interpret.ml 24.8 KB
Newer Older
1 2 3 4 5 6 7
(* This module is in charge of handling the [--interpret] option,
   if it is present. *)

module I = Invariant (* artificial dependency; ensures that [Invariant] runs first *)

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

8
open Grammar
9 10
open SentenceParserAux

11 12 13 14 15
(* A delimiter. *)

type delimiter =
  string

16
(* An error message. *)
17

18 19
type message =
  string
20

21 22 23
(* A run is a series of sentences or comments,
   followed with a delimiter (at least one blank line; comments),
   followed with an error message. *)
24 25

type run =
26 27 28
  located_sentence or_comment list *
  delimiter *
  message
29

30 31 32 33 34 35 36 37 38
(* A targeted sentence is a located sentence together with the target into
   which it leads. A target tells us which state a sentence leads to, as well
   as which spurious reductions are performed at the end. *)

type target =
  ReferenceInterpreter.target

let target2state (s, _spurious) =
  s
39

40
type maybe_targeted_sentence =
41
  located_sentence * target option
42

43
type targeted_sentence =
44
  located_sentence * target
45 46 47

(* A targeted run is a series of targeted sentences or comments together with
   an error message. *)
48

49
type maybe_targeted_run =
50 51 52
  maybe_targeted_sentence or_comment list *
  delimiter *
  message
53

54
type targeted_run =
55 56 57
  targeted_sentence or_comment list *
  delimiter *
  message
58 59 60 61 62

(* A filtered targeted run is a series of targeted sentences together with an
   error message. (The comments have been filtered out.) *)

type filtered_targeted_run =
63 64
  targeted_sentence list *
  message
65 66

(* --------------------------------------------------------------------------- *)
POTTIER Francois's avatar
POTTIER Francois committed
67

68
(* Display and debugging. *)
69 70 71 72

let print_sentence (nto, terminals) : string =
  let b = Buffer.create 128 in
  Option.iter (fun nt ->
73
    Printf.bprintf b "%s: " (Nonterminal.print false nt)
74 75 76 77 78 79 80
  ) nto;
  List.iter (fun t ->
    Printf.bprintf b "%s " (Terminal.print t)
  ) terminals;
  Printf.bprintf b "\n";
  Buffer.contents b

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

(* [stream] turns a finite list of terminals into a stream of terminals. *)

exception EndOfStream

let stream (toks : Terminal.t list) : unit -> Terminal.t * Lexing.position * Lexing.position =
  let toks = ref toks in
  fun () ->

    let tok =
      match !toks with
      | tok :: more ->

	  (* Take a token off the list, and return it. *)

	  toks := more;
	  tok

      | [] ->

	  (* The finite list has been exhausted. Here, two plausible behaviors
	     come to mind.

	     The first behavior consists in raising an exception. In that case,
	     we are creating a finite stream, and it is up to the parser to not
	     read past its end.

	     The second behavior consists in returning a designated token. In
	     that case, we are creating an infinite, eventually constant,
	     stream.

	     The choice between these two behaviors is somewhat arbitrary;
	     furthermore, in the second case, the choice of the designated
	     token is arbitrary as well. Here, we adopt the second behavior if
	     and only if the grammar has an EOF token, and we use EOF as the
	     designated token. Again, this is arbitrary, and could be changed
	     in the future. *)

	  match Terminal.eof with
	  | Some eof ->
	      eof
	  | None ->
	      raise EndOfStream

    in

    (* For now, return dummy positions. *)

    tok, Lexing.dummy_pos, Lexing.dummy_pos

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

134 135 136 137 138 139 140
(* [start sentence] returns the start symbol that we should use to interpret
   the sentence [sentence]. *)

(* If a start symbol was explicitly provided as part of the sentence, we use
   it. Otherwise, we use the grammar's unique start symbol, if there is
   one. *)

141
let start poss ((nto, _) : sentence) : Nonterminal.t =
142 143 144 145 146 147
  match nto with
  | Some nt ->
      nt
  | None ->
      match ProductionMap.is_singleton Lr1.entry with
      | None ->
148
          Error.error poss
149 150 151 152 153 154 155 156 157
            "Because the grammar has multiple start symbols, each of the\n\
             sentences provided on the standard input channel must be of the\n\
             form: <start symbol>: <token>*"
      | Some (prod, _) ->
          match Production.classify prod with
          | Some nt ->
              nt
          | None ->
              assert false
158

159
(* --------------------------------------------------------------------------- *)
160

161
(* [interpret] interprets a sentence. *)
162

163
let interpret ((_, toks) as sentence) : unit =
164

165
  let nt = start [] sentence in
166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208

  (* Run the reference interpreter. This can produce a concrete syntax tree
     ([Some cst]), fail with a parser error ([None]), or fail with a lexer error
     ([EndOfStream]). *)

  (* In either case, we produce just one line of output, so it should be clear
     to the user which outcomes correspond to which sentences (should multiple
     sentences be supplied). *)

  begin try
    match
      MenhirLib.Convert.Simplified.traditional2revised
	(ReferenceInterpreter.interpret Settings.trace nt)
	(stream toks)
    with

    | Some cst ->

	(* Success. *)

	Printf.printf "ACCEPT";
	if Settings.interpret_show_cst then begin
	  print_newline();
	  Cst.show stdout cst
	end

    | None ->

	(* Parser failure. *)

	Printf.printf "REJECT"

  with EndOfStream ->

    (* Lexer failure. *)
    
    Printf.printf "OVERSHOOT"

  end;
  print_newline()

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

209 210
(* [interpret_error_aux] interprets a sentence, expecting it to end in an
   error. Failure or success is reported via two continuations. *)
211

212 213
let interpret_error_aux poss ((_, terminals) as sentence) fail succeed =
  let nt = start poss sentence in
214
  let open ReferenceInterpreter in
215
  match check_error_path nt terminals with
216
  | OInputReadPastEnd ->
217
      fail "No syntax error occurs."
218
  | OInputNotFullyConsumed ->
219
      fail "A syntax error occurs before the last token is reached."
220
  | OUnexpectedAccept ->
221
      fail "No syntax error occurs; in fact, this input is accepted."
222 223
  | OK target ->
      succeed nt terminals target
224 225 226 227 228 229 230 231 232 233

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

(* This default error message is produced by [--list-errors] when it creates a
   [.messages] file, and is recognized by [--compare-errors] when it compares
   two such files. *)

let default_message =
  "<YOUR SYNTAX ERROR MESSAGE HERE>\n"

234
(* [print_messages_auto] displays just the sentence and the auto-generated
235
   comments. [otarget] may be [None], in which case the auto-generated comment
236 237
   is just a warning that this sentence does not end in an error. *)

238
let print_messages_auto (nt, sentence, otarget) : unit =
239 240
  (* Print the sentence, followed with auto-generated comments. *)
  print_string (print_sentence (Some nt, sentence));
241
  match (otarget : target option) with
242 243 244 245 246 247
  | None ->
      Printf.printf
        "##\n\
         ## WARNING: This sentence does NOT end with a syntax error, as it should.\n\
         ##\n"
  | Some (s', spurious) ->
248
      Printf.printf
249 250 251 252
        "##\n\
         ## Ends in an error in state: %d.\n\
         ##\n\
         %s##\n"
253 254 255 256
        (Lr1.number s')
        (* [Lr0.print] or [Lr0.print_closure] could be used here. The latter
           could sometimes be helpful, but is usually intolerably verbose. *)
        (Lr0.print "## " (Lr1.state s'))
257
      ;
258 259 260 261 262 263
      Printf.printf
        "## The known suffix of the stack is as follows:\n\
         ## %s\n\
         ##\n"
        (Invariant.print (Invariant.stack s'))
      ;
264 265 266 267 268 269 270
      if spurious <> [] then begin
        Printf.printf
          "## WARNING: This example involves spurious reductions.\n\
           ## This implies that, although the LR(1) items shown above provide an\n\
           ## accurate view of the past (what has been recognized so far), they\n\
           ## may provide an INCOMPLETE view of the future (what was expected next).\n"
        ;
271
        List.iter (fun (s, prod) ->
272
          Printf.printf
273 274
            "## In state %d, spurious reduction of production %s\n"
            (Lr1.number s)
275 276 277 278
            (Production.print prod)
        ) spurious;
        Printf.printf "##\n"
      end
279

280
(* [print_messages_item] displays one data item. The item is of the form [nt,
281 282 283 284
   sentence, target], which means that beginning at the start symbol [nt], the
   sentence [sentence] ends in an error in the target state given by [target].
   [target] also contains information about which spurious reductions are
   performed at the end. The display obeys the [.messages] file format. *)
285

286
let print_messages_item (nt, sentence, target) : unit =
287
  (* Print the sentence, followed with auto-generated comments. *)
288
  print_messages_auto (nt, sentence, Some target);
289 290 291 292 293
  (* Then, print a proposed error message, between two blank lines. *)
  Printf.printf "\n%s\n" default_message

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

294 295 296 297 298
(* [write_run run] writes a run into a new [.messages] file. Manually-written
   comments are preserved. New auto-generated comments are produced. *)

let write_run : maybe_targeted_run or_comment -> unit =
  function
299
  | Thing (sentences_or_comments, delimiter, message) ->
300 301 302 303 304 305 306 307 308 309
      (* First, print every sentence and human comment. *)
      List.iter (fun sentence_or_comment ->
        match sentence_or_comment with
        | Thing ((poss, ((_, toks) as sentence)), target) ->
            let nt = start poss sentence in
            (* Every sentence is followed with newly generated auto-comments. *)
            print_messages_auto (nt, toks, target)
        | Comment c ->
            print_string c
      ) sentences_or_comments;
310 311 312 313 314 315 316
      (* Then, print the delimiter, which must begin with a blank line
         and may include comments. *)
      print_string delimiter;
      (* Then, print the error message. *)
      print_string message
      (* No need for another blank line. It will be printed as part of a
         separate [Comment]. *)
317
  | Comment comments ->
318
      (* Must begin with a blank line. *)
319
      print_string comments
320

321
(* --------------------------------------------------------------------------- *)
322

323 324 325 326 327
(* [interpret_error] interprets a sentence, expecting it to end in an error.
   Failure or success is reported on the standard output channel. This is
   used by [--interpret-error]. *)

let fail msg =
328
  Error.error [] msg
329

330 331
let succeed nt terminals target =
  print_messages_item (nt, terminals, target);
332 333 334 335 336 337 338
  exit 0

let interpret_error sentence =
  interpret_error_aux [] sentence fail succeed

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

339 340 341 342
(* [target_sentence] interprets a (located) sentence, expecting it to end in
   an error, computes the state in which the error is obtained, and constructs
   a targeted sentence. *)

343
let target_sentence signal : located_sentence -> maybe_targeted_sentence =
344
  fun (poss, sentence) ->
345
    (poss, sentence),
346
    interpret_error_aux poss sentence
347 348 349 350 351 352 353 354 355
      (* failure: *)
      (fun msg ->
        signal poss (Printf.sprintf
          "This sentence does not end with a syntax error, as it should.\n%s"
          msg
        );
        None
      )
      (* success: *)
356
      (fun _nt _terminals target -> Some target)
357 358

let target_run_1 signal : run -> maybe_targeted_run =
359 360 361 362
  fun (sentences, delimiter, message) ->
    List.map (or_comment_map (target_sentence signal)) sentences,
    delimiter,
    message
363

364
let target_run_2 : maybe_targeted_run -> targeted_run =
365
  fun (sentences, delimiter, message) ->
366
    let aux (x, y) = (x, Misc.unSome y) in
367 368 369
    List.map (or_comment_map aux) sentences,
    delimiter,
    message
370 371 372

let target_runs : run list -> targeted_run list =
  fun runs ->
373
    (* Interpret all sentences, possibly displaying multiple errors. *)
374
    let runs = List.map (target_run_1 Error.signal) runs in
375
    (* Abort if an error occurred. *)
376
    if Error.errors() then exit 1;
377 378
    (* Remove the options introduced by the first phase above. *)
    let runs = List.map target_run_2 runs in
379
    runs
380

381
(* --------------------------------------------------------------------------- *)
382

383 384 385 386 387 388
(* [filter_things] filters out the comments in a list of things or comments. *)

let filter_things : 'a or_comment list -> 'a list =
  fun things -> List.flatten (List.map unThing things)

(* [filter_run] filters out the comments within a run. *)
389 390

let filter_run : targeted_run -> filtered_targeted_run =
391
  fun (sentences, _, message) ->
392
    filter_things sentences, message
393 394 395

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

396 397
(* [setup()] returns a function [read] which reads one sentence from the
   standard input channel. *)
398

399
let setup () : unit -> sentence option =
400

401 402 403
  let open Lexing in
  let lexbuf = from_channel stdin in
  lexbuf.lex_curr_p <- { lexbuf.lex_curr_p with pos_fname = "(stdin)" };
404

405 406
  let read () =
    try
407
      SentenceParser.optional_sentence SentenceLexer.lex lexbuf
408 409 410
    with Parsing.Parse_error ->
      Error.error (Positions.lexbuf lexbuf) "Ill-formed input sentence."
  in
411

412 413 414 415 416 417
  read

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

(* If [--interpret] is set, interpret the sentences found on the standard
   input channel, then stop, without generating a parser. *)
418

419 420 421 422 423 424 425
(* We read a series of sentences from the standard input channel. To allow
   interactive use, we interpret each sentence as soon as it is read. *)

let () =
  if Settings.interpret then
    let read = setup() in
    while true do
426 427
      match read() with
      | None ->
428
  	  exit 0
429
      | Some sentence ->
430 431
	  interpret sentence
    done
432

433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448
(* --------------------------------------------------------------------------- *)

(* If [--interpret-error] is set, interpret one sentence found on the standard
   input channel, then stop, without generating a parser. *)

(* We read just one sentence, confirm that this sentence ends in an error, and
   (if that is the case) display the number of the state that is reached. *)

let () =
  if Settings.interpret_error then
    let read = setup() in
    match read() with
    | None ->
      exit 1 (* abnormal: no input *)
    | Some sentence ->
        interpret_error sentence (* never returns *)
449

450 451
(* --------------------------------------------------------------------------- *)

452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476
(* Display an informational message about the contents of a [.messages] file.  *)

let stats (runs : run or_comment list) =
  (* [s] counts the sample input sentences. [m] counts the error messages. *)
  let s = ref 0
  and m = ref 0 in
  List.iter (function
  | Thing (sentences, _, _) ->
      incr m;
      List.iter (function
      | Thing _ ->
          incr s
      | Comment _ ->
          ()
      ) sentences
  | Comment _ ->
      ()
  ) runs;
  Printf.eprintf
    "Read %d sample input sentences and %d error messages.\n%!"
    !s !m;
  runs

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

477 478
(* Reading a [.messages] file. *)

479 480 481 482 483 484
(* Our life is slightly complicated by the fact that the whitespace between
   two runs can contain comments, which we wish to preserve when performing
   [--update-errors]. *)

let read_messages filename : run or_comment list =
  let open Segment in
485
  (* Read and segment the file. *)
486
  let segments : (tag * string * Lexing.lexbuf) list = segment filename in
487 488 489 490 491 492 493
  (* Process the segments, two by two. We expect one segment to contain
     a non-empty series of sentences, and the next segment to contain
     free-form text. *)
  let rec loop accu segments =
    match segments with
    | [] ->
        List.rev accu
494 495 496
    | (Whitespace, comments, _) :: segments ->
         loop (Comment comments :: accu) segments
    | (Segment, _, lexbuf) :: segments ->
497
        (* Read a series of located sentences. *)
498 499
        match SentenceParser.entry SentenceLexer.lex lexbuf with
        | exception Parsing.Parse_error ->
500
            Error.error
501
              (Positions.one (Lexing.lexeme_start_p lexbuf))
502
              "Ill-formed sentence."
503
        | sentences ->
504 505 506
            (* In principle, we should now find a segment of whitespace
               followed with a segment of text. By construction, the two
               kinds of segments alternate. *)
507
            match segments with
508 509
            | (Whitespace, comments, _) ::
              (Segment, message, _) ::
510
              segments ->
511 512
                let run : run = sentences, comments, message in
                loop (Thing run :: accu) segments
513 514 515 516 517 518 519 520 521 522
            | []
            | [ _ ] ->
                Error.error
                  (Positions.one (Lexing.lexeme_end_p lexbuf))
                  "Syntax error: missing a final message. I may be desynchronized."
            | (Segment, _, _) :: _
            | (Whitespace, _, _) :: (Whitespace, _, _) :: _ ->
                (* Should not happen, thanks to the alternation between the
                   two kinds of segments. *)
                assert false
523
  in
524
  stats (loop [] segments)
525 526 527

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

528
(* [message_table] converts a list of targeted runs to a table (a mapping) of
529 530
   states to located sentences and messages. Optionally, it can detect that
   two sentences lead to the same state, and report an error. *)
531

532
let message_table (detect_redundancy : bool) (runs : filtered_targeted_run list)
533
  : (located_sentence * message) Lr1.NodeMap.t =
534 535

  let table =
536
    List.fold_left (fun table (sentences_and_states, message) ->
537 538
      List.fold_left (fun table (sentence2, target) ->
        let s = target2state target in
539
        match Lr1.NodeMap.find s table with
540
        | sentence1, _ ->
541 542 543 544 545 546 547
            if detect_redundancy then
              Error.signal (fst sentence1 @ fst sentence2)
                (Printf.sprintf
                   "Redundancy: these sentences both cause an error in state %d."
                   (Lr1.number s));
            table
        | exception Not_found ->
548
            Lr1.NodeMap.add s (sentence2, message) table
549 550 551 552 553 554 555 556 557 558 559 560
      ) table sentences_and_states
    ) Lr1.NodeMap.empty runs
  in
  if Error.errors() then exit 1;
  table

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

(* [compile_runs] converts a list of targeted runs to OCaml code that encodes
   a mapping of state numbers to error messages. The code is sent to the
   standard output channel. *)

561
let compile_runs filename (runs : filtered_targeted_run list) : unit =
562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578

  (* We wish to produce a function that maps a state number to a message.
     By convention, we call this function [message]. *)

  let name = "message" in

  let open IL in
  let open CodeBits in
  let default = {
    branchpat  = PWildcard;
    branchbody = eraisenotfound
  (* The default branch raises an exception, which can be caught by
     the user, who can then produce a generic error message. *)
  } in
  let branches =
    List.fold_left (fun branches (sentences_and_states, message) ->
      (* Create an or-pattern for these states. *)
579 580
      let states = List.map (fun (_, target) ->
        let s = target2state target in
581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
        pint (Lr1.number s)
      ) sentences_and_states in
      (* Map all these states to this message. *)
      { branchpat = POr states;
        branchbody = EStringConst message } :: branches
    ) [ default ] runs
  in
  let messagedef = {
    valpublic = true;
    valpat = PVar name;
    valval = EFun ([ PVar "s" ], EMatch (EVar "s", branches))
  } in
  let program = [
    SIComment (Printf.sprintf
      "This file was auto-generated based on \"%s\"." filename);
    SIComment (Printf.sprintf
      "Please note that the function [%s] can raise [Not_found]." name);
    SIValDefs (false,
      [ messagedef ]);
  ] in

  (* Write this program to the standard output channel. *)

  let module P = Printer.Make (struct
    let f = stdout
    let locate_stretches = None
  end) in
  P.program program

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

612 613 614 615 616
(* If [--compile-errors <filename>] is set, compile the error message
   descriptions found in file [filename] down to OCaml code, then stop. *)

let () =
  Settings.compile_errors |> Option.iter (fun filename ->
617

618
    (* Read the file. *)
619 620 621
    let runs : run or_comment list = read_messages filename in
    (* Drop the comments in between two runs. *)
    let runs : run list = filter_things runs in
622

623 624
    (* Convert every sentence to a state number. We signal an error if a
       sentence does not end in an error, as expected. *)
625
    let runs : targeted_run list = target_runs runs in
626

627 628
    (* Remove comments within the runs. *)
    let runs : filtered_targeted_run list = List.map filter_run runs in
629

630 631
    (* Build a mapping of states to located sentences. This allows us to
       detect if two sentences lead to the same state. *)
632
    let _ = message_table true runs in
633

634 635 636 637 638 639
    (* In principle, we would like to check whether this set of sentences is
       complete (i.e., covers all states where an error can arise), but this
       may be costly -- it requires running [LRijkstra]. Instead, we offer a
       separate facility for comparing two [.messages] files, one of which can
       be produced via [--list-errors]. This can be used to ensure
       completeness. *)
POTTIER Francois's avatar
POTTIER Francois committed
640

641 642 643
    (* Now, compile this information down to OCaml code. We wish to
       produce a function that maps a state number to a message. By
       convention, we call this function [message]. *)
644
    compile_runs filename runs;
645

646 647 648
    exit 0
  )

649 650 651 652 653 654 655 656 657 658 659 660 661
(* --------------------------------------------------------------------------- *)

(* If two [--compare-errors <filename>] directives are provided, compare the
   two message descriptions files, and stop. We wish to make sure that every
   state that appears on the left-hand side appears on the right-hand side as
   well. *)

let () =
  Settings.compare_errors |> Option.iter (fun (filename1, filename2) ->

    (* Read and convert both files, as above. *)
    let runs1 = read_messages filename1
    and runs2 = read_messages filename2 in
662 663
    let runs1 = filter_things runs1
    and runs2 = filter_things runs2 in
664 665
    let runs1 = target_runs runs1
    and runs2 = target_runs runs2 in (* here, it would be OK to ignore errors *)
666 667
    let runs1 = List.map filter_run runs1
    and runs2 = List.map filter_run runs2 in
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
    let table1 = message_table false runs1
    and table2 = message_table false runs2 in
    
    (* Check that the domain of [table1] is a subset of the domain of [table2]. *)
    table1 |> Lr1.NodeMap.iter (fun s ((poss1, _), _) ->
      if not (Lr1.NodeMap.mem s table2) then
        Error.signal poss1 (Printf.sprintf
          "This sentence leads to an error in state %d.\n\
           No sentence that leads to this state exists in \"%s\"."
          (Lr1.number s) filename2
        )
    );

    (* Check that [table1] is a subset of [table2], that is, for every state
       [s] in the domain of [table1], [s] is mapped by [table1] and [table2]
683 684 685 686 687
       to the same error message. As an exception, if the message found in
       [table1] is the default message, then no comparison takes place. This
       allows using [--list-errors] and [--compare-errors] in conjunction to
       ensure that a [.messages] file is complete, without seeing warnings
       about different messages. *)
688
    table1 |> Lr1.NodeMap.iter (fun s ((poss1, _), message1) ->
689 690 691 692 693 694 695 696 697 698 699
      if message1 <> default_message then
        try
          let (poss2, _), message2 = Lr1.NodeMap.find s table2 in
          if message1 <> message2 then
            Error.warning (poss1 @ poss2) (Printf.sprintf
              "These sentences lead to an error in state %d.\n\
               The corresponding messages in \"%s\" and \"%s\" differ."
              (Lr1.number s) filename1 filename2
            )
        with Not_found ->
          ()
700 701 702 703 704 705 706
    );

    if Error.errors() then exit 1;
    exit 0

  )

707 708 709 710 711 712 713 714 715 716 717
(* --------------------------------------------------------------------------- *)

(* If [--update-errors <filename>] is set, update the error message
   descriptions found in file [filename]. The idea is to re-generate
   the auto-comments, which are marked with ##, while leaving the
   rest untouched. *)

let () =
  Settings.update_errors |> Option.iter (fun filename ->

    (* Read the file. *)
718
    let runs : run or_comment list = read_messages filename in
719 720 721

    (* Convert every sentence to a state number. Warn, but do not
       fail, if a sentence does not end in an error, as it should. *)
722 723 724
    let runs : maybe_targeted_run or_comment list =
      List.map (or_comment_map (target_run_1 Error.warning)) runs
    in
725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741

    (* We might wish to detect if two sentences lead to the same state. We
       might also wish to detect if this set of sentences is incomplete,
       and complete it automatically. However, the first task is carried
       out by [--compile-errors] already, and the second task is carried
       out by [--list-errors] and [--compare-errors] together. For now,
       let's try and keep things as simple as possible. The task of
       [--update-errors] should be to update the auto-generated comments,
       without failing, and without adding or removing sentences. *)

    (* Now, write a new [.messages] to the standard output channel, with
       new auto-generated comments. *)
    List.iter write_run runs;

    exit 0
  )