec aux n =
      try List.assq n !mem
      with Not_found ->
	let r = mk_delayed () in
	mem := (n,r) :: !mem;
	let rec aux2 n =
951
	  match n.desc with
952
953
954
955
956
	    | ILink n' -> aux2 n'
	    | IOr (a,b) -> ior (aux a) (aux b)
	    | ITimes (a,b) -> mk (ITimes (a, aux b))
	    | IConcat (a,b) -> elim_concat n; aux2 n
	    | IType (t,_) -> elim_concat_type t q
957
	    | _ -> raise (Patterns.Error "Cannot compute concatenation")
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
	in
	r.desc <- ILink (aux2 n);
	r
    in
    aux n
  and elim_concat_type t q =
    if not (Types.subtype t Sequence.any) then
      raise (Patterns.Error "Left argument of concatenation is not a sequence type");
    (* TODO: check t <= [ Any* ] *)
    let mem = H.create 17 in
    let rec aux t =
      try H.find mem t 
      with Not_found ->
	let n = mk_delayed () in
	H.add mem t n;
	let d = 
	  List.fold_left
	    (fun accu (t1,t2) -> ior accu (mk (ITimes (itype t1, aux t2))))
	    (if Types.Atom.has_atom t Sequence.nil_atom then q else iempty)
	    (Types.Product.get t) in
	n.desc <- d.desc;
	n
    in
    aux t
    
   
    
  let elim_concats () =
    try
      List.iter elim_concat !concats;
      List.iter (fun n -> n.sid <- 0) !concats;
      concats := []
    with exn ->
      List.iter (fun n -> n.sid <- 0) !concats;
      concats := [];
      raise exn
994

995
996
  let derec penv p =
    let d = derecurs penv p in
997
    elim_concats ();
998
999
1000
    check_delayed ();
    internalize d;
    d
1001
1002


1003
(* API *)
1004
1005
1006

  module Ids = Set.Make(Id)
  let type_defs env b =
1007
1008
1009
1010
1011
1012
    let _,b' = derecurs_def (penv env) b in
    elim_concats ();
    check_delayed ();
    let aux loc d =
      internalize d;
      check_no_fv loc d;
1013
      try typ d
1014
      with Patterns.Error s -> raise_loc_generic loc s
1015
    in
1016
    let b = 
1017
1018
1019
      List.map2 
	(fun (loc,v,p) (v',_,d) ->
	   let t = aux loc d in
1020
1021
1022
1023
	   if (loc <> noloc) && (Types.is_empty t) then
	     warning loc 
	       ("This definition yields an empty type for " ^ (U.to_string v));
	   let v = ident env loc v in
1024
	   (v',t)) b b' in
1025
1026
    List.iter (fun (v,t) -> Types.Print.register_global 
		 (Types.CompUnit.get_current ()) (Id.value v) t) b;
1027
1028
    b

1029
1030
1031
1032
  let type_defs env b =
    try type_defs env b
    with exn -> clean_on_err (); raise exn

1033

1034
  let typ_descr d =
1035
1036
    try internalize d; typ d
    with exn -> clean_on_err (); raise exn
1037

1038
  let typ env t = 
1039
1040
1041
1042
1043
1044
    try
      let d = derec (penv env) t in
      check_no_fv t.loc d;
      try typ_node d
      with Patterns.Error s -> raise_loc_generic t.loc s
    with exn -> clean_on_err (); raise exn
1045
1046

  let pat env t = 
1047
1048
1049
1050
1051
    try
      let d = derec (penv env) t in
      try pat_node d
      with Patterns.Error s -> raise_loc_generic t.loc s
    with exn -> clean_on_err (); raise exn
1052
end
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068

let typ = IType.typ
let pat = IType.pat
let type_defs = IType.type_defs

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

let dump_ns ppf env =
  Ns.dump_table ppf env.ns



1069

1070
1071
(* II. Build skeleton *)

1072

1073
type type_fun = Types.t -> bool -> Types.t
1074

1075
module Fv = IdSet
1076

1077
1078
1079
type branch = Branch of Typed.branch * branch list

let cur_branch : branch list ref = ref []
1080

1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
let exp' loc e = 
  { Typed.exp_loc = loc; Typed.exp_typ = Types.empty; Typed.exp_descr = e; }

let exp loc fv e = fv, exp' loc e

let exp_nil = exp' noloc (Typed.Cst Sequence.nil_cst)

let pat_true = 
  let n = Patterns.make Fv.empty in
  Patterns.define n (Patterns.constr Builtin_defs.true_type);
  n

let pat_false =   
  let n = Patterns.make Fv.empty in
  Patterns.define n (Patterns.constr Builtin_defs.false_type);
  n

1098

1099
let ops = Hashtbl.create 13
1100
1101
let register_op op arity f = Hashtbl.add ops op (arity,f)
let typ_op op = snd (Hashtbl.find ops op)
1102

1103
1104
1105
1106
1107
let fun_name env a =
  match a.fun_name with
    | None -> None
    | Some (loc,s) -> Some (ident env loc s)

1108
let is_op env s = 
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
  if (Env.mem s env.ids) then None
  else
    let (ns,s) = Id.value s in
    if Ns.equal ns Ns.empty then
      let s = U.get_str s in
      try 
	let o = Hashtbl.find ops s in
	Some (s, fst o)
      with Not_found -> None
    else None
1119

1120
1121
let rec expr env loc = function
  | LocatedExpr (loc,e) -> expr env loc e
1122
  | Forget (e,t) ->
1123
      let (fv,e) = expr env loc e and t = typ env t in
1124
      exp loc fv (Typed.Forget (e,t))
1125
1126
  | Check (e,t) ->
      let (fv,e) = expr env loc e and t = typ env t in
1127
      exp loc fv (Typed.Check (ref Types.empty,e,t))
1128
  | Var s -> var env loc s
1129
  | Apply (e1,e2) -> 
1130
1131
1132
1133
1134
1135
1136
1137
      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
1138
  | (Integer _ | Char _ | Atom _ | Const _) as c -> 
1139
      exp loc Fv.empty (Typed.Cst (const env loc c))
1140
  | Pair (e1,e2) ->
1141
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
1142
1143
      exp loc (Fv.cup fv1 fv2) (Typed.Pair (e1,e2))
  | Xml (e1,e2) ->
1144
      let (fv1,e1) = expr env loc e1 and (fv2,e2) = expr env loc e2 in
1145
      exp loc (Fv.cup fv1 fv2) (Typed.Xml (e1,e2))