engineTypes.ml 11.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
(* This file defines several types and module types that are used in the
   specification of module [Engine]. *)

(* --------------------------------------------------------------------------- *)

(* It would be nice if we could keep the structure of stacks and environments
   hidden. However, stacks and environments must be accessible to semantic
   actions, so the following data structure definitions must be public. *)

(* --------------------------------------------------------------------------- *)

(* A stack is a linked list of cells. A sentinel cell -- which is its own
   successor -- is used to mark the bottom of the stack. The sentinel cell
   itself is not significant -- it contains dummy values. *)

type ('state, 'semantic_value) stack = {

  (* The state that we should go back to if we pop this stack cell. *)

  (* This convention means that the state contained in the top stack cell is
     not the current state [env.current]. It also means that the state found
     within the sentinel is a dummy -- it is never consulted. This convention
     is the same as that adopted by the code-based back-end. *)

  state: 'state;

  (* The semantic value associated with the chunk of input that this cell
     represents. *)

  semv: 'semantic_value;

  (* The start and end positions of the chunk of input that this cell
     represents. *)

  startp: Lexing.position;
  endp: Lexing.position;

  (* The next cell down in the stack. If this is a self-pointer, then this
     cell is the sentinel, and the stack is conceptually empty. *)

  next: ('state, 'semantic_value) stack;

}

(* --------------------------------------------------------------------------- *)

47 48
(* A parsing environment contains all of the parser's state (except for the
   current program point). *)
49 50 51

type ('state, 'semantic_value, 'token) env = {

52 53 54 55 56 57
  (* If this flag is true, then the first component of [env.triple] should
     be ignored, as it has been logically overwritten with the [error]
     pseudo-token. *)

  error: bool;

58
  (* The last token that was obtained from the lexer, together with its start
59 60
     and end positions. Warning: before the first call to the lexer has taken
     place, a dummy (and possibly invalid) token is stored here. *)
61

62
  triple: 'token * Lexing.position * Lexing.position;
63 64 65 66

  (* The stack. In [CodeBackend], it is passed around on its own,
     whereas, here, it is accessed via the environment. *)

67
  stack: ('state, 'semantic_value) stack;
68 69 70 71

  (* The current state. In [CodeBackend], it is passed around on its
     own, whereas, here, it is accessed via the environment. *)

72
  current: 'state;
73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191

}

(* --------------------------------------------------------------------------- *)

(* This signature describes the parameters that must be supplied to the LR
   engine. *)

module type TABLE = sig

  (* The type of automaton states. *)

  type state

  (* The type of tokens. These can be thought of as real tokens, that is,
     tokens returned by the lexer. They carry a semantic value. This type
     does not include the [error] pseudo-token. *)

  type token

  (* The type of terminal symbols. These can be thought of as integer codes.
     They do not carry a semantic value. This type does include the [error]
     pseudo-token. *)

  type terminal

  (* The type of semantic values. *)

  type semantic_value

  (* A token is conceptually a pair of a (non-[error]) terminal symbol and
     a semantic value. The following two functions are the pair projections. *)

  val token2terminal: token -> terminal
  val token2value: token -> semantic_value

  (* Even though the [error] pseudo-token is not a real token, it is a
     terminal symbol. Furthermore, for regularity, it must have a semantic
     value. *)

  val error_terminal: terminal
  val error_value: semantic_value

  (* The type of productions. *)

  type production

  (* If a state [s] has a default reduction on production [prod], then, upon
     entering [s], the automaton should reduce [prod] without consulting the
     lookahead token. The following function allows determining which states
     have default reductions. *)

  (* Instead of returning a value of a sum type -- either [DefRed prod], or
     [NoDefRed] -- it accepts two continuations, and invokes just one of
     them. This mechanism allows avoiding a memory allocation. *)

  val default_reduction:
    state ->
    ('env -> production -> 'answer) ->
    ('env -> 'answer) ->
    'env -> 'answer

  (* An LR automaton can normally take three kinds of actions: shift, reduce,
     or fail. (Acceptance is a particular case of reduction: it consists in
     reducing a start production.) *)

  (* There are two variants of the shift action. [shift/discard s] instructs
     the automaton to discard the current token, request a new one from the
     lexer, and move to state [s]. [shift/nodiscard s] instructs it to move to
     state [s] without requesting a new token. This instruction should be used
     when [s] has a default reduction on [#]. See [CodeBackend.gettoken] for
     details. *)

  (* This is the automaton's action table. It maps a pair of a state and a
     terminal symbol to an action. *)

  (* Instead of returning a value of a sum type -- one of shift/discard,
     shift/nodiscard, reduce, or fail -- this function accepts three
     continuations, and invokes just one them. This mechanism allows avoiding
     a memory allocation. *)

  (* In summary, the parameters to [action] are as follows:

     - the first two parameters, a state and a terminal symbol, are used to
       look up the action table;

     - the next parameter is the semantic value associated with the above
       terminal symbol; it is not used, only passed along to the shift
       continuation, as explained below;

     - the shift continuation expects an environment; a flag that tells
       whether to discard the current token; the terminal symbol that
       is being shifted; its semantic value; and the target state of
       the transition;

     - the reduce continuation expects an environment and a production;

     - the fail continuation expects an environment;

     - the last parameter is the environment; it is not used, only passed
       along to the selected continuation. *)

  val action:
    state ->
    terminal ->
    semantic_value ->
    ('env -> bool -> terminal -> semantic_value -> state -> 'answer) ->
    ('env -> production -> 'answer) ->
    ('env -> 'answer) ->
    'env -> 'answer

  (* This is the automaton's goto table. It maps a pair of a state and a
     production to a new state.

     This convention is slightly different from the textbook approach. The
     goto table is usually indexed by a state and a non-terminal symbol. *)

  val goto: state -> production -> state

192 193 194 195
  (* [is_start prod] tells whether the production [prod] is a start production. *)

  val is_start: production -> bool

196 197 198 199 200
  (* By convention, a semantic action is responsible for:

     1. fetching whatever semantic values and positions it needs off the stack;

     2. popping an appropriate number of cells off the stack, as dictated
201
        by the length of the right-hand side of the production;
202 203 204 205

     3. computing a new semantic value, as well as new start and end positions;

     4. pushing a new stack cell, which contains the three values
206 207 208 209 210
        computed in step 3;

     5. returning the new stack computed in steps 2 and 4. The environment
        is not affected: the caller of the semantic action is responsible
        for writing the new stack into [env.stack].
211 212 213 214 215 216 217 218 219 220 221

     Point 1 is essentially forced upon us: if semantic values were fetched
     off the stack by this interpreter, then the calling convention for
     semantic actions would be variadic: not all semantic actions would have
     the same number of arguments. The rest follows rather naturally. *)

  (* Semantic actions are allowed to raise [Error]. *)

  exception Error

  type semantic_action =
222
      (state, semantic_value, token) env -> (state, semantic_value) stack
223 224 225 226 227 228 229 230

  val semantic_action: production -> semantic_action

  (* The LR engine requires a number of hooks, which are used for logging. *)

  (* The comments below indicate the conventional messages that correspond
     to these hooks in the code-based back-end; see [CodeBackend]. *)

231 232 233 234 235 236 237
  (* If the flag [log] is false, then the logging functions are guaranteed
     to do nothing, so it is not necessary to call them. If [log] is true,
     the logging functions may or may not have an effect. This flag is
     logically superfluous, but saves time in the table-based back-end. *)

  val log : bool

238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
  module Log : sig

    (* State %d: *)

    val state: state -> unit

    (* Shifting (<terminal>) to state <state> *)

    val shift: terminal -> state -> unit

    (* Reducing a production should be logged either as a reduction
       event (for regular productions) or as an acceptance event (for
       start productions). *)

    (* Reducing production <production> / Accepting *)

    val reduce_or_accept: production -> unit

    (* Lookahead token is now <terminal> (<pos>-<pos>) *)

258
    val lookahead_token: terminal -> Lexing.position -> Lexing.position -> unit
259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277

    (* Initiating error handling *)

    val initiating_error_handling: unit -> unit

    (* Resuming error handling *)

    val resuming_error_handling: unit -> unit

    (* Handling error in state <state> *)

    val handling_error: state -> unit

  end

end

(* --------------------------------------------------------------------------- *)

278
(* This signature describes the monolithic (traditional) LR engine. *)
279

280 281 282
(* In this interface, the parser controls the lexer. *)

module type MONOLITHIC_ENGINE = sig
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301

  type state

  type token

  type semantic_value

  (* An entry point to the engine requires a start state, a lexer, and a lexing
     buffer. It either succeeds and produces a semantic value, or fails and
     raises [Error]. *)

  exception Error

  val entry:
    state ->
    (Lexing.lexbuf -> token) ->
    Lexing.lexbuf ->
    semantic_value

302 303 304 305
end

(* --------------------------------------------------------------------------- *)

306
(* The following signatures describe the incremental LR engine. *)
307

308
(* First, see [INCREMENTAL_ENGINE] in the file [IncrementalEngine.ml]. *)
309

310 311 312 313
(* The [start] function is set apart because we do not wish to publish
   it as part of the generated [parser.mli] file. Instead, the table
   back-end will publish specialized versions of it, with a suitable
   type cast. *)
314

315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
module type INCREMENTAL_ENGINE_START = sig

  (* [start] is an entry point. It requires just a start state, and begins
     the parsing process. It produces a result, which usually will be an
     [InputNeeded] result. (It could be [Accepted] if this starting state
     accepts only the empty word. It could be [Rejected] if this starting
     state accepts no word at all.) It does not raise any exception. *)

  (* [start s] should really produce a result of type ['a result], for a
     fixed ['a] that depends on the state [s]. We cannot express this, so
     we use [semantic_value result], which is safe. The table back-end
     uses [Obj.magic] to produce safe specialized versions of [start]. *)

  type state
  type semantic_value
  type 'a result

  val start:
    state ->
    semantic_value result

end

338 339 340 341 342 343 344 345
(* --------------------------------------------------------------------------- *)

(* This signature describes the LR engine, which combines the monolithic
   and incremental interfaces. *)

module type ENGINE = sig

  include MONOLITHIC_ENGINE
346

347
  include IncrementalEngine.INCREMENTAL_ENGINE
348
    with type token := token
349
     and type 'a lr1state = state (* useful for us; hidden from the end user *)
350 351

  include INCREMENTAL_ENGINE_START
352 353
    with type state := state
     and type semantic_value := semantic_value
354
     and type 'a result := 'a result
355 356

end
357