Source file pre_sexp.ml

(* Sexp: Module for handling S-expressions (I/O, etc.) *)

open Format
open Bigarray
module Sexplib = Sexplib0
module Conv = Sexplib.Sexp_conv (* conv.ml depends on us so we can
                                   only use this module *)

include Type

type bigstring = (char, int8_unsigned_elt, c_layout) Array1.t
include (Sexplib.Sexp : module type of struct include Sexplib.Sexp end with type t := t)
include Private

let rec compare_list compare_elt a b =
  match a, b with
  | [] , [] -> 0
  | [] , _  -> -1
  | _  , [] -> 1
  | x::xs, y::ys ->
    let res = compare_elt x y in
    if res <> 0 then res
    else compare_list compare_elt xs ys

let rec compare a b =
  if a == b then
    0
  else
    match a, b with
    | Atom a, Atom b -> String.compare a b
    | Atom _, _ -> -1
    | _, Atom _ -> 1
    | List a, List b ->
      compare_list compare a b

(* Output of S-expressions to I/O-channels *)

let with_new_buffer oc f =
  let buf = buffer () in
  f buf;
  Buffer.output_buffer oc buf

let output_hum oc sexp =
  with_new_buffer oc (fun buf -> to_buffer_hum sexp ~buf)

let output_hum_indent indent oc sexp =
  with_new_buffer oc (fun buf -> to_buffer_hum ~indent sexp ~buf)

let output_mach oc sexp =
  with_new_buffer oc (fun buf -> to_buffer_mach sexp ~buf)

let output = output_mach


(* Output of S-expressions to file *)

(* The temp file functions in the OCaml Filename module do not support
   permissions.  But opening a file with given permissions is different
   from opening it and chmoding it to these permissions, because the umask
   is taken in account.  Under Unix there's no easy way to get the umask in
   a thread-safe way. *)
module Tmp_file = struct
  let prng = ref None

  let temp_file_name prefix suffix =
    let rand_state = match !prng with
      | Some v -> v
      | None ->
          let ret = Random.State.make_self_init () in
          prng := Some ret;
          ret
    in
    let rnd = (Random.State.bits rand_state) land 0xFFFFFF in
    Printf.sprintf "%s%06x%s" prefix rnd suffix

  (* Keep the permissions loose. Sexps are usually shared and rarely private*)
  let open_temp_file ?(perm = 0o666) prefix suffix =
    let rec try_name counter =
      let name = temp_file_name prefix suffix in
      try
        let oc =
          open_out_gen [Open_wronly; Open_creat; Open_excl; Open_text] perm name
        in
        name, oc
      with Sys_error _ as e ->
        if counter >= 1000 then raise e else try_name (counter + 1)
    in
    try_name 0
end

let save_of_output ?perm output_function file sexp =
  let tmp_name, oc = Tmp_file.open_temp_file ?perm file "tmp" in
  begin
    try
      output_function oc sexp;
      close_out oc;
    with e ->
      close_out_noerr oc;
      begin try Sys.remove tmp_name with _ -> () end;
      raise e
  end;
  Sys.rename tmp_name file

let output_sexp_nl do_output oc sexp =
  do_output oc sexp;
  output_string oc "\n"

let save_hum ?perm file sexp =
  save_of_output ?perm (output_sexp_nl output_hum) file sexp

let save_mach ?perm file sexp = save_of_output ?perm output_mach file sexp
let save = save_mach

let output_sexps_nl do_output oc sexps =
  List.iter (output_sexp_nl do_output oc) sexps

let save_sexps_hum ?perm file sexps =
  save_of_output ?perm (output_sexps_nl output_hum) file sexps

let save_sexps_mach ?perm file sexps =
  save_of_output ?perm (output_sexps_nl output_mach) file sexps

let save_sexps = save_sexps_mach


(* Scan functions *)

let scan_sexp ?buf lexbuf = Parser.sexp (Lexer.main ?buf) lexbuf
let scan_sexp_opt ?buf lexbuf = Parser.sexp_opt (Lexer.main ?buf) lexbuf
let scan_sexps ?buf lexbuf = Parser.sexps (Lexer.main ?buf) lexbuf
let scan_rev_sexps ?buf lexbuf = Parser.rev_sexps (Lexer.main ?buf) lexbuf

let get_main_buf buf =
  let buf =
    match buf with
    | None -> Buffer.create 128
    | Some buf -> buf in
  Lexer.main ~buf

let scan_fold_sexps ?buf ~f ~init lexbuf =
  let main = get_main_buf buf in
  let rec loop acc =
    match Parser.sexp_opt main lexbuf with
    | None -> acc
    | Some sexp -> loop (f acc sexp) in
  loop init

let scan_iter_sexps ?buf ~f lexbuf =
  scan_fold_sexps ?buf lexbuf ~init:() ~f:(fun () sexp -> f sexp)

let scan_sexps_conv ?buf ~f lexbuf =
  let coll acc sexp = f sexp :: acc in
  List.rev (scan_fold_sexps ?buf ~f:coll ~init:[] lexbuf)


(* Partial parsing *)

module Annot = struct
  type pos = { line : int; col : int; offset : int }
  type range = { start_pos : pos; end_pos : pos }
  type t = Atom of range * Type.t | List of range * t list * Type.t
  type 'a conv = [ `Result of 'a | `Error of exn * t ]

  exception Conv_exn of string * exn
  let () =
    Conv.Exn_converter.add ~finalise:false [%extension_constructor Conv_exn]
      (function
        | Conv_exn (loc, exn) ->
            List [
              Atom "Sexplib.Sexp.Annotated.Conv_exn";
              Atom loc;
              Conv.sexp_of_exn exn;
            ]
        | _ -> assert false)

  type stack = {
    mutable positions : pos list;
    mutable stack : t list list;
  }

  let get_sexp = function Atom (_, sexp) | List (_, _, sexp) -> sexp
  let get_range = function Atom (range, _) | List (range, _, _) -> range

  exception Annot_sexp of t

  let find_sexp annot_sexp sexp =
    let rec loop annot_sexp =
      match annot_sexp with
      | Atom (_, sub_sexp)
      | List (_, _, sub_sexp) when sexp == sub_sexp ->
          raise (Annot_sexp annot_sexp)
      | List (_, annots, _) -> List.iter loop annots
      | Atom _ -> ()
    in
    try loop annot_sexp; None
    with Annot_sexp res -> Some res
end

module Parse_pos = struct
  type t =
    {
      mutable text_line : int;
      mutable text_char : int;
      mutable global_offset : int;
      mutable buf_pos : int;
    }

  let create
        ?(text_line = 1) ?(text_char = 0)
        ?(buf_pos = 0) ?(global_offset = 0) () =
    let fail msg = failwith ("Sexplib.Sexp.Parse_pos.create: " ^ msg) in
    if text_line < 1 then fail "text_line < 1"
    else if text_char < 0 then fail "text_char < 0"
    else if global_offset < 0 then fail "global_offset < 0"
    else if buf_pos < 0 then fail "buf_pos < 0"
    else { text_line; text_char; global_offset; buf_pos }

  let with_buf_pos t buf_pos = { t with buf_pos }
end

module Cont_state = struct
  type t =
    | Parsing_toplevel_whitespace
    | Parsing_nested_whitespace
    | Parsing_atom
    | Parsing_list
    | Parsing_sexp_comment
    | Parsing_block_comment

  let to_string = function
    | Parsing_toplevel_whitespace -> "Parsing_toplevel_whitespace"
    | Parsing_nested_whitespace -> "Parsing_nested_whitespace"
    | Parsing_atom -> "Parsing_atom"
    | Parsing_list -> "Parsing_list"
    | Parsing_sexp_comment -> "Parsing_sexp_comment"
    | Parsing_block_comment -> "Parsing_block_comment"
end

type ('a, 't) parse_result =
  | Done of 't * Parse_pos.t
  | Cont of Cont_state.t * ('a, 't) parse_fun

and ('a, 't) parse_fun = pos : int -> len : int -> 'a -> ('a, 't) parse_result

type 't parse_state =
  {
    parse_pos : Parse_pos.t;
    mutable pstack : 't;
    pbuf : Buffer.t;
  }

type parse_error =
  {
    location : string;
    err_msg : string;
    parse_state :
      [
      | `Sexp of t list list parse_state
      | `Annot of Annot.stack parse_state
      ]
  }

exception Parse_error of parse_error
let () =
  Conv.Exn_converter.add ~finalise:false
    [%extension_constructor Parse_error]
    (function
      | Parse_error pe ->
        let ppos =
          match pe.parse_state with
          | `Sexp { parse_pos; pstack=_; pbuf=_ } | `Annot { parse_pos; pstack=_; pbuf=_ } -> parse_pos
        in
        List [
          Atom "Sexplib.Sexp.Parse_error";
          List [
            List [Atom "location"; Atom pe.location];
            List [Atom "err_msg"; Atom pe.err_msg];
            List [Atom "text_line"; Conv.sexp_of_int ppos.Parse_pos.text_line];
            List [Atom "text_char"; Conv.sexp_of_int ppos.Parse_pos.text_char];
            List [
              Atom "global_offset"; Conv.sexp_of_int ppos.Parse_pos.global_offset
            ];
            List [Atom "buf_pos"; Conv.sexp_of_int ppos.Parse_pos.buf_pos];
          ]
        ]
      | _ -> assert false)

let bump_text_line { parse_pos; _ } =
  parse_pos.Parse_pos.text_line <- parse_pos.Parse_pos.text_line + 1;
  parse_pos.Parse_pos.text_char <- 0

let bump_text_pos { parse_pos; _ } =
  parse_pos.Parse_pos.text_char <- parse_pos.Parse_pos.text_char + 1

let bump_pos_cont state str ~max_pos ~pos cont =
  bump_text_pos state;
  cont state str ~max_pos ~pos:(pos + 1)

let bump_line_cont state str ~max_pos ~pos cont =
  bump_text_line state;
  cont state str ~max_pos ~pos:(pos + 1)

let add_bump bump state str ~max_pos ~pos c cont =
  Buffer.add_char state.pbuf c;
  bump state;
  cont state str ~max_pos ~pos:(pos + 1)

let add_bump_pos state str ~max_pos ~pos c cont =
  add_bump bump_text_pos state str ~max_pos ~pos c cont

let add_bump_line state str ~max_pos ~pos c cont =
  add_bump bump_text_line state str ~max_pos ~pos c cont

let set_parse_pos parse_pos buf_pos =
  let len = buf_pos - parse_pos.Parse_pos.buf_pos in
  parse_pos.Parse_pos.buf_pos <- buf_pos;
  parse_pos.Parse_pos.global_offset <- parse_pos.Parse_pos.global_offset + len

let mk_parse_pos { parse_pos; _ } buf_pos =
  set_parse_pos parse_pos buf_pos;
  parse_pos

let raise_parse_error parse_state location buf_pos err_msg =
  match parse_state with
  | `Sexp { parse_pos; _ } | `Annot { parse_pos; _ } ->
      set_parse_pos parse_pos buf_pos;
      let parse_error = { location; err_msg; parse_state } in
      raise (Parse_error parse_error)

let raise_unexpected_char parse_state location buf_pos c =
  let err_msg = sprintf "unexpected character: '%c'" c in
  raise_parse_error parse_state location buf_pos err_msg

let mk_cont_parser cont_parse = (); fun _state str ~max_pos ~pos ->
  let len = max_pos - pos + 1 in
  cont_parse ~pos ~len str

module Safe_empty_parse = struct
  (* A type indicating whether [eof] in a certain parser state should result in valid
     parse of empty sexp list.

     [True] means that it should be considered a valid empty parse,
     [False] means that it should be considered an incomplete parse,
     [Should_not_be_empty] means that this state will guarantee [pstack] or [pbuf]
     are non-empty, so it's incomplete parse anyway.
  *)
  type t =
    | True
    | False
    | Should_not_be_empty
end

(* Macro for generating parsers *)
#define MK_PARSER( \
    TYPE, GET_LEN, PARSE, GET_CHAR, \
    GET_PSTACK, SET_PSTACK, \
    REGISTER_POS, REGISTER_POS1, \
    MK_ATOM, MK_LIST, INIT_PSTACK, MK_PARSE_STATE) \
  let bump_found_atom bump state str ~max_pos ~pos cont = \
    let pbuf = state.pbuf in \
    let pbuf_str = Buffer.contents pbuf in \
    let atom = MK_ATOM in \
    match GET_PSTACK with \
    | [] -> Done (atom, mk_parse_pos state pos) \
    | rev_sexp_lst :: sexp_stack -> \
        Buffer.clear pbuf; \
        let pstack = (atom :: rev_sexp_lst) :: sexp_stack in \
        SET_PSTACK; \
        bump state; \
        cont state str ~max_pos ~pos:(pos + 1) \
  \
  let check_str_bounds loc ~pos ~len (str : TYPE) = \
    if pos < 0 then invalid_arg (loc ^ ": pos < 0"); \
    if len < 0 then invalid_arg (loc ^ ": len < 0"); \
    let str_len = GET_LEN str in \
    let pos_len = pos + len in \
    if pos_len > str_len then invalid_arg (loc ^ ": pos + len > str_len"); \
    pos_len - 1 \
  \
  let mk_cont_state name cont state ~cont_state = \
    let parse_fun = \
      let used_ref = ref false in \
      fun ~pos ~len str -> \
        if !used_ref then \
          failwith "Sexplib.Sexp: parser continuation called twice" \
        else begin \
          used_ref := true; \
          let max_pos = check_str_bounds name ~pos ~len str in \
          cont state str ~max_pos ~pos \
        end \
    in \
    Cont (cont_state, parse_fun) \
  \
  let mk_cont name ~(eof_safe : Safe_empty_parse.t) cont state = \
    let cont_state = \
      match GET_PSTACK = [], Buffer.length state.pbuf = 0 with \
      | true, true -> \
        begin match eof_safe with \
        | True -> Cont_state.Parsing_toplevel_whitespace \
        | False -> Cont_state.Parsing_nested_whitespace \
        | Should_not_be_empty -> \
          failwith ("should never be parsing whitespace in "^ name) \
        end \
      | false, true -> Cont_state.Parsing_list \
      | _, false -> Cont_state.Parsing_atom \
    in \
    mk_cont_state name cont state ~cont_state \
  \
  let rec PARSE state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse" ~eof_safe:True PARSE state \
    else \
      match GET_CHAR with \
      | '(' -> \
          REGISTER_POS \
          let pstack = [] :: GET_PSTACK in \
          SET_PSTACK; \
          bump_pos_cont state str ~max_pos ~pos PARSE \
      | ')' as c -> \
          (match GET_PSTACK with \
          | [] -> raise_unexpected_char (MK_PARSE_STATE state) "parse" pos c \
          | rev_sexp_lst :: sexp_stack -> \
              let sexp_lst = List.rev rev_sexp_lst in \
              let sexp = MK_LIST in \
              match sexp_stack with \
              | [] -> Done (sexp, mk_parse_pos state (pos + 1)) \
              | higher_rev_sexp_lst :: higher_sexp_stack -> \
                  let pstack = \
                    (sexp :: higher_rev_sexp_lst) :: higher_sexp_stack \
                  in \
                  SET_PSTACK; \
                  bump_pos_cont state str ~max_pos ~pos PARSE) \
      | ' ' | '\009' | '\012' -> bump_pos_cont state str ~max_pos ~pos PARSE \
      | '\010' -> bump_line_cont state str ~max_pos ~pos PARSE \
      | '\013' -> bump_pos_cont state str ~max_pos ~pos parse_nl \
      | ';' -> bump_pos_cont state str ~max_pos ~pos parse_comment \
      | '"' -> \
          REGISTER_POS1 \
          bump_pos_cont state str ~max_pos ~pos parse_quoted \
      | c -> \
          REGISTER_POS \
          let parse = \
            match c with \
            | '#' -> maybe_parse_comment \
            | '|' -> maybe_parse_close_comment \
            | _ -> parse_atom \
          in \
          add_bump_pos state str ~max_pos ~pos c parse \
  \
  and parse_nl state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_nl" ~eof_safe:False parse_nl state \
    else \
      let c = GET_CHAR in \
      if c = '\010' then bump_line_cont state str ~max_pos ~pos PARSE \
      else raise_unexpected_char (MK_PARSE_STATE state) "parse_nl" pos c \
  \
  and parse_comment state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_comment" ~eof_safe:True parse_comment state \
    else \
      match GET_CHAR with \
      | '\010' -> bump_line_cont state str ~max_pos ~pos PARSE \
      | '\013' -> bump_pos_cont state str ~max_pos ~pos parse_nl \
      | _ -> bump_pos_cont state str ~max_pos ~pos parse_comment \
  \
  and maybe_parse_comment state str ~max_pos ~pos = \
    if pos > max_pos then \
      mk_cont "maybe_parse_comment" ~eof_safe:Should_not_be_empty maybe_parse_comment state \
    else \
      match GET_CHAR with \
      | ';' -> bump_pos_cont state str ~max_pos ~pos parse_sexp_comment \
      | '|' -> bump_pos_cont state str ~max_pos ~pos parse_block_comment \
      | _ -> parse_atom state str ~max_pos ~pos \
  \
  and maybe_parse_close_comment state str ~max_pos ~pos = \
    if pos > max_pos then \
      mk_cont "maybe_parse_close_comment" ~eof_safe:Should_not_be_empty maybe_parse_close_comment state \
    else \
      if GET_CHAR <> '#' then parse_atom state str ~max_pos ~pos \
      else \
        let err_msg = "end of block comment without start" in \
        raise_parse_error (MK_PARSE_STATE state) \
          "maybe_parse_close_comment" pos err_msg \
  \
  and parse_sexp_comment state str ~max_pos ~pos = \
    let pbuf_str = "" in \
    ignore (MK_ATOM); \
    Buffer.clear state.pbuf; \
    let old_pstack = GET_PSTACK in \
    let pstack = [] in \
    SET_PSTACK; \
    let rec loop parse state str ~max_pos ~pos = \
      match parse state str ~max_pos ~pos with \
      | Done (_sexp, { Parse_pos.buf_pos = pos; _ }) -> \
          Buffer.clear state.pbuf; \
          let pstack = old_pstack in \
          SET_PSTACK; \
          PARSE state str ~max_pos ~pos \
      | Cont (_, cont_parse) -> \
          Buffer.clear state.pbuf; \
          let parse = mk_cont_parser cont_parse in \
          mk_cont_state "parse_sexp_comment" (loop parse) state \
            ~cont_state:Cont_state.Parsing_sexp_comment \
    in \
    loop PARSE state str ~max_pos ~pos \
  \
  and parse_block_comment state str ~max_pos ~pos = \
    let pbuf_str = "" in \
    ignore (MK_ATOM); \
    Buffer.clear state.pbuf; \
    let old_pstack = GET_PSTACK in \
    let pstack = [] in \
    SET_PSTACK; \
    let rec loop depth state str ~max_pos ~pos = \
      let rec parse_block_depth state str ~max_pos ~pos = \
        if pos > max_pos then \
          mk_cont "parse_block_depth" ~eof_safe:False parse_block_depth state \
        else \
          match GET_CHAR with \
          | '\010' -> bump_line_cont state str ~max_pos ~pos parse_block_depth \
          | '"' -> \
              REGISTER_POS1 \
              let rec parse_block_quote parse state str ~max_pos ~pos = \
                match parse state str ~max_pos ~pos with \
                | Done (_sexp, { Parse_pos.buf_pos = pos; _ }) -> \
                    Buffer.clear state.pbuf; \
                    parse_block_depth state str ~max_pos ~pos \
                | Cont (_, cont_parse) -> \
                    Buffer.clear state.pbuf; \
                    let parse = mk_cont_parser cont_parse in \
                    mk_cont_state "parse_block_quote" \
                      (parse_block_quote parse) state \
                      ~cont_state:Cont_state.Parsing_block_comment \
              in \
              bump_pos_cont state str ~max_pos ~pos \
                (parse_block_quote parse_quoted) \
          | '#' -> bump_pos_cont state str ~max_pos ~pos parse_open_block \
          | '|' -> bump_pos_cont state str ~max_pos ~pos parse_close_block \
          | _ -> bump_pos_cont state str ~max_pos ~pos parse_block_depth \
      and parse_open_block state str ~max_pos ~pos = \
        if pos > max_pos then \
          mk_cont "parse_open_block" ~eof_safe:False parse_open_block state \
        else \
          if GET_CHAR = '|' then \
            bump_pos_cont state str ~max_pos ~pos (loop (depth + 1)) \
          else parse_block_depth state str ~max_pos ~pos \
      and parse_close_block state str ~max_pos ~pos = \
        if pos > max_pos then \
          mk_cont "parse_close_block" ~eof_safe:False parse_close_block state \
        else if GET_CHAR = '#' then \
          let parse = \
            if depth = 1 then \
              let () = Buffer.clear state.pbuf in \
              let pstack = old_pstack in \
              SET_PSTACK; \
              PARSE \
            else loop (depth - 1) \
          in \
          bump_pos_cont state str ~max_pos ~pos parse \
        else parse_block_depth state str ~max_pos ~pos \
      in \
      parse_block_depth state str ~max_pos ~pos \
    in \
    loop 1 state str ~max_pos ~pos \
  \
  and parse_atom state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_atom" ~eof_safe:Should_not_be_empty parse_atom state \
    else \
      match GET_CHAR with \
      | ' ' | '\009' | '\012' -> \
          bump_found_atom bump_text_pos state str ~max_pos ~pos PARSE \
      | '#' as c -> \
          add_bump_pos state str ~max_pos ~pos c maybe_parse_bad_atom_hash \
      | '|' as c -> \
          add_bump_pos state str ~max_pos ~pos c maybe_parse_bad_atom_pipe \
      | '(' -> \
          let pbuf = state.pbuf in \
          let pbuf_str = Buffer.contents pbuf in \
          let atom = MK_ATOM in \
          (match GET_PSTACK with \
          | [] -> Done (atom, mk_parse_pos state pos) \
          | rev_sexp_lst :: sexp_stack -> \
              REGISTER_POS \
              Buffer.clear pbuf; \
              let pstack = [] :: (atom :: rev_sexp_lst) :: sexp_stack in \
              SET_PSTACK; \
              bump_pos_cont state str ~max_pos ~pos PARSE) \
      | ')' -> \
          let pbuf = state.pbuf in \
          let pbuf_str = Buffer.contents pbuf in \
          let atom = MK_ATOM in \
          (match GET_PSTACK with \
          | [] -> Done (atom, mk_parse_pos state pos) \
          | rev_sexp_lst :: sexp_stack -> \
              let sexp_lst = List.rev_append rev_sexp_lst [atom] in \
              let sexp = MK_LIST in \
              match sexp_stack with \
              | [] -> Done (sexp, mk_parse_pos state (pos + 1)) \
              | higher_rev_sexp_lst :: higher_sexp_stack -> \
                  Buffer.clear pbuf; \
                  let pstack = \
                    (sexp :: higher_rev_sexp_lst) :: higher_sexp_stack \
                  in \
                  SET_PSTACK; \
                  bump_pos_cont state str ~max_pos ~pos PARSE) \
      | '\010' -> bump_found_atom bump_text_line state str ~max_pos ~pos PARSE \
      | '\013' -> \
          bump_found_atom bump_text_pos state str ~max_pos ~pos parse_nl \
      | ';' -> \
          bump_found_atom bump_text_pos state str ~max_pos ~pos parse_comment \
      | '"' -> \
          bump_found_atom \
            bump_text_pos state str ~max_pos ~pos reg_parse_quoted \
      | c -> \
          (* This is [add_bump_pos state str ~max_pos ~pos c parse_atom] inlined by \
             hand, see https://github.com/janestreet/sexplib/pull/14 for details: *) \
          Buffer.add_char state.pbuf c; \
          bump_text_pos state; \
          parse_atom state str ~max_pos ~pos:(pos + 1) \
  \
  and maybe_parse_bad_atom_pipe state str ~max_pos ~pos = \
    if pos > max_pos then \
      mk_cont "maybe_parse_bad_atom_pipe" \
        ~eof_safe:Should_not_be_empty maybe_parse_bad_atom_pipe state \
    else \
      match GET_CHAR with \
      | '#' -> \
          let err_msg = "illegal end of block comment in unquoted atom" in \
          raise_parse_error (MK_PARSE_STATE state) "maybe_parse_bad_atom_pipe" \
            pos err_msg \
      | _ -> parse_atom state str ~max_pos ~pos \
  \
  and maybe_parse_bad_atom_hash state str ~max_pos ~pos = \
    if pos > max_pos then \
      mk_cont "maybe_parse_bad_atom_hash" \
        ~eof_safe:Should_not_be_empty maybe_parse_bad_atom_hash state \
    else \
      match GET_CHAR with \
      | '|' -> \
          let err_msg = "illegal start of block comment in unquoted atom" in \
          raise_parse_error (MK_PARSE_STATE state) "maybe_parse_bad_atom_hash" \
            pos err_msg \
      | _ -> parse_atom state str ~max_pos ~pos \
  \
  and reg_parse_quoted state str ~max_pos ~pos = \
    REGISTER_POS \
    parse_quoted state str ~max_pos ~pos \
  \
  and parse_quoted state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_quoted" ~eof_safe:False parse_quoted state \
    else \
      match GET_CHAR with \
      | '"' -> \
          let pbuf = state.pbuf in \
          let pbuf_str = Buffer.contents pbuf in \
          let atom = MK_ATOM in \
          (match GET_PSTACK with \
          | [] -> Done (atom, mk_parse_pos state (pos + 1)) \
          | rev_sexp_lst :: sexp_stack -> \
              Buffer.clear pbuf; \
              let pstack = (atom :: rev_sexp_lst) :: sexp_stack in \
              SET_PSTACK; \
              bump_pos_cont state str ~max_pos ~pos PARSE) \
      | '\\' -> bump_pos_cont state str ~max_pos ~pos parse_escaped \
      | '\010' as c -> add_bump_line state str ~max_pos ~pos c parse_quoted \
      | c -> \
          (* This is [add_bump_pos state str ~max_pos ~pos c parse_quoted] inlined by \
             hand, see https://github.com/janestreet/sexplib/pull/14 for details: *) \
          Buffer.add_char state.pbuf c; \
          bump_text_pos state; \
          parse_quoted state str ~max_pos ~pos:(pos + 1) \
  \
  and parse_escaped state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_escaped" ~eof_safe:False parse_escaped state \
    else \
      match GET_CHAR with \
      | '\010' -> bump_line_cont state str ~max_pos ~pos parse_skip_ws \
      | '\013' -> bump_pos_cont state str ~max_pos ~pos parse_skip_ws_nl \
      | '0' .. '9' as c -> \
          bump_text_pos state; \
          let d = Char.code c - 48 in \
          parse_dec state str ~max_pos ~pos:(pos + 1) ~count:2 ~d \
      | 'x' -> \
          bump_text_pos state; \
          parse_hex state str ~max_pos ~pos:(pos + 1) ~count:2 ~d:0 \
      | ('\\' | '"' | '\'' ) as c -> \
          add_bump_pos state str ~max_pos ~pos c parse_quoted \
      | 'n' -> add_bump_pos state str ~max_pos ~pos '\n' parse_quoted \
      | 't' -> add_bump_pos state str ~max_pos ~pos '\t' parse_quoted \
      | 'b' -> add_bump_pos state str ~max_pos ~pos '\b' parse_quoted \
      | 'r' -> add_bump_pos state str ~max_pos ~pos '\r' parse_quoted \
      | c -> \
          Buffer.add_char state.pbuf '\\'; \
          add_bump_pos state str ~max_pos ~pos c parse_quoted \
  \
  and parse_skip_ws state str ~max_pos ~pos = \
    if pos > max_pos then mk_cont "parse_skip_ws" ~eof_safe:False parse_skip_ws state \
    else \
      match GET_CHAR with \
      | ' ' | '\009' -> bump_pos_cont state str ~max_pos ~pos parse_skip_ws \
      | _ -> parse_quoted state str ~max_pos ~pos \
  \
  and parse_skip_ws_nl state str ~max_pos ~pos = \
      if pos > max_pos then \
        mk_cont "parse_skip_ws_nl" ~eof_safe:False parse_skip_ws_nl state \
    else \
      if GET_CHAR = '\010' then \
        bump_line_cont state str ~max_pos ~pos parse_skip_ws \
      else begin \
        Buffer.add_char state.pbuf '\013'; \
        parse_quoted state str ~max_pos ~pos \
      end \
  \
  and parse_dec state str ~max_pos ~pos ~count ~d = \
    if pos > max_pos then mk_cont "parse_dec" ~eof_safe:False (parse_dec ~count ~d) state \
    else \
      match GET_CHAR with \
      | '0' .. '9' as c -> \
          let d = 10 * d + Char.code c - 48 in \
          if count = 1 then \
            if d > 255 then \
              let err_msg = sprintf "illegal decimal escape: \\%d" d in \
              raise_parse_error (MK_PARSE_STATE state) "parse_dec" pos err_msg \
            else \
              add_bump_pos state str ~max_pos ~pos (Char.chr d) parse_quoted \
          else ( \
            bump_text_pos state; \
            parse_dec state str ~max_pos ~pos:(pos + 1) ~count:(count - 1) ~d) \
      | c -> raise_unexpected_char (MK_PARSE_STATE state) "parse_dec" pos c \
  \
  and parse_hex state str ~max_pos ~pos ~count ~d = \
    if pos > max_pos then mk_cont "parse_hex" ~eof_safe:False (parse_hex ~count ~d) state \
    else \
      match GET_CHAR with \
      | '0' .. '9' | 'a' .. 'f' | 'A' .. 'F' as c -> \
          let corr = \
            if c >= 'a' then 87 \
            else if c >= 'A' then 55 \
            else 48 \
          in \
          let d = 16 * d + Char.code c - corr in \
          if count = 1 then \
            if d > 255 then \
              let err_msg = sprintf "illegal hexadecimal escape: \\%x" d in \
              raise_parse_error (MK_PARSE_STATE state) "parse_hex" pos err_msg \
            else \
              add_bump_pos state str ~max_pos ~pos (Char.chr d) parse_quoted \
          else ( \
            bump_text_pos state; \
            parse_hex state str ~max_pos ~pos:(pos + 1) ~count:(count - 1) ~d) \
      | c -> raise_unexpected_char (MK_PARSE_STATE state) "parse_hex" pos c \
  \
  let PARSE ?(parse_pos = Parse_pos.create ()) ?len str = \
    let pos = parse_pos.Parse_pos.buf_pos in \
    let len = \
      match len with \
      | Some len -> len \
      | None -> GET_LEN str - pos \
    in \
    let max_pos = check_str_bounds "parse" ~pos ~len str in \
    let state = \
      { \
        parse_pos; \
        pstack = INIT_PSTACK; \
        pbuf = Buffer.create 128; \
      } \
    in \
    PARSE state str ~max_pos ~pos

MK_PARSER(
  string, String.length, parse_str, str.[pos],
  state.pstack, state.pstack <- pstack,
  ,,
  Atom pbuf_str, List sexp_lst, [], `Sexp
)

let parse = parse_str


(* Annot parsers *)

let get_glob_ofs parse_pos pos =
  parse_pos.Parse_pos.global_offset + pos - parse_pos.Parse_pos.buf_pos

let mk_annot_pos
      ({ Parse_pos.text_line = line; text_char = col; _ } as parse_pos) pos =
  { Annot.line; col; offset = get_glob_ofs parse_pos pos }

let mk_annot_pos1
      ({ Parse_pos.text_line = line; text_char = col; _ } as parse_pos) pos =
  { Annot.line; col = col + 1; offset = get_glob_ofs parse_pos pos }

let add_annot_pos { parse_pos; pstack; pbuf = _ } pos =
  pstack.Annot.positions <- mk_annot_pos parse_pos pos :: pstack.Annot.positions

let add_annot_pos1 { parse_pos; pstack; pbuf = _ } pos =
  pstack.Annot.positions <-
    mk_annot_pos1 parse_pos pos :: pstack.Annot.positions

let get_annot_range { parse_pos; pstack; pbuf = _ } pos =
  let start_pos =
    match pstack.Annot.positions with
    | [] -> assert false  (* impossible *)
    | h :: t -> pstack.Annot.positions <- t; h
  in
  let end_pos =
    {
      Annot.
      line = parse_pos.Parse_pos.text_line;
      col = parse_pos.Parse_pos.text_char;
      offset = get_glob_ofs parse_pos pos;
    }
  in
  { Annot.start_pos; end_pos }

let mk_annot_atom parse_state str pos =
  Annot.Atom (get_annot_range parse_state pos, Atom str)

let mk_annot_list parse_state annot_lst pos =
  let range = get_annot_range parse_state pos in
  let sexp = List (List.rev (List.rev_map Annot.get_sexp annot_lst)) in
  Annot.List (range, annot_lst, sexp)

let init_annot_pstate () = { Annot.positions = []; stack = [] }

MK_PARSER(
  string, String.length, parse_str_annot, str.[pos],
  state.pstack.Annot.stack, state.pstack.Annot.stack <- pstack,
  add_annot_pos state pos;,add_annot_pos1 state pos;,
  mk_annot_atom state pbuf_str pos, mk_annot_list state sexp_lst pos,
  init_annot_pstate (), `Annot
)


(* Partial parsing from bigstrings *)

(* NOTE: this is really an awful duplication of the code for parsing
   strings, but since OCaml does not inline higher-order functions known
   at compile, other solutions would sacrifice a lot of efficiency. *)

MK_PARSER(
  bigstring, Array1.dim, parse_bigstring, str.{pos},
  state.pstack, state.pstack <- pstack,
  ,,
  Atom pbuf_str, List sexp_lst, [], `Sexp
)

MK_PARSER(
  bigstring, Array1.dim, parse_bigstring_annot, str.{pos},
  state.pstack.Annot.stack, state.pstack.Annot.stack <- pstack,
  add_annot_pos state pos;,add_annot_pos1 state pos;,
  mk_annot_atom state pbuf_str pos, mk_annot_list state sexp_lst pos,
  init_annot_pstate (), `Annot
)


(* Input functions *)

let mk_this_parse ?parse_pos my_parse = (); fun ~pos ~len str ->
  let parse_pos =
    match parse_pos with
    | None -> Parse_pos.create ~buf_pos:pos ()
    | Some parse_pos -> parse_pos.Parse_pos.buf_pos <- pos; parse_pos
  in
  my_parse ?parse_pos:(Some parse_pos) ?len:(Some len) str

(* [ws_buf] must contain a single space character *)
let feed_end_of_input ~this_parse ~ws_buf =
  (* When parsing atoms, the incremental parser cannot tell whether
     it is at the end until it hits whitespace.  We therefore feed it
     one space to determine whether it is finished. *)
  match this_parse ~pos:0 ~len:1 ws_buf with
  | Done (sexp, _) -> Ok sexp
  | Cont (cont_state, _) -> Error cont_state

let gen_input_sexp my_parse ?parse_pos ic =
  let buf = Bytes.create 1 in
  let rec loop this_parse =
    match input_char ic with
    | exception End_of_file ->
      (match feed_end_of_input ~this_parse ~ws_buf:" " with
       | Ok sexp -> sexp
       | Error _ -> raise End_of_file)
    | c ->
      Bytes.set buf 0 c;
      match this_parse ~pos:0 ~len:1 (Bytes.unsafe_to_string buf) with
      | Done (sexp, _) -> sexp
      | Cont (_, this_parse) -> loop this_parse
  in
  loop (mk_this_parse ?parse_pos my_parse)

let input_sexp ?parse_pos ic = gen_input_sexp parse ?parse_pos ic

let gen_input_rev_sexps my_parse ~ws_buf ?parse_pos ?(buf = Bytes.create 8192) ic =
  let rev_sexps_ref = ref [] in
  let buf_len = Bytes.length buf in
  let rec loop this_parse ~pos ~len =
    if len > 0 then
      match this_parse ~pos ~len (Bytes.unsafe_to_string buf) with
      | Done (sexp, ({ Parse_pos.buf_pos; _ } as parse_pos)) ->
        rev_sexps_ref := sexp :: !rev_sexps_ref;
        let n_parsed = buf_pos - pos in
        let this_parse = mk_this_parse ~parse_pos my_parse in
        if n_parsed = len then
          let new_len = input ic buf 0 buf_len in
          loop this_parse ~pos:0 ~len:new_len
        else loop this_parse ~pos:buf_pos ~len:(len - n_parsed)
      | Cont (_, this_parse) ->
        loop this_parse ~pos:0 ~len:(input ic buf 0 buf_len)
    else
      match feed_end_of_input ~this_parse ~ws_buf with
      | Ok sexp ->
        sexp :: !rev_sexps_ref
      | Error Parsing_toplevel_whitespace ->
        !rev_sexps_ref
      | Error cont_state ->
        failwith (
          "Sexplib.Sexp.input_rev_sexps: reached EOF while in state "
          ^ Cont_state.to_string cont_state)
  in
  let len = input ic buf 0 buf_len in
  let this_parse = mk_this_parse ?parse_pos my_parse in
  loop this_parse ~pos:0 ~len

let input_rev_sexps ?parse_pos ?buf ic =
  gen_input_rev_sexps parse ~ws_buf:" " ?parse_pos ?buf ic

let input_sexps ?parse_pos ?buf ic =
  List.rev (input_rev_sexps ?parse_pos ?buf ic)


(* of_string and of_bigstring *)

let of_string_bigstring loc my_parse ws_buf get_len get_sub str =
  match my_parse ?parse_pos:None ?len:None str with
  | Done (sexp, parse_pos) ->
    begin
      match my_parse ?parse_pos:(Some parse_pos) ?len:None str with
      | Done (_sexp2, _) ->
        failwith (sprintf (
          "Sexplib.Sexp.%s: got multiple S-expressions where only one was expected."
        ) loc)
      | Cont (Cont_state.Parsing_toplevel_whitespace, _) ->
        sexp
      | Cont (_, _) ->
        (* not using [feed_end_of_input] here means "a b" will end up here and not in
           "multiple S-expressions" branch, but it doesn't matter that much *)
        failwith (sprintf (
          "Sexplib.Sexp.%s: S-expression followed by data at position %d...")
          loc parse_pos.buf_pos)
    end
  | Cont (_, this_parse) ->
    match feed_end_of_input ~this_parse ~ws_buf with
    | Ok sexp -> sexp
    | Error cont_state ->
      let cont_state_str = Cont_state.to_string cont_state in
      failwith (
        sprintf
          "Sexplib.Sexp.%s: incomplete S-expression while in state %s: %s"
          loc cont_state_str (get_sub str 0 (get_len str)))

let of_string str =
  of_string_bigstring "of_string" parse " " String.length String.sub str

let get_bstr_sub_str bstr pos len =
  let str = Bytes.create len in
  for i = 0 to len - 1 do Bytes.set str i bstr.{pos + i} done;
  Bytes.unsafe_to_string str

let bstr_ws_buf = Array1.create char c_layout 1
let () = bstr_ws_buf.{0} <- ' '

let of_bigstring bstr =
  of_string_bigstring
    "of_bigstring"
    parse_bigstring
    bstr_ws_buf Array1.dim get_bstr_sub_str bstr


(* Loading *)

let gen_load_rev_sexps input_rev_sexps ?buf file =
  let ic = open_in file in
  try
    let sexps = input_rev_sexps ?parse_pos:None ?buf ic in
    close_in ic;
    sexps
  with exc -> close_in_noerr ic; raise exc

let load_rev_sexps ?buf file = gen_load_rev_sexps input_rev_sexps ?buf file

let load_sexps ?buf file = List.rev (load_rev_sexps ?buf file)

let gen_load_sexp_loc = "Sexplib.Sexp.gen_load_sexp"

let gen_load_sexp my_parse ?(strict = true) ?(buf = Bytes.create 8192) file =
  let buf_len = Bytes.length buf in
  let ic = open_in file in
  let rec loop this_parse =
    let len = input ic buf 0 buf_len in
    if len = 0 then
      match feed_end_of_input ~this_parse ~ws_buf:" " with
      | Ok sexp -> sexp
      | Error cont_state ->
        failwith (
          sprintf "%s: EOF in %s while in state %s"
            gen_load_sexp_loc file (Cont_state.to_string cont_state))
    else
      match this_parse ~pos:0 ~len (Bytes.unsafe_to_string buf) with
      | Done (sexp, ({ Parse_pos.buf_pos; _ } as parse_pos)) when strict ->
        let rec strict_loop this_parse ~pos ~len =
            match this_parse ~pos ~len (Bytes.unsafe_to_string buf) with
            | Done _ ->
                failwith (
                  sprintf "%s: more than one S-expression in file %s"
                    gen_load_sexp_loc file)
            | Cont (cont_state, this_parse) ->
                let len = input ic buf 0 buf_len in
                if len > 0 then strict_loop this_parse ~pos:0 ~len
                else if cont_state = Cont_state.Parsing_toplevel_whitespace then sexp
                else
                  failwith (
                    sprintf "%s: %s in state %s loading file %s"
                      gen_load_sexp_loc "additional incomplete data"
                      (Cont_state.to_string cont_state) file)
          in
          let this_parse = mk_this_parse ~parse_pos my_parse in
          strict_loop this_parse ~pos:buf_pos ~len:(len - buf_pos)
      | Done (sexp, _) -> sexp
      | Cont (_, this_parse) -> loop this_parse
  in
  try
    let sexp = loop (mk_this_parse my_parse) in
    close_in ic;
    sexp
  with exc -> close_in_noerr ic; raise exc

let load_sexp ?strict ?buf file = gen_load_sexp parse ?strict ?buf file

module Annotated = struct
  include Annot

  let parse = parse_str_annot
  let parse_bigstring = parse_bigstring_annot

  let input_rev_sexps ?parse_pos ?buf ic =
    gen_input_rev_sexps parse ~ws_buf:" " ?parse_pos ?buf ic

  let input_sexp ?parse_pos ic = gen_input_sexp parse ?parse_pos ic

  let input_sexps ?parse_pos ?buf ic =
    List.rev (input_rev_sexps ?parse_pos ?buf ic)

  let of_string str =
    of_string_bigstring
      "Annotated.of_string" parse
      " " String.length String.sub str

  let of_bigstring bstr =
    of_string_bigstring
      "Annotated.of_bigstring"
      parse_bigstring
      bstr_ws_buf Array1.dim get_bstr_sub_str bstr

  let load_rev_sexps ?buf file = gen_load_rev_sexps input_rev_sexps ?buf file
  let load_sexps ?buf file = List.rev (load_rev_sexps ?buf file)
  let load_sexp ?strict ?buf file = gen_load_sexp parse ?strict ?buf file

  let conv f annot_sexp =
    let sexp = get_sexp annot_sexp in
    try `Result (f sexp)
    with Of_sexp_error (exc, bad_sexp) as e ->
      match find_sexp annot_sexp bad_sexp with
      | None -> raise e
      | Some bad_annot_sexp -> `Error (exc, bad_annot_sexp)

  let get_conv_exn ~file ~exc annot_sexp =
    let range = get_range annot_sexp in
    let { start_pos = { line; col; offset = _ }; end_pos = _ } = range in
    let loc = sprintf "%s:%d:%d" file line col in
    Of_sexp_error (Annot.Conv_exn (loc, exc), get_sexp annot_sexp)
end

let load_sexp_conv ?(strict = true) ?(buf = String.create 8192) file f =
  let sexp = load_sexp ~strict ~buf file in
  try `Result (f sexp)
  with Of_sexp_error _ ->
    Annotated.conv f (Annotated.load_sexp ~strict ~buf file)

let raise_conv_exn ~file = function
  | `Result res -> res
  | `Error (exc, annot_sexp) ->
      raise (Annotated.get_conv_exn ~file ~exc annot_sexp)

let load_sexp_conv_exn ?strict ?buf file f =
  raise_conv_exn ~file (load_sexp_conv ?strict ?buf file f)

let load_sexps_conv ?(buf = String.create 8192) file f =
  let rev_sexps = load_rev_sexps ~buf file in
  try List.rev_map (fun sexp -> `Result (f sexp)) rev_sexps
  with Of_sexp_error _ as e ->
    match Annotated.load_rev_sexps ~buf file with
    | [] ->
        (* File is now empty - perhaps it was a temporary file handle? *)
        raise e
    | rev_annot_sexps ->
        List.rev_map (fun annot_sexp -> Annotated.conv f annot_sexp)
          rev_annot_sexps

let load_sexps_conv_exn ?(buf = String.create 8192) file f =
  let rev_sexps = load_rev_sexps ~buf file in
  try List.rev_map f rev_sexps
  with Of_sexp_error _ as e ->
    match Annotated.load_rev_sexps ~buf file with
    | [] ->
        (* File is now empty - perhaps it was a temporary file handle? *)
        raise e
    | rev_annot_sexps ->
        List.rev_map
          (fun annot_sexp -> raise_conv_exn ~file (Annotated.conv f annot_sexp))
          rev_annot_sexps

let gen_of_string_conv of_string annot_of_string str f =
  let sexp = of_string str in
  try `Result (f sexp)
  with Of_sexp_error _ -> Annotated.conv f (annot_of_string str)

let of_string_conv str f =
  gen_of_string_conv of_string Annotated.of_string str f

let of_bigstring_conv bstr f =
  gen_of_string_conv of_bigstring Annotated.of_bigstring bstr f

module Of_string_conv_exn = struct
  type t = { exc : exn; sexp : Type.t; sub_sexp : Type.t }

  exception E of t

  let () =
    Conv.Exn_converter.add ~finalise:false [%extension_constructor E]
      (function
        | E osce ->
          List [
            Atom "Sexplib.Sexp.Of_string_conv_exn.E";
            List [
              List [Atom "exc"; Conv.sexp_of_exn osce.exc];
              List [Atom "sexp"; osce.sexp];
              List [Atom "sub_sexp"; osce.sub_sexp];
            ]
            ]
        | _ -> assert false)
end

let gen_of_string_conv_exn of_string str f =
  let sexp = of_string str in
  try f sexp
  with Of_sexp_error (exc, sub_sexp) ->
    raise (Of_string_conv_exn.E { Of_string_conv_exn.exc; sexp; sub_sexp })

let of_string_conv_exn str f = gen_of_string_conv_exn of_string str f
let of_bigstring_conv_exn bstr f = gen_of_string_conv_exn of_bigstring bstr f


(* Utilities for automated type conversions *)

let unit = List []

external sexp_of_t : t -> t = "%identity"
external t_of_sexp : t -> t = "%identity"


(* Utilities for conversion error handling *)

type found = [ `Found | `Pos of int * found ]
type search_result = [ `Not_found | found ]

let rec search_physical sexp ~contained =
  if sexp == contained then `Found
  else
    match sexp with
    | Atom _ -> `Not_found
    | List lst ->
        let rec loop i = function
          | [] -> `Not_found
          | h :: t ->
              let res = search_physical h ~contained in
              match res with
              | `Not_found -> loop (i + 1) t
              | #found as found -> `Pos (i, found)
        in
        loop 0 lst

let rec subst_found sexp ~subst = function
  | `Found -> subst
  | `Pos (pos, found) ->
      match sexp with
      | Atom _ ->
          failwith "Sexplib.Sexp.subst_found: atom when position requested"
      | List lst ->
          let rec loop acc pos = function
            | [] ->
                failwith
                  "Sexplib.Sexp.subst_found: short list when position requested"
            | h :: t when pos <> 0 -> loop (h :: acc) (pos - 1) t
            | h :: t ->
                List (List.rev_append acc (subst_found h ~subst found :: t))
          in
          loop [] pos lst