src/Pure/Syntax/parser.ML
author wenzelm
Sat Jan 26 22:14:07 2008 +0100 (2008-01-26 ago)
changeset 25986 26f1e4c172c3
parent 24245 4ffeb1dd048a
child 25987 bfda3f3beccd
permissions -rw-r--r--
syntax error: reduced spam -- print expected nonterminals instead of terminals;
tuned pretty_gram;
wenzelm@18
     1
(*  Title:      Pure/Syntax/parser.ML
wenzelm@46
     2
    ID:         $Id$
wenzelm@2186
     3
    Author:     Carsten Clasohm, Sonia Mahjoub, and Markus Wenzel, TU Muenchen
wenzelm@18
     4
wenzelm@15979
     5
General context-free parser for the inner syntax of terms, types, etc.
wenzelm@18
     6
*)
wenzelm@18
     7
wenzelm@18
     8
signature PARSER =
wenzelm@15752
     9
sig
paulson@1507
    10
  type gram
paulson@1507
    11
  val empty_gram: gram
paulson@1507
    12
  val extend_gram: gram -> SynExt.xprod list -> gram
wenzelm@15752
    13
  val make_gram: SynExt.xprod list -> gram
paulson@1507
    14
  val merge_grams: gram -> gram -> gram
paulson@1507
    15
  val pretty_gram: gram -> Pretty.T list
paulson@1507
    16
  datatype parsetree =
paulson@1507
    17
    Node of string * parsetree list |
paulson@1507
    18
    Tip of Lexicon.token
paulson@1507
    19
  val parse: gram -> string -> Lexicon.token list -> parsetree list
paulson@1507
    20
  val branching_level: int ref
wenzelm@15752
    21
end;
paulson@1507
    22
wenzelm@15752
    23
structure Parser: PARSER =
wenzelm@18
    24
struct
wenzelm@15752
    25
wenzelm@237
    26
open Lexicon SynExt;
wenzelm@18
    27
wenzelm@15752
    28
wenzelm@18
    29
(** datatype gram **)
wenzelm@18
    30
clasohm@1175
    31
type nt_tag = int;              (*production for the NTs are stored in an array
clasohm@1175
    32
                                  so we can identify NTs by their index*)
clasohm@1147
    33
clasohm@1147
    34
datatype symb = Terminal of token
clasohm@1175
    35
              | Nonterminal of nt_tag * int;              (*(tag, precedence)*)
clasohm@1147
    36
clasohm@1175
    37
type nt_gram = ((nt_tag list * token list) *
clasohm@1147
    38
                (token option * (symb list * string * int) list) list);
clasohm@1147
    39
                                     (*(([dependent_nts], [start_tokens]),
clasohm@1147
    40
                                        [(start_token, [(rhs, name, prio)])])*)
clasohm@1175
    41
                              (*depent_nts is a list of all NTs whose lookahead
clasohm@1175
    42
                                depends on this NT's lookahead*)
clasohm@1147
    43
clasohm@1147
    44
datatype gram =
clasohm@1147
    45
  Gram of {nt_count: int, prod_count: int,
clasohm@1147
    46
           tags: nt_tag Symtab.table,
clasohm@1147
    47
           chains: (nt_tag * nt_tag list) list,              (*[(to, [from])]*)
clasohm@1147
    48
           lambdas: nt_tag list,
clasohm@1175
    49
           prods: nt_gram Array.array};
clasohm@1175
    50
                       (*"tags" is used to map NT names (i.e. strings) to tags;
clasohm@1175
    51
                         chain productions are not stored as normal productions
clasohm@1175
    52
                         but instead as an entry in "chains";
clasohm@1175
    53
                         lambda productions are stored as normal productions
clasohm@1175
    54
                         and also as an entry in "lambdas"*)
clasohm@1147
    55
wenzelm@25986
    56
fun reverted_tags (tags: nt_tag Symtab.table) =
wenzelm@25986
    57
  the o Inttab.lookup (Inttab.make (map swap (Symtab.dest tags)));
wenzelm@25986
    58
clasohm@1175
    59
val UnknownStart = EndToken;       (*productions for which no starting token is
clasohm@1175
    60
                                     known yet are associated with this token*)
clasohm@1147
    61
wenzelm@15752
    62
(* get all NTs that are connected with a list of NTs
clasohm@1175
    63
   (used for expanding chain list)*)
wenzelm@23909
    64
fun connected_with _ ([]: nt_tag list) relatives = relatives
clasohm@1147
    65
  | connected_with chains (root :: roots) relatives =
wenzelm@19305
    66
    let val branches = subtract (op =) relatives (these (AList.lookup (op =) chains root));
clasohm@1147
    67
    in connected_with chains (branches @ roots) (branches @ relatives) end;
wenzelm@18
    68
clasohm@1147
    69
(* convert productions to grammar;
clasohm@1175
    70
   N.B. that the chains parameter has the form [(from, [to])];
skalberg@15531
    71
   prod_count is of type "int option" and is only updated if it is <> NONE*)
clasohm@1147
    72
fun add_prods _ chains lambdas prod_count [] = (chains, lambdas, prod_count)
clasohm@1147
    73
  | add_prods prods chains lambdas prod_count
clasohm@1147
    74
              ((lhs, new_prod as (rhs, name, pri)) :: ps) =
clasohm@1147
    75
    let
wenzelm@15752
    76
      val chain_from = case (pri, rhs) of (~1, [Nonterminal (id, ~1)]) => SOME id | _ => NONE;
wenzelm@15752
    77
wenzelm@15752
    78
      (*store chain if it does not already exist*)
wenzelm@22573
    79
      val (new_chain, chains') = case chain_from of NONE => (NONE, chains) | SOME from_ =>
wenzelm@22573
    80
        let val old_tos = these (AList.lookup (op =) chains from_) in
wenzelm@20664
    81
          if member (op =) old_tos lhs then (NONE, chains)
wenzelm@22573
    82
          else (SOME from_, AList.update (op =) (from_, insert (op =) lhs old_tos) chains)
clasohm@1175
    83
        end;
clasohm@1147
    84
clasohm@1175
    85
      (*propagate new chain in lookahead and lambda lists;
clasohm@1175
    86
        added_starts is used later to associate existing
clasohm@1175
    87
        productions with new starting tokens*)
clasohm@1147
    88
      val (added_starts, lambdas') =
clasohm@1147
    89
        if is_none new_chain then ([], lambdas) else
clasohm@1175
    90
        let (*lookahead of chain's source*)
wenzelm@15979
    91
            val ((from_nts, from_tks), _) = Array.sub (prods, the new_chain);
clasohm@1147
    92
clasohm@1175
    93
            (*copy from's lookahead to chain's destinations*)
clasohm@1147
    94
            fun copy_lookahead [] added = added
clasohm@1147
    95
              | copy_lookahead (to :: tos) added =
clasohm@1175
    96
                let
clasohm@1175
    97
                  val ((to_nts, to_tks), ps) = Array.sub (prods, to);
clasohm@330
    98
wenzelm@19305
    99
                  val new_tks = subtract (op =) to_tks from_tks;  (*added lookahead tokens*)
clasohm@1175
   100
                in Array.update (prods, to, ((to_nts, to_tks @ new_tks), ps));
clasohm@1147
   101
                   copy_lookahead tos (if null new_tks then added
clasohm@1147
   102
                                       else (to, new_tks) :: added)
clasohm@1147
   103
                end;
clasohm@1147
   104
clasohm@1147
   105
            val tos = connected_with chains' [lhs] [lhs];
clasohm@1175
   106
        in (copy_lookahead tos [],
wenzelm@23909
   107
            gen_union (op =) (if member (op =) lambdas lhs then tos else [], lambdas))
clasohm@1147
   108
        end;
clasohm@1147
   109
clasohm@1175
   110
      (*test if new production can produce lambda
clasohm@1175
   111
        (rhs must either be empty or only consist of lambda NTs)*)
clasohm@1175
   112
      val (new_lambda, lambdas') =
wenzelm@20664
   113
        if forall (fn (Nonterminal (id, _)) => member (op =) lambdas' id
clasohm@1175
   114
                    | (Terminal _) => false) rhs then
wenzelm@23909
   115
          (true, gen_union (op =) (lambdas', connected_with chains' [lhs] [lhs]))
clasohm@1175
   116
        else
clasohm@1175
   117
          (false, lambdas');
clasohm@1147
   118
clasohm@1175
   119
      (*list optional terminal and all nonterminals on which the lookahead
clasohm@1175
   120
        of a production depends*)
skalberg@15531
   121
      fun lookahead_dependency _ [] nts = (NONE, nts)
skalberg@15531
   122
        | lookahead_dependency _ ((Terminal tk) :: _) nts = (SOME tk, nts)
clasohm@1147
   123
        | lookahead_dependency lambdas ((Nonterminal (nt, _)) :: symbs) nts =
wenzelm@20664
   124
            if member (op =) lambdas nt then
clasohm@1147
   125
              lookahead_dependency lambdas symbs (nt :: nts)
skalberg@15531
   126
            else (NONE, nt :: nts);
clasohm@1147
   127
clasohm@1175
   128
      (*get all known starting tokens for a nonterminal*)
clasohm@1175
   129
      fun starts_for_nt nt = snd (fst (Array.sub (prods, nt)));
clasohm@1175
   130
clasohm@1147
   131
      val token_union = gen_union matching_tokens;
clasohm@330
   132
clasohm@1147
   133
      (*update prods, lookaheads, and lambdas according to new lambda NTs*)
clasohm@1147
   134
      val (added_starts', lambdas') =
clasohm@1147
   135
        let
clasohm@1147
   136
          (*propagate added lambda NT*)
clasohm@1175
   137
          fun propagate_lambda [] added_starts lambdas= (added_starts, lambdas)
clasohm@1147
   138
            | propagate_lambda (l :: ls) added_starts lambdas =
clasohm@1147
   139
              let
clasohm@1175
   140
                (*get lookahead for lambda NT*)
clasohm@1175
   141
                val ((dependent, l_starts), _) = Array.sub (prods, l);
clasohm@1147
   142
wenzelm@15752
   143
                (*check productions whose lookahead may depend on lambda NT*)
clasohm@1147
   144
                fun examine_prods [] add_lambda nt_dependencies added_tks
clasohm@1175
   145
                                  nt_prods =
clasohm@1147
   146
                      (add_lambda, nt_dependencies, added_tks, nt_prods)
clasohm@1147
   147
                  | examine_prods ((p as (rhs, _, _)) :: ps) add_lambda
clasohm@1147
   148
                      nt_dependencies added_tks nt_prods =
clasohm@1147
   149
                    let val (tk, nts) = lookahead_dependency lambdas rhs [];
clasohm@1147
   150
                    in
wenzelm@20664
   151
                      if member (op =) nts l then       (*update production's lookahead*)
clasohm@1147
   152
                      let
clasohm@1147
   153
                        val new_lambda = is_none tk andalso nts subset lambdas;
clasohm@1147
   154
wenzelm@19305
   155
                        val new_tks = subtract (op =) l_starts
wenzelm@19305
   156
                          ((if is_some tk then [the tk] else []) @
wenzelm@19305
   157
                            Library.foldl token_union ([], map starts_for_nt nts));
wenzelm@15752
   158
clasohm@1147
   159
                        val added_tks' = token_union (new_tks, added_tks);
clasohm@1147
   160
wenzelm@23909
   161
                        val nt_dependencies' = gen_union (op =) (nts, nt_dependencies);
clasohm@1147
   162
clasohm@1175
   163
                        (*associate production with new starting tokens*)
wenzelm@23909
   164
                        fun copy ([]: token option list) nt_prods = nt_prods
clasohm@1175
   165
                          | copy (tk :: tks) nt_prods =
haftmann@17192
   166
                            let val old_prods = these (AList.lookup (op =) nt_prods tk);
clasohm@1175
   167
clasohm@1175
   168
                                val prods' = p :: old_prods;
haftmann@17192
   169
                            in nt_prods
haftmann@17192
   170
                               |> AList.update (op =) (tk, prods')
haftmann@17192
   171
                               |> copy tks
clasohm@1175
   172
                            end;
wenzelm@15752
   173
clasohm@1147
   174
                        val nt_prods' =
skalberg@15531
   175
                          let val new_opt_tks = map SOME new_tks;
skalberg@15531
   176
                          in copy ((if new_lambda then [NONE] else []) @
clasohm@1175
   177
                                   new_opt_tks) nt_prods
clasohm@1147
   178
                          end;
clasohm@1147
   179
                      in examine_prods ps (add_lambda orelse new_lambda)
clasohm@1147
   180
                           nt_dependencies' added_tks' nt_prods'
clasohm@1147
   181
                      end
clasohm@1175
   182
                      else                                  (*skip production*)
clasohm@1175
   183
                        examine_prods ps add_lambda nt_dependencies
clasohm@1175
   184
                                      added_tks nt_prods
clasohm@1147
   185
                    end;
clasohm@1147
   186
clasohm@1147
   187
                (*check each NT whose lookahead depends on new lambda NT*)
clasohm@1147
   188
                fun process_nts [] added_lambdas added_starts =
clasohm@1147
   189
                      (added_lambdas, added_starts)
clasohm@1147
   190
                  | process_nts (nt :: nts) added_lambdas added_starts =
clasohm@1147
   191
                    let
clasohm@1147
   192
                      val (lookahead as (old_nts, old_tks), nt_prods) =
clasohm@1147
   193
                        Array.sub (prods, nt);
clasohm@1147
   194
clasohm@1175
   195
                      (*existing productions whose lookahead may depend on l*)
clasohm@1175
   196
                      val tk_prods =
haftmann@17192
   197
                        (these o AList.lookup (op =) nt_prods)
skalberg@15570
   198
                               (SOME (hd l_starts  handle Empty => UnknownStart));
clasohm@1147
   199
clasohm@1175
   200
                      (*add_lambda is true if an existing production of the nt
clasohm@1175
   201
                        produces lambda due to the new lambda NT l*)
clasohm@1147
   202
                      val (add_lambda, nt_dependencies, added_tks, nt_prods') =
clasohm@1147
   203
                        examine_prods tk_prods false [] [] nt_prods;
clasohm@1147
   204
wenzelm@19305
   205
                      val added_nts = subtract (op =) old_nts nt_dependencies;
clasohm@1147
   206
clasohm@1147
   207
                      val added_lambdas' =
clasohm@1147
   208
                        if add_lambda then nt :: added_lambdas
clasohm@1147
   209
                        else added_lambdas;
clasohm@1147
   210
                    in Array.update (prods, nt,
clasohm@1175
   211
                                   ((added_nts @ old_nts, old_tks @ added_tks),
clasohm@1175
   212
                                    nt_prods'));
clasohm@1175
   213
                                          (*N.B. that because the tks component
clasohm@1175
   214
                                            is used to access existing
clasohm@1175
   215
                                            productions we have to add new
clasohm@1175
   216
                                            tokens at the _end_ of the list*)
clasohm@1175
   217
clasohm@1147
   218
                       if null added_tks then
clasohm@1147
   219
                         process_nts nts added_lambdas' added_starts
clasohm@1147
   220
                       else
clasohm@1147
   221
                         process_nts nts added_lambdas'
clasohm@1147
   222
                                      ((nt, added_tks) :: added_starts)
clasohm@1147
   223
                    end;
clasohm@1147
   224
clasohm@1147
   225
                val (added_lambdas, added_starts') =
clasohm@1147
   226
                  process_nts dependent [] added_starts;
clasohm@330
   227
wenzelm@19305
   228
                val added_lambdas' = subtract (op =) lambdas added_lambdas;
clasohm@1147
   229
              in propagate_lambda (ls @ added_lambdas') added_starts'
clasohm@1147
   230
                                  (added_lambdas' @ lambdas)
clasohm@1147
   231
              end;
wenzelm@19305
   232
        in propagate_lambda (subtract (op =) lambdas lambdas') added_starts lambdas' end;
clasohm@1147
   233
clasohm@1147
   234
      (*insert production into grammar*)
clasohm@1147
   235
      val (added_starts', prod_count') =
wenzelm@15979
   236
        if is_some chain_from then (added_starts', prod_count)  (*don't store chain production*)
clasohm@1147
   237
        else let
clasohm@1175
   238
          (*lookahead tokens of new production and on which
clasohm@1175
   239
            NTs lookahead depends*)
clasohm@1175
   240
          val (start_tk, start_nts) = lookahead_dependency lambdas' rhs [];
clasohm@1147
   241
skalberg@15570
   242
          val start_tks = Library.foldl token_union
wenzelm@15979
   243
                          (if is_some start_tk then [the start_tk] else [],
clasohm@1175
   244
                           map starts_for_nt start_nts);
clasohm@1175
   245
skalberg@15531
   246
          val opt_starts = (if new_lambda then [NONE]
skalberg@15531
   247
                            else if null start_tks then [SOME UnknownStart]
skalberg@15531
   248
                            else []) @ (map SOME start_tks);
clasohm@1147
   249
clasohm@1147
   250
          (*add lhs NT to list of dependent NTs in lookahead*)
clasohm@1147
   251
          fun add_nts [] = ()
clasohm@1147
   252
            | add_nts (nt :: nts) =
clasohm@1147
   253
              let val ((old_nts, old_tks), ps) = Array.sub (prods, nt);
wenzelm@20664
   254
              in if member (op =) old_nts lhs then ()
clasohm@1175
   255
                 else Array.update (prods, nt, ((lhs :: old_nts, old_tks), ps))
clasohm@1147
   256
              end;
clasohm@330
   257
clasohm@1175
   258
          (*add new start tokens to chained NTs' lookahead list;
clasohm@1147
   259
            also store new production for lhs NT*)
clasohm@1147
   260
          fun add_tks [] added prod_count = (added, prod_count)
clasohm@1147
   261
            | add_tks (nt :: nts) added prod_count =
clasohm@1147
   262
              let
clasohm@1175
   263
                val ((old_nts, old_tks), nt_prods) = Array.sub (prods, nt);
clasohm@1147
   264
haftmann@20951
   265
                val new_tks = subtract matching_tokens old_tks start_tks;
clasohm@1147
   266
clasohm@1175
   267
                (*store new production*)
clasohm@1147
   268
                fun store [] prods is_new =
wenzelm@15979
   269
                      (prods, if is_some prod_count andalso is_new then
skalberg@15570
   270
                                Option.map (fn x => x+1) prod_count
clasohm@1175
   271
                              else prod_count, is_new)
clasohm@1175
   272
                  | store (tk :: tks) prods is_new =
haftmann@17192
   273
                    let val tk_prods = (these o AList.lookup (op =) prods) tk;
clasohm@330
   274
skalberg@15531
   275
                        (*if prod_count = NONE then we can assume that
clasohm@1175
   276
                          grammar does not contain new production already*)
clasohm@1175
   277
                        val (tk_prods', is_new') =
wenzelm@15979
   278
                          if is_some prod_count then
wenzelm@20664
   279
                            if member (op =) tk_prods new_prod then (tk_prods, false)
clasohm@1147
   280
                            else (new_prod :: tk_prods, true)
clasohm@1147
   281
                          else (new_prod :: tk_prods, true);
clasohm@1175
   282
haftmann@17192
   283
                        val prods' = prods
wenzelm@23909
   284
                          |> is_new' ? AList.update (op =) (tk: token option, tk_prods');
clasohm@1175
   285
                    in store tks prods' (is_new orelse is_new') end;
clasohm@330
   286
clasohm@1175
   287
                val (nt_prods', prod_count', changed) =
clasohm@1175
   288
                  if nt = lhs then store opt_starts nt_prods false
clasohm@1175
   289
                              else (nt_prods, prod_count, false);
clasohm@1175
   290
              in if not changed andalso null new_tks then ()
clasohm@1175
   291
                 else Array.update (prods, nt, ((old_nts, old_tks @ new_tks),
clasohm@1175
   292
                                                nt_prods'));
clasohm@1147
   293
                 add_tks nts (if null new_tks then added
clasohm@1147
   294
                              else (nt, new_tks) :: added) prod_count'
clasohm@1147
   295
              end;
clasohm@1147
   296
        in add_nts start_nts;
clasohm@1147
   297
           add_tks (connected_with chains' [lhs] [lhs]) [] prod_count
clasohm@330
   298
        end;
clasohm@330
   299
clasohm@1175
   300
      (*associate productions with new lookaheads*)
clasohm@1147
   301
      val dummy =
clasohm@1147
   302
        let
clasohm@1147
   303
          (*propagate added start tokens*)
clasohm@1147
   304
          fun add_starts [] = ()
clasohm@1147
   305
            | add_starts ((changed_nt, new_tks) :: starts) =
clasohm@1147
   306
              let
clasohm@1175
   307
                (*token under which old productions which
clasohm@1175
   308
                  depend on changed_nt could be stored*)
clasohm@1175
   309
                val key =
wenzelm@20664
   310
                 case find_first (not o member (op =) new_tks)
clasohm@1175
   311
                                 (starts_for_nt changed_nt) of
skalberg@15531
   312
                      NONE => SOME UnknownStart
clasohm@1175
   313
                    | t => t;
clasohm@1175
   314
clasohm@1175
   315
                (*copy productions whose lookahead depends on changed_nt;
skalberg@15531
   316
                  if key = SOME UnknownToken then tk_prods is used to hold
clasohm@1175
   317
                  the productions not copied*)
clasohm@1147
   318
                fun update_prods [] result = result
wenzelm@23909
   319
                  | update_prods ((p as (rhs, _: string, _: nt_tag)) :: ps)
clasohm@1147
   320
                      (tk_prods, nt_prods) =
clasohm@1147
   321
                    let
clasohm@1175
   322
                      (*lookahead dependency for production*)
clasohm@1147
   323
                      val (tk, depends) = lookahead_dependency lambdas' rhs [];
wenzelm@18
   324
clasohm@1175
   325
                      (*test if this production has to be copied*)
wenzelm@20664
   326
                      val update = member (op =) depends changed_nt;
clasohm@1175
   327
clasohm@1175
   328
                      (*test if production could already be associated with
clasohm@1175
   329
                        a member of new_tks*)
clasohm@1147
   330
                      val lambda = length depends > 1 orelse
wenzelm@15979
   331
                                   not (null depends) andalso is_some tk
wenzelm@20664
   332
                                   andalso member (op =) new_tks (the tk);
clasohm@330
   333
clasohm@1175
   334
                      (*associate production with new starting tokens*)
wenzelm@23909
   335
                      fun copy ([]: token list) nt_prods = nt_prods
clasohm@1147
   336
                        | copy (tk :: tks) nt_prods =
clasohm@1147
   337
                          let
haftmann@17192
   338
                            val tk_prods = (these o AList.lookup (op =) nt_prods) (SOME tk);
clasohm@330
   339
clasohm@1147
   340
                            val tk_prods' =
clasohm@1147
   341
                              if not lambda then p :: tk_prods
haftmann@20854
   342
                              else insert (op =) p tk_prods;
clasohm@1147
   343
                                      (*if production depends on lambda NT we
clasohm@1147
   344
                                        have to look for duplicates*)
haftmann@17192
   345
                         in
haftmann@17192
   346
                           nt_prods
haftmann@17192
   347
                           |> AList.update (op =) (SOME tk, tk_prods')
haftmann@17192
   348
                           |> copy tks
clasohm@1147
   349
                         end;
clasohm@1147
   350
                      val result =
clasohm@1175
   351
                        if update then
clasohm@1175
   352
                          (tk_prods, copy new_tks nt_prods)
skalberg@15531
   353
                        else if key = SOME UnknownStart then
clasohm@1175
   354
                          (p :: tk_prods, nt_prods)
clasohm@1175
   355
                        else (tk_prods, nt_prods);
clasohm@1147
   356
                    in update_prods ps result end;
clasohm@330
   357
clasohm@1147
   358
                (*copy existing productions for new starting tokens*)
clasohm@1175
   359
                fun process_nts [] added = added
clasohm@1175
   360
                  | process_nts (nt :: nts) added =
clasohm@1147
   361
                    let
clasohm@1147
   362
                      val (lookahead as (old_nts, old_tks), nt_prods) =
clasohm@1147
   363
                        Array.sub (prods, nt);
clasohm@377
   364
haftmann@17192
   365
                      val tk_prods = (these o AList.lookup (op =) nt_prods) key;
clasohm@1147
   366
clasohm@1175
   367
                      (*associate productions with new lookahead tokens*)
clasohm@1147
   368
                      val (tk_prods', nt_prods') =
clasohm@1147
   369
                        update_prods tk_prods ([], nt_prods);
clasohm@330
   370
clasohm@1147
   371
                      val nt_prods' =
haftmann@17192
   372
                        nt_prods'
wenzelm@21565
   373
                        |> (key = SOME UnknownStart) ? AList.update (op =) (key, tk_prods')
wenzelm@18
   374
clasohm@1175
   375
                      val added_tks =
haftmann@20951
   376
                        subtract matching_tokens old_tks new_tks;
clasohm@1147
   377
                    in if null added_tks then
clasohm@1147
   378
                         (Array.update (prods, nt, (lookahead, nt_prods'));
clasohm@1175
   379
                          process_nts nts added)
clasohm@1147
   380
                       else
clasohm@1147
   381
                         (Array.update (prods, nt,
clasohm@1147
   382
                            ((old_nts, added_tks @ old_tks), nt_prods'));
clasohm@1175
   383
                          process_nts nts ((nt, added_tks) :: added))
clasohm@1147
   384
                    end;
wenzelm@18
   385
clasohm@1175
   386
                val ((dependent, _), _) = Array.sub (prods, changed_nt);
clasohm@1175
   387
              in add_starts (starts @ (process_nts dependent [])) end;
clasohm@1147
   388
        in add_starts added_starts' end;
clasohm@1147
   389
  in add_prods prods chains' lambdas' prod_count ps end;
wenzelm@237
   390
wenzelm@18
   391
wenzelm@237
   392
(* pretty_gram *)
wenzelm@18
   393
clasohm@1147
   394
fun pretty_gram (Gram {tags, prods, chains, ...}) =
wenzelm@237
   395
  let
wenzelm@237
   396
    fun pretty_name name = [Pretty.str (name ^ " =")];
wenzelm@18
   397
wenzelm@25986
   398
    val nt_name = reverted_tags tags;
clasohm@1147
   399
berghofe@14600
   400
    fun pretty_symb (Terminal (Token s)) = Pretty.quote (Pretty.str s)
wenzelm@237
   401
      | pretty_symb (Terminal tok) = Pretty.str (str_of_token tok)
wenzelm@25986
   402
      | pretty_symb (Nonterminal (tag, p)) = Pretty.str (nt_name tag ^ "[" ^ string_of_int p ^ "]");
wenzelm@18
   403
wenzelm@237
   404
    fun pretty_const "" = []
berghofe@14600
   405
      | pretty_const c = [Pretty.str ("=> " ^ Library.quote c)];
wenzelm@237
   406
wenzelm@237
   407
    fun pretty_pri p = [Pretty.str ("(" ^ string_of_int p ^ ")")];
wenzelm@237
   408
clasohm@1147
   409
    fun pretty_prod name (symbs, const, pri) =
wenzelm@237
   410
      Pretty.block (Pretty.breaks (pretty_name name @
wenzelm@237
   411
        map pretty_symb symbs @ pretty_const const @ pretty_pri pri));
clasohm@1147
   412
clasohm@1147
   413
    fun pretty_nt (name, tag) =
clasohm@1147
   414
      let
wenzelm@22573
   415
        fun prod_of_chain from_ = ([Nonterminal (from_, ~1)], "", ~1);
clasohm@1147
   416
clasohm@1147
   417
        val nt_prods =
wenzelm@23909
   418
          Library.foldl (gen_union op =) ([], map snd (snd (Array.sub (prods, tag)))) @
haftmann@17192
   419
          map prod_of_chain ((these o AList.lookup (op =) chains) tag);
clasohm@1147
   420
      in map (pretty_prod name) nt_prods end;
wenzelm@15752
   421
wenzelm@25986
   422
  in maps pretty_nt (sort_wrt fst (Symtab.dest tags)) end;
clasohm@1147
   423
clasohm@1147
   424
clasohm@1438
   425
(** Operations on gramars **)
clasohm@1147
   426
clasohm@1438
   427
(*The mother of all grammars*)
clasohm@1147
   428
val empty_gram = Gram {nt_count = 0, prod_count = 0,
wenzelm@4487
   429
                       tags = Symtab.empty, chains = [], lambdas = [],
clasohm@1147
   430
                       prods = Array.array (0, (([], []), []))};
clasohm@1147
   431
clasohm@1438
   432
clasohm@1438
   433
(*Invert list of chain productions*)
clasohm@1147
   434
fun inverse_chains [] result = result
wenzelm@23909
   435
  | inverse_chains ((root, branches: nt_tag list) :: cs) result =
wenzelm@23909
   436
    let fun add ([]: nt_tag list) result = result
clasohm@1147
   437
          | add (id :: ids) result =
haftmann@17192
   438
            let val old = (these o AList.lookup (op =) result) id;
haftmann@17192
   439
            in add ids (AList.update (op =) (id, root :: old) result) end;
clasohm@1147
   440
    in inverse_chains cs (add branches result) end;
clasohm@1147
   441
clasohm@1438
   442
clasohm@1438
   443
(*Add productions to a grammar*)
clasohm@1147
   444
fun extend_gram gram [] = gram
clasohm@1147
   445
  | extend_gram (Gram {nt_count, prod_count, tags, chains, lambdas, prods})
clasohm@1147
   446
                xprods =
clasohm@1147
   447
  let
clasohm@1438
   448
    (*Get tag for existing nonterminal or create a new one*)
clasohm@1147
   449
    fun get_tag nt_count tags nt =
wenzelm@17412
   450
      case Symtab.lookup tags nt of
skalberg@15531
   451
        SOME tag => (nt_count, tags, tag)
wenzelm@17412
   452
      | NONE => (nt_count+1, Symtab.update_new (nt, nt_count) tags,
clasohm@1147
   453
                 nt_count);
clasohm@1438
   454
clasohm@1438
   455
    (*Convert symbols to the form used by the parser;
clasohm@1438
   456
      delimiters and predefined terms are stored as terminals,
clasohm@1438
   457
      nonterminals are converted to integer tags*)
clasohm@1147
   458
    fun symb_of [] nt_count tags result = (nt_count, tags, rev result)
clasohm@1147
   459
      | symb_of ((Delim s) :: ss) nt_count tags result =
clasohm@1147
   460
          symb_of ss nt_count tags ((Terminal (Token s)) :: result)
clasohm@1147
   461
      | symb_of ((Argument (s, p)) :: ss) nt_count tags result =
clasohm@1147
   462
          let
clasohm@1147
   463
            val (nt_count', tags', new_symb) =
clasohm@1147
   464
              case predef_term s of
skalberg@15531
   465
                NONE =>
clasohm@1147
   466
                  let val (nt_count', tags', s_tag) = get_tag nt_count tags s;
clasohm@1147
   467
                  in (nt_count', tags', Nonterminal (s_tag, p)) end
skalberg@15531
   468
              | SOME tk => (nt_count, tags, Terminal tk);
clasohm@1147
   469
          in symb_of ss nt_count' tags' (new_symb :: result) end
clasohm@1147
   470
      | symb_of (_ :: ss) nt_count tags result =
clasohm@1147
   471
          symb_of ss nt_count tags result;
clasohm@1147
   472
clasohm@1438
   473
    (*Convert list of productions by invoking symb_of for each of them*)
clasohm@1147
   474
    fun prod_of [] nt_count prod_count tags result =
clasohm@1147
   475
          (nt_count, prod_count, tags, result)
clasohm@1147
   476
      | prod_of ((XProd (lhs, xsymbs, const, pri)) :: ps)
clasohm@1147
   477
                nt_count prod_count tags result =
clasohm@1147
   478
        let val (nt_count', tags', lhs_tag) = get_tag nt_count tags lhs;
clasohm@1147
   479
clasohm@1147
   480
            val (nt_count'', tags'', prods) =
clasohm@1147
   481
              symb_of xsymbs nt_count' tags' [];
clasohm@1147
   482
        in prod_of ps nt_count'' (prod_count+1) tags''
clasohm@1147
   483
                   ((lhs_tag, (prods, const, pri)) :: result)
clasohm@1147
   484
        end;
clasohm@1147
   485
clasohm@1438
   486
    val (nt_count', prod_count', tags', xprods') =
clasohm@1147
   487
      prod_of xprods nt_count prod_count tags [];
clasohm@1147
   488
clasohm@1438
   489
    (*Copy array containing productions of old grammar;
clasohm@1438
   490
      this has to be done to preserve the old grammar while being able
clasohm@1438
   491
      to change the array's content*)
clasohm@1147
   492
    val prods' =
clasohm@1147
   493
      let fun get_prod i = if i < nt_count then Array.sub (prods, i)
clasohm@1147
   494
                           else (([], []), []);
clasohm@1147
   495
      in Array.tabulate (nt_count', get_prod) end;
clasohm@1147
   496
clasohm@1147
   497
    val fromto_chains = inverse_chains chains [];
clasohm@1147
   498
clasohm@1438
   499
    (*Add new productions to old ones*)
clasohm@1147
   500
    val (fromto_chains', lambdas', _) =
skalberg@15531
   501
      add_prods prods' fromto_chains lambdas NONE xprods';
clasohm@1147
   502
clasohm@1147
   503
    val chains' = inverse_chains fromto_chains' [];
clasohm@1177
   504
  in Gram {nt_count = nt_count', prod_count = prod_count', tags = tags',
clasohm@1147
   505
           chains = chains', lambdas = lambdas', prods = prods'}
wenzelm@18
   506
  end;
wenzelm@18
   507
wenzelm@15752
   508
val make_gram = extend_gram empty_gram;
wenzelm@15752
   509
wenzelm@18
   510
clasohm@1438
   511
(*Merge two grammars*)
clasohm@1147
   512
fun merge_grams gram_a gram_b =
clasohm@1147
   513
  let
clasohm@1147
   514
    (*find out which grammar is bigger*)
clasohm@1147
   515
    val (Gram {nt_count = nt_count1, prod_count = prod_count1, tags = tags1,
clasohm@1147
   516
               chains = chains1, lambdas = lambdas1, prods = prods1},
clasohm@1147
   517
         Gram {nt_count = nt_count2, prod_count = prod_count2, tags = tags2,
clasohm@1147
   518
               chains = chains2, lambdas = lambdas2, prods = prods2}) =
clasohm@1147
   519
      let val Gram {prod_count = count_a, ...} = gram_a;
clasohm@1147
   520
          val Gram {prod_count = count_b, ...} = gram_b;
clasohm@1147
   521
      in if count_a > count_b then (gram_a, gram_b)
clasohm@1147
   522
                              else (gram_b, gram_a)
clasohm@1147
   523
      end;
clasohm@1147
   524
clasohm@1147
   525
    (*get existing tag from grammar1 or create a new one*)
clasohm@1147
   526
    fun get_tag nt_count tags nt =
wenzelm@17412
   527
      case Symtab.lookup tags nt of
skalberg@15531
   528
        SOME tag => (nt_count, tags, tag)
wenzelm@17412
   529
      | NONE => (nt_count+1, Symtab.update_new (nt, nt_count) tags,
clasohm@1147
   530
                nt_count)
clasohm@1147
   531
clasohm@1147
   532
    val ((nt_count1', tags1'), tag_table) =
clasohm@1147
   533
      let val tag_list = Symtab.dest tags2;
clasohm@1147
   534
clasohm@1147
   535
          val table = Array.array (nt_count2, ~1);
clasohm@1147
   536
clasohm@1147
   537
          fun store_tag nt_count tags ~1 = (nt_count, tags)
clasohm@1147
   538
            | store_tag nt_count tags tag =
clasohm@1147
   539
              let val (nt_count', tags', tag') =
clasohm@1147
   540
                   get_tag nt_count tags
wenzelm@15979
   541
                     (fst (the (find_first (fn (n, t) => t = tag) tag_list)));
clasohm@1147
   542
              in Array.update (table, tag, tag');
clasohm@1147
   543
                 store_tag nt_count' tags' (tag-1)
clasohm@1147
   544
              end;
clasohm@1147
   545
      in (store_tag nt_count1 tags1 (nt_count2-1), table) end;
wenzelm@15752
   546
clasohm@1147
   547
    (*convert grammar2 tag to grammar1 tag*)
clasohm@1147
   548
    fun convert_tag tag = Array.sub (tag_table, tag);
clasohm@1147
   549
clasohm@1147
   550
    (*convert chain list to raw productions*)
clasohm@1147
   551
    fun mk_chain_prods [] result = result
clasohm@1147
   552
      | mk_chain_prods ((to, froms) :: cs) result =
clasohm@1147
   553
        let
clasohm@1147
   554
          val to_tag = convert_tag to;
clasohm@1147
   555
clasohm@1147
   556
          fun make [] result = result
wenzelm@22573
   557
            | make (from_ :: froms) result = make froms ((to_tag,
wenzelm@22573
   558
                ([Nonterminal (convert_tag from_, ~1)], "", ~1)) :: result);
clasohm@1147
   559
        in mk_chain_prods cs (make froms [] @ result) end;
wenzelm@15752
   560
clasohm@1147
   561
    val chain_prods = mk_chain_prods chains2 [];
clasohm@1147
   562
clasohm@1147
   563
    (*convert prods2 array to productions*)
clasohm@1147
   564
    fun process_nt ~1 result = result
clasohm@1147
   565
      | process_nt nt result =
clasohm@1147
   566
        let
wenzelm@23909
   567
          val nt_prods = Library.foldl (gen_union op =)
clasohm@1147
   568
                             ([], map snd (snd (Array.sub (prods2, nt))));
clasohm@1147
   569
          val lhs_tag = convert_tag nt;
clasohm@1147
   570
clasohm@1147
   571
          (*convert tags in rhs*)
clasohm@1147
   572
          fun process_rhs [] result = result
clasohm@1147
   573
            | process_rhs (Terminal tk :: rhs) result =
clasohm@1147
   574
                process_rhs rhs (result @ [Terminal tk])
clasohm@1147
   575
            | process_rhs (Nonterminal (nt, prec) :: rhs) result =
clasohm@1147
   576
                process_rhs rhs
clasohm@1147
   577
                            (result @ [Nonterminal (convert_tag nt, prec)]);
clasohm@1147
   578
clasohm@1147
   579
          (*convert tags in productions*)
clasohm@1147
   580
          fun process_prods [] result = result
clasohm@1147
   581
            | process_prods ((rhs, id, prec) :: ps) result =
clasohm@1147
   582
                process_prods ps ((lhs_tag, (process_rhs rhs [], id, prec))
clasohm@1147
   583
                                  :: result);
clasohm@1147
   584
        in process_nt (nt-1) (process_prods nt_prods [] @ result) end;
clasohm@1147
   585
clasohm@1147
   586
    val raw_prods = chain_prods @ process_nt (nt_count2-1) [];
clasohm@1147
   587
clasohm@1147
   588
    val prods1' =
clasohm@1147
   589
      let fun get_prod i = if i < nt_count1 then Array.sub (prods1, i)
clasohm@1147
   590
                           else (([], []), []);
clasohm@1147
   591
      in Array.tabulate (nt_count1', get_prod) end;
clasohm@1147
   592
clasohm@1147
   593
    val fromto_chains = inverse_chains chains1 [];
clasohm@1147
   594
skalberg@15531
   595
    val (fromto_chains', lambdas', SOME prod_count1') =
skalberg@15531
   596
      add_prods prods1' fromto_chains lambdas1 (SOME prod_count1) raw_prods;
clasohm@1147
   597
clasohm@1147
   598
    val chains' = inverse_chains fromto_chains' [];
clasohm@1147
   599
  in Gram {nt_count = nt_count1', prod_count = prod_count1',
clasohm@1147
   600
           tags = tags1', chains = chains', lambdas = lambdas',
clasohm@1147
   601
           prods = prods1'}
clasohm@1147
   602
  end;
clasohm@1147
   603
wenzelm@18
   604
clasohm@1438
   605
(** Parser **)
wenzelm@18
   606
wenzelm@237
   607
datatype parsetree =
wenzelm@237
   608
  Node of string * parsetree list |
wenzelm@237
   609
  Tip of token;
wenzelm@237
   610
wenzelm@18
   611
type state =
clasohm@1147
   612
  nt_tag * int *                (*identification and production precedence*)
clasohm@1147
   613
  parsetree list *              (*already parsed nonterminals on rhs*)
clasohm@1147
   614
  symb list *                   (*rest of rhs*)
clasohm@1147
   615
  string *                      (*name of production*)
clasohm@1147
   616
  int;                          (*index for previous state list*)
wenzelm@18
   617
wenzelm@18
   618
clasohm@330
   619
(*Get all rhss with precedence >= minPrec*)
skalberg@15570
   620
fun getRHS minPrec = List.filter (fn (_, _, prec:int) => prec >= minPrec);
wenzelm@237
   621
clasohm@330
   622
(*Get all rhss with precedence >= minPrec and < maxPrec*)
clasohm@330
   623
fun getRHS' minPrec maxPrec =
skalberg@15570
   624
  List.filter (fn (_, _, prec:int) => prec >= minPrec andalso prec < maxPrec);
wenzelm@18
   625
clasohm@330
   626
(*Make states using a list of rhss*)
clasohm@330
   627
fun mkStates i minPrec lhsID rhss =
clasohm@330
   628
  let fun mkState (rhs, id, prodPrec) = (lhsID, prodPrec, [], rhs, id, i);
clasohm@330
   629
  in map mkState rhss end;
clasohm@697
   630
wenzelm@15752
   631
(*Add parse tree to list and eliminate duplicates
clasohm@330
   632
  saving the maximum precedence*)
wenzelm@16668
   633
fun conc (t: parsetree list, prec:int) [] = (NONE, [(t, prec)])
clasohm@330
   634
  | conc (t, prec) ((t', prec') :: ts) =
clasohm@330
   635
      if t = t' then
wenzelm@15752
   636
        (SOME prec', if prec' >= prec then (t', prec') :: ts
clasohm@330
   637
                     else (t, prec) :: ts)
clasohm@330
   638
      else
clasohm@330
   639
        let val (n, ts') = conc (t, prec) ts
clasohm@330
   640
        in (n, (t', prec') :: ts') end;
wenzelm@18
   641
clasohm@330
   642
(*Update entry in used*)
wenzelm@16668
   643
fun update_trees ((B: nt_tag, (i, ts)) :: used) (A, t) =
wenzelm@237
   644
  if A = B then
wenzelm@237
   645
    let val (n, ts') = conc t ts
wenzelm@237
   646
    in ((A, (i, ts')) :: used, n) end
wenzelm@237
   647
  else
clasohm@697
   648
    let val (used', n) = update_trees used (A, t)
wenzelm@237
   649
    in ((B, (i, ts)) :: used', n) end;
wenzelm@18
   650
clasohm@330
   651
(*Replace entry in used*)
wenzelm@16668
   652
fun update_prec (A: nt_tag, prec) used =
clasohm@697
   653
  let fun update ((hd as (B, (_, ts))) :: used, used') =
clasohm@330
   654
        if A = B
clasohm@330
   655
        then used' @ ((A, (prec, ts)) :: used)
clasohm@330
   656
        else update (used, hd :: used')
clasohm@330
   657
  in update (used, []) end;
wenzelm@18
   658
clasohm@330
   659
fun getS A maxPrec Si =
skalberg@15570
   660
  List.filter
clasohm@1147
   661
    (fn (_, _, _, Nonterminal (B, prec) :: _, _, _)
clasohm@330
   662
          => A = B andalso prec <= maxPrec
wenzelm@237
   663
      | _ => false) Si;
wenzelm@18
   664
clasohm@330
   665
fun getS' A maxPrec minPrec Si =
skalberg@15570
   666
  List.filter
clasohm@1147
   667
    (fn (_, _, _, Nonterminal (B, prec) :: _, _, _)
clasohm@330
   668
          => A = B andalso prec > minPrec andalso prec <= maxPrec
wenzelm@237
   669
      | _ => false) Si;
wenzelm@18
   670
clasohm@330
   671
fun getStates Estate i ii A maxPrec =
skalberg@15570
   672
  List.filter
clasohm@1147
   673
    (fn (_, _, _, Nonterminal (B, prec) :: _, _, _)
clasohm@330
   674
          => A = B andalso prec <= maxPrec
wenzelm@237
   675
      | _ => false)
wenzelm@237
   676
    (Array.sub (Estate, ii));
wenzelm@18
   677
wenzelm@18
   678
clasohm@1147
   679
fun movedot_term (A, j, ts, Terminal a :: sa, id, i) c =
wenzelm@237
   680
  if valued_token c then
clasohm@697
   681
    (A, j, ts @ [Tip c], sa, id, i)
wenzelm@237
   682
  else (A, j, ts, sa, id, i);
wenzelm@18
   683
clasohm@1147
   684
fun movedot_nonterm ts (A, j, tss, Nonterminal _ :: sa, id, i) =
wenzelm@237
   685
  (A, j, tss @ ts, sa, id, i);
wenzelm@18
   686
wenzelm@237
   687
fun movedot_lambda _ [] = []
clasohm@1147
   688
  | movedot_lambda (B, j, tss, Nonterminal (A, k) :: sa, id, i) ((t, ki) :: ts) =
wenzelm@237
   689
      if k <= ki then
wenzelm@237
   690
        (B, j, tss @ t, sa, id, i) ::
clasohm@1147
   691
          movedot_lambda (B, j, tss, Nonterminal (A, k) :: sa, id, i) ts
clasohm@1147
   692
      else movedot_lambda (B, j, tss, Nonterminal (A, k) :: sa, id, i) ts;
wenzelm@18
   693
wenzelm@18
   694
kleing@13057
   695
val branching_level = ref 600;                   (*trigger value for warnings*)
wenzelm@18
   696
clasohm@1147
   697
(*get all productions of a NT and NTs chained to it which can
clasohm@1147
   698
  be started by specified token*)
clasohm@1147
   699
fun prods_for prods chains include_none tk nts =
wenzelm@15979
   700
let
wenzelm@15979
   701
    fun token_assoc (list, key) =
clasohm@1147
   702
      let fun assoc [] result = result
clasohm@1147
   703
            | assoc ((keyi, pi) :: pairs) result =
wenzelm@15979
   704
                if is_some keyi andalso matching_tokens (the keyi, key)
wenzelm@15752
   705
                   orelse include_none andalso is_none keyi then
clasohm@1147
   706
                  assoc pairs (pi @ result)
clasohm@1147
   707
                else assoc pairs result;
clasohm@1147
   708
      in assoc list [] end;
clasohm@1147
   709
clasohm@1147
   710
    fun get_prods [] result = result
clasohm@1147
   711
      | get_prods (nt :: nts) result =
clasohm@1147
   712
        let val nt_prods = snd (Array.sub (prods, nt));
wenzelm@15979
   713
        in get_prods nts ((token_assoc (nt_prods, tk)) @ result) end;
clasohm@1147
   714
in get_prods (connected_with chains nts nts) [] end;
clasohm@1147
   715
clasohm@1147
   716
wenzelm@23941
   717
fun PROCESSS warned prods chains Estate i c states =
wenzelm@18
   718
let
clasohm@1147
   719
fun all_prods_for nt = prods_for prods chains true c [nt];
clasohm@330
   720
wenzelm@237
   721
fun processS used [] (Si, Sii) = (Si, Sii)
wenzelm@237
   722
  | processS used (S :: States) (Si, Sii) =
wenzelm@237
   723
      (case S of
clasohm@1147
   724
        (_, _, _, Nonterminal (nt, minPrec) :: _, _, _) =>
clasohm@330
   725
          let                                       (*predictor operation*)
clasohm@697
   726
            val (used', new_states) =
haftmann@17314
   727
              (case AList.lookup (op =) used nt of
skalberg@15531
   728
                SOME (usedPrec, l) =>       (*nonterminal has been processed*)
clasohm@330
   729
                  if usedPrec <= minPrec then
clasohm@330
   730
                                      (*wanted precedence has been processed*)
clasohm@330
   731
                    (used, movedot_lambda S l)
clasohm@330
   732
                  else            (*wanted precedence hasn't been parsed yet*)
clasohm@1147
   733
                    let
clasohm@1147
   734
                      val tk_prods = all_prods_for nt;
wenzelm@15752
   735
clasohm@1147
   736
                      val States' = mkStates i minPrec nt
clasohm@1147
   737
                                      (getRHS' minPrec usedPrec tk_prods);
wenzelm@15752
   738
                    in (update_prec (nt, minPrec) used,
clasohm@330
   739
                        movedot_lambda S l @ States')
wenzelm@237
   740
                    end
wenzelm@18
   741
skalberg@15531
   742
              | NONE =>           (*nonterminal is parsed for the first time*)
clasohm@1147
   743
                  let val tk_prods = all_prods_for nt;
clasohm@1147
   744
                      val States' = mkStates i minPrec nt
clasohm@1147
   745
                                      (getRHS minPrec tk_prods);
clasohm@1147
   746
                  in ((nt, (minPrec, [])) :: used, States') end);
clasohm@697
   747
clasohm@1147
   748
            val dummy =
clasohm@1147
   749
              if not (!warned) andalso
clasohm@1147
   750
                 length (new_states @ States) > (!branching_level) then
wenzelm@19305
   751
                (warning "Currently parsed expression could be extremely ambiguous.";
clasohm@1147
   752
                 warned := true)
clasohm@1147
   753
              else ();
wenzelm@18
   754
          in
clasohm@697
   755
            processS used' (new_states @ States) (S :: Si, Sii)
wenzelm@18
   756
          end
clasohm@1147
   757
      | (_, _, _, Terminal a :: _, _, _) =>               (*scanner operation*)
wenzelm@237
   758
          processS used States
wenzelm@237
   759
            (S :: Si,
wenzelm@237
   760
              if matching_tokens (a, c) then movedot_term S c :: Sii else Sii)
clasohm@330
   761
      | (A, prec, ts, [], id, j) =>                   (*completer operation*)
clasohm@697
   762
          let val tt = if id = "" then ts else [Node (id, ts)] in
clasohm@330
   763
            if j = i then                             (*lambda production?*)
wenzelm@237
   764
              let
clasohm@697
   765
                val (used', O) = update_trees used (A, (tt, prec));
wenzelm@237
   766
              in
clasohm@1147
   767
                case O of
skalberg@15531
   768
                  NONE =>
clasohm@1147
   769
                    let val Slist = getS A prec Si;
clasohm@1147
   770
                        val States' = map (movedot_nonterm tt) Slist;
clasohm@1147
   771
                    in processS used' (States' @ States) (S :: Si, Sii) end
skalberg@15531
   772
                | SOME n =>
clasohm@1147
   773
                    if n >= prec then processS used' States (S :: Si, Sii)
wenzelm@237
   774
                    else
clasohm@1147
   775
                      let val Slist = getS' A prec n Si;
clasohm@1147
   776
                          val States' = map (movedot_nonterm tt) Slist;
clasohm@1147
   777
                      in processS used' (States' @ States) (S :: Si, Sii) end
wenzelm@15752
   778
              end
wenzelm@237
   779
            else
clasohm@1147
   780
              let val Slist = getStates Estate i j A prec
clasohm@1147
   781
              in processS used (map (movedot_nonterm tt) Slist @ States)
clasohm@1147
   782
                          (S :: Si, Sii)
clasohm@682
   783
              end
wenzelm@237
   784
          end)
clasohm@682
   785
in processS [] states ([], []) end;
wenzelm@18
   786
wenzelm@18
   787
wenzelm@25986
   788
fun produce warned prods tags chains stateset i indata prev_token =
wenzelm@237
   789
  (case Array.sub (stateset, i) of
wenzelm@25986
   790
    [] =>
wenzelm@25986
   791
      let
wenzelm@25986
   792
        val toks =
wenzelm@25986
   793
          if prev_token = EndToken then indata
wenzelm@25986
   794
          else prev_token :: indata;
clasohm@373
   795
wenzelm@25986
   796
        val nt_name = reverted_tags tags;
wenzelm@25986
   797
        val nts =
wenzelm@25986
   798
          fold (fn (_, _, _, Nonterminal nt :: _, _, _) => insert (op =) nt | _ => I)
wenzelm@25986
   799
            (Array.sub (stateset, i - 1)) []
wenzelm@25986
   800
          |> map (fn (a, prec) => nt_name a ^ "(" ^ signed_string_of_int prec ^ ")");
clasohm@373
   801
wenzelm@25986
   802
        val msg =
wenzelm@25986
   803
          (if null toks then Pretty.str "Inner syntax error: unexpected end of input"
wenzelm@25986
   804
           else
wenzelm@25986
   805
             Pretty.block (Pretty.str "Inner syntax error at: \"" ::
wenzelm@25986
   806
               Pretty.breaks (map (Pretty.str o str_of_token) (#1 (split_last toks))) @
wenzelm@25986
   807
               [Pretty.str "\""])) ::
wenzelm@25986
   808
          (if null nts then []
wenzelm@25986
   809
           else [Pretty.block
wenzelm@25986
   810
            (Pretty.str "Expected syntax categories: " :: Pretty.commas (map Pretty.str nts))]);
wenzelm@25986
   811
      in error (Pretty.string_of (Pretty.chunks msg)) end
wenzelm@237
   812
  | s =>
wenzelm@237
   813
    (case indata of
clasohm@1147
   814
       [] => Array.sub (stateset, i)
clasohm@1147
   815
     | c :: cs =>
wenzelm@23941
   816
       let val (si, sii) = PROCESSS warned prods chains stateset i c s;
clasohm@1147
   817
       in Array.update (stateset, i, si);
clasohm@1147
   818
          Array.update (stateset, i + 1, sii);
wenzelm@25986
   819
          produce warned prods tags chains stateset (i + 1) cs c
clasohm@1147
   820
       end));
wenzelm@18
   821
wenzelm@18
   822
wenzelm@19482
   823
fun get_trees l = map_filter (fn (_, _, [pt], _, _, _) => SOME pt | _ => NONE)
paulson@2229
   824
                            l;
wenzelm@18
   825
clasohm@1147
   826
fun earley prods tags chains startsymbol indata =
wenzelm@237
   827
  let
wenzelm@17412
   828
    val start_tag = case Symtab.lookup tags startsymbol of
skalberg@15531
   829
                       SOME tag => tag
wenzelm@15752
   830
                     | NONE   => error ("parse: Unknown startsymbol " ^
clasohm@624
   831
                                        quote startsymbol);
clasohm@1147
   832
    val S0 = [(~1, 0, [], [Nonterminal (start_tag, 0), Terminal EndToken],
clasohm@1147
   833
               "", 0)];
clasohm@330
   834
    val s = length indata + 1;
wenzelm@237
   835
    val Estate = Array.array (s, []);
wenzelm@237
   836
  in
wenzelm@237
   837
    Array.update (Estate, 0, S0);
wenzelm@25986
   838
    get_trees (produce (ref false) prods tags chains Estate 0 indata EndToken)
wenzelm@237
   839
  end;
wenzelm@18
   840
wenzelm@18
   841
clasohm@1147
   842
fun parse (Gram {tags, prods, chains, ...}) start toks =
clasohm@624
   843
let val r =
wenzelm@24245
   844
  (case earley prods tags chains start (toks @ [Lexicon.EndToken]) of
wenzelm@237
   845
    [] => sys_error "parse: no parse trees"
clasohm@330
   846
  | pts => pts);
clasohm@624
   847
in r end
wenzelm@18
   848
wenzelm@18
   849
end;