[pspp] / src / data / sys-file-reader.c

/* PSPP - a program for statistical analysis.
   Copyright (C) 1997-2000, 2006-2007, 2009-2013 Free Software Foundation, Inc.

   This program is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation, either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <http://www.gnu.org/licenses/>. */

#include <config.h>

#include "data/sys-file-reader.h"
#include "data/sys-file-private.h"

#include <errno.h>
#include <float.h>
#include <inttypes.h>
#include <setjmp.h>
#include <stdlib.h>

#include "data/attributes.h"
#include "data/case.h"
#include "data/casereader-provider.h"
#include "data/casereader.h"
#include "data/dictionary.h"
#include "data/file-handle-def.h"
#include "data/file-name.h"
#include "data/format.h"
#include "data/identifier.h"
#include "data/missing-values.h"
#include "data/mrset.h"
#include "data/short-names.h"
#include "data/value-labels.h"
#include "data/value.h"
#include "data/variable.h"
#include "libpspp/array.h"
#include "libpspp/assertion.h"
#include "libpspp/compiler.h"
#include "libpspp/i18n.h"
#include "libpspp/message.h"
#include "libpspp/misc.h"
#include "libpspp/pool.h"
#include "libpspp/str.h"
#include "libpspp/stringi-set.h"

#include "gl/c-strtod.h"
#include "gl/c-ctype.h"
#include "gl/inttostr.h"
#include "gl/localcharset.h"
#include "gl/minmax.h"
#include "gl/unlocked-io.h"
#include "gl/xalloc.h"
#include "gl/xsize.h"

#include "gettext.h"
#define _(msgid) gettext (msgid)
#define N_(msgid) (msgid)

enum
  {
    /* subtypes 0-2 unknown */
    EXT_INTEGER       = 3,      /* Machine integer info. */
    EXT_FLOAT         = 4,      /* Machine floating-point info. */
    EXT_VAR_SETS      = 5,      /* Variable sets. */
    EXT_DATE          = 6,      /* DATE. */
    EXT_MRSETS        = 7,      /* Multiple response sets. */
    EXT_DATA_ENTRY    = 8,      /* SPSS Data Entry. */
    /* subtypes 9-10 unknown */
    EXT_DISPLAY       = 11,     /* Variable display parameters. */
    /* subtype 12 unknown */
    EXT_LONG_NAMES    = 13,     /* Long variable names. */
    EXT_LONG_STRINGS  = 14,     /* Long strings. */
    /* subtype 15 unknown */
    EXT_NCASES        = 16,     /* Extended number of cases. */
    EXT_FILE_ATTRS    = 17,     /* Data file attributes. */
    EXT_VAR_ATTRS     = 18,     /* Variable attributes. */
    EXT_MRSETS2       = 19,     /* Multiple response sets (extended). */
    EXT_ENCODING      = 20,     /* Character encoding. */
    EXT_LONG_LABELS   = 21,     /* Value labels for long strings. */
    EXT_DATAVIEW      = 24      /* "Format properties in dataview table". */
  };

/* Fields from the top-level header record. */
struct sfm_header_record
  {
    char magic[5];              /* First 4 bytes of file, then null. */
    int weight_idx;             /* 0 if unweighted, otherwise a var index. */
    int nominal_case_size;      /* Number of var positions. */

    /* These correspond to the members of struct sfm_file_info or a dictionary
       but in the system file's encoding rather than ASCII. */
    char creation_date[10];     /* "dd mmm yy". */
    char creation_time[9];      /* "hh:mm:ss". */
    char eye_catcher[61];       /* Eye-catcher string, then product name. */
    char file_label[65];        /* File label. */
  };

struct sfm_var_record
  {
    off_t pos;
    int width;
    char name[8];
    int print_format;
    int write_format;
    int missing_value_code;
    uint8_t missing[24];
    char *label;
    struct variable *var;
  };

struct sfm_value_label
  {
    uint8_t value[8];
    char *label;
  };

struct sfm_value_label_record
  {
    off_t pos;
    struct sfm_value_label *labels;
    size_t n_labels;

    int *vars;
    size_t n_vars;
  };

struct sfm_document_record
  {
    off_t pos;
    char *documents;
    size_t n_lines;
  };

struct sfm_extension_record
  {
    off_t pos;                  /* Starting offset in file. */
    size_t size;                /* Size of data elements. */
    size_t count;               /* Number of data elements. */
    void *data;                 /* Contents. */
  };

/* System file reader. */
struct sfm_reader
  {
    /* Resource tracking. */
    struct pool *pool;          /* All system file state. */
    jmp_buf bail_out;           /* longjmp() target for error handling. */

    /* File state. */
    struct file_handle *fh;     /* File handle. */
    struct fh_lock *lock;       /* Mutual exclusion for file handle. */
    FILE *file;                 /* File stream. */
    off_t pos;                  /* Position in file. */
    bool error;                 /* I/O or corruption error? */
    struct caseproto *proto;    /* Format of output cases. */

    /* File format. */
    enum integer_format integer_format; /* On-disk integer format. */
    enum float_format float_format; /* On-disk floating point format. */
    struct sfm_var *sfm_vars;   /* Variables. */
    size_t sfm_var_cnt;         /* Number of variables. */
    casenumber case_cnt;        /* Number of cases */
    const char *encoding;       /* String encoding. */

    /* Decompression. */
    bool compressed;            /* File is compressed? */
    double bias;                /* Compression bias, usually 100.0. */
    uint8_t opcodes[8];         /* Current block of opcodes. */
    size_t opcode_idx;          /* Next opcode to interpret, 8 if none left. */
    bool corruption_warning;    /* Warned about possible corruption? */
  };

static const struct casereader_class sys_file_casereader_class;

static bool close_reader (struct sfm_reader *);

static struct variable *lookup_var_by_index (struct sfm_reader *, off_t,
                                             const struct sfm_var_record *,
                                             size_t n, int idx);

static void sys_msg (struct sfm_reader *r, off_t, int class,
                     const char *format, va_list args)
     PRINTF_FORMAT (4, 0);
static void sys_warn (struct sfm_reader *, off_t, const char *, ...)
     PRINTF_FORMAT (3, 4);
static void sys_error (struct sfm_reader *, off_t, const char *, ...)
     PRINTF_FORMAT (3, 4)
     NO_RETURN;

static void read_bytes (struct sfm_reader *, void *, size_t);
static bool try_read_bytes (struct sfm_reader *, void *, size_t);
static int read_int (struct sfm_reader *);
static double read_float (struct sfm_reader *);
static void read_string (struct sfm_reader *, char *, size_t);
static void skip_bytes (struct sfm_reader *, size_t);

static int parse_int (struct sfm_reader *, const void *data, size_t ofs);
static double parse_float (struct sfm_reader *, const void *data, size_t ofs);

static void read_variable_record (struct sfm_reader *,
                                  struct sfm_var_record *);
static void read_value_label_record (struct sfm_reader *,
                                     struct sfm_value_label_record *,
                                     size_t n_vars);
static struct sfm_document_record *read_document_record (struct sfm_reader *);
static struct sfm_extension_record *read_extension_record (
  struct sfm_reader *, int subtype);
static void skip_extension_record (struct sfm_reader *, int subtype);

static const char *choose_encoding (
  struct sfm_reader *,
  const struct sfm_header_record *,
  const struct sfm_extension_record *ext_integer,
  const struct sfm_extension_record *ext_encoding);

static struct text_record *open_text_record (
  struct sfm_reader *, const struct sfm_extension_record *,
  bool recode_to_utf8);
static void close_text_record (struct sfm_reader *,
                               struct text_record *);
static bool read_variable_to_value_pair (struct sfm_reader *,
                                         struct dictionary *,
                                         struct text_record *,
                                         struct variable **var, char **value);
static void text_warn (struct sfm_reader *r, struct text_record *text,
                       const char *format, ...)
  PRINTF_FORMAT (3, 4);
static char *text_get_token (struct text_record *,
                             struct substring delimiters, char *delimiter);
static bool text_match (struct text_record *, char c);
static bool text_read_variable_name (struct sfm_reader *, struct dictionary *,
                                     struct text_record *,
                                     struct substring delimiters,
                                     struct variable **);
static bool text_read_short_name (struct sfm_reader *, struct dictionary *,
                                  struct text_record *,
                                  struct substring delimiters,
                                  struct variable **);
static const char *text_parse_counted_string (struct sfm_reader *,
                                              struct text_record *);
static size_t text_pos (const struct text_record *);

static bool close_reader (struct sfm_reader *r);
\f
/* Dictionary reader. */

enum which_format
  {
    PRINT_FORMAT,
    WRITE_FORMAT
  };

static void read_header (struct sfm_reader *, struct sfm_read_info *,
                         struct sfm_header_record *);
static void parse_header (struct sfm_reader *,
                          const struct sfm_header_record *,
                          struct sfm_read_info *, struct dictionary *);
static void parse_variable_records (struct sfm_reader *, struct dictionary *,
                                    struct sfm_var_record *, size_t n);
static void parse_format_spec (struct sfm_reader *, off_t pos,
                               unsigned int format, enum which_format,
                               struct variable *, int *format_warning_cnt);
static void parse_document (struct dictionary *, struct sfm_document_record *);
static void parse_display_parameters (struct sfm_reader *,
                                      const struct sfm_extension_record *,
                                      struct dictionary *);
static void parse_machine_integer_info (struct sfm_reader *,
                                        const struct sfm_extension_record *,
                                        struct sfm_read_info *);
static void parse_machine_float_info (struct sfm_reader *,
                                      const struct sfm_extension_record *);
static void parse_mrsets (struct sfm_reader *,
                          const struct sfm_extension_record *,
                          struct dictionary *);
static void parse_long_var_name_map (struct sfm_reader *,
                                     const struct sfm_extension_record *,
                                     struct dictionary *);
static void parse_long_string_map (struct sfm_reader *,
                                   const struct sfm_extension_record *,
                                   struct dictionary *);
static void parse_value_labels (struct sfm_reader *, struct dictionary *,
                                const struct sfm_var_record *,
                                size_t n_var_recs,
                                const struct sfm_value_label_record *);
static void parse_data_file_attributes (struct sfm_reader *,
                                        const struct sfm_extension_record *,
                                        struct dictionary *);
static void parse_variable_attributes (struct sfm_reader *,
                                       const struct sfm_extension_record *,
                                       struct dictionary *);
static void parse_long_string_value_labels (struct sfm_reader *,
                                            const struct sfm_extension_record *,
                                            struct dictionary *);

/* Frees the strings inside INFO. */
void
sfm_read_info_destroy (struct sfm_read_info *info)
{
  if (info)
    {
      free (info->creation_date);
      free (info->creation_time);
      free (info->product);
    }
}

/* Opens the system file designated by file handle FH for reading.  Reads the
   system file's dictionary into *DICT.

   Ordinarily the reader attempts to automatically detect the character
   encoding based on the file's contents.  This isn't always possible,
   especially for files written by old versions of SPSS or PSPP, so specifying
   a nonnull ENCODING overrides the choice of character encoding.

   If INFO is non-null, then it receives additional info about the system file,
   which the caller must eventually free with sfm_read_info_destroy() when it
   is no longer needed. */
struct casereader *
sfm_open_reader (struct file_handle *fh, const char *volatile encoding,
                 struct dictionary **dictp, struct sfm_read_info *infop)
{
  struct sfm_reader *volatile r = NULL;
  struct sfm_read_info *volatile info;

  struct sfm_header_record header;

  struct sfm_var_record *vars;
  size_t n_vars, allocated_vars;

  struct sfm_value_label_record *labels;
  size_t n_labels, allocated_labels;

  struct sfm_document_record *document;

  struct sfm_extension_record *extensions[32];

  struct dictionary *volatile dict = NULL;
  size_t i;

  /* Create and initialize reader. */
  r = pool_create_container (struct sfm_reader, pool);
  r->fh = fh_ref (fh);
  r->lock = NULL;
  r->file = NULL;
  r->pos = 0;
  r->error = false;
  r->opcode_idx = sizeof r->opcodes;
  r->corruption_warning = false;

  info = infop ? infop : xmalloc (sizeof *info);
  memset (info, 0, sizeof *info);

  /* TRANSLATORS: this fragment will be interpolated into
     messages in fh_lock() that identify types of files. */
  r->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_READ, false);
  if (r->lock == NULL)
    goto error;

  r->file = fn_open (fh_get_file_name (fh), "rb");
  if (r->file == NULL)
    {
      msg (ME, _("Error opening `%s' for reading as a system file: %s."),
           fh_get_file_name (r->fh), strerror (errno));
      goto error;
    }

  if (setjmp (r->bail_out))
    goto error;

  /* Read header. */
  read_header (r, info, &header);

  vars = NULL;
  n_vars = allocated_vars = 0;

  labels = NULL;
  n_labels = allocated_labels = 0;

  document = NULL;

  memset (extensions, 0, sizeof extensions);

  for (;;)
    {
      int subtype;
      int type;

      type = read_int (r);
      if (type == 999)
        {
          read_int (r);         /* Skip filler. */
          break;
        }

      switch (type)
        {
        case 2:
          if (n_vars >= allocated_vars)
            vars = pool_2nrealloc (r->pool, vars, &allocated_vars,
                                   sizeof *vars);
          read_variable_record (r, &vars[n_vars++]);
          break;

        case 3:
          if (n_labels >= allocated_labels)
            labels = pool_2nrealloc (r->pool, labels, &allocated_labels,
                                     sizeof *labels);
          read_value_label_record (r, &labels[n_labels++], n_vars);
          break;

        case 4:
          /* A Type 4 record is always immediately after a type 3 record,
             so the code for type 3 records reads the type 4 record too. */
          sys_error (r, r->pos, _("Misplaced type 4 record."));

        case 6:
          if (document != NULL)
            sys_error (r, r->pos, _("Duplicate type 6 (document) record."));
          document = read_document_record (r);
          break;

        case 7:
          subtype = read_int (r);
          if (subtype < 0 || subtype >= sizeof extensions / sizeof *extensions)
            {
              sys_warn (r, r->pos,
                        _("Unrecognized record type 7, subtype %d.  Please "
                          "send a copy of this file, and the syntax which "
                          "created it to %s."),
                        subtype, PACKAGE_BUGREPORT);
              skip_extension_record (r, subtype);
            }
          else if (extensions[subtype] != NULL)
            {
              sys_warn (r, r->pos,
                        _("Record type 7, subtype %d found here has the same "
                          "type as the record found near offset 0x%llx.  "
                          "Please send a copy of this file, and the syntax "
                          "which created it to %s."),
                        subtype, (long long int) extensions[subtype]->pos,
                        PACKAGE_BUGREPORT);
              skip_extension_record (r, subtype);
            }
          else
            extensions[subtype] = read_extension_record (r, subtype);
          break;

        default:
          sys_error (r, r->pos, _("Unrecognized record type %d."), type);
          goto error;
        }
    }

  /* Now actually parse what we read.

     First, figure out the correct character encoding, because this determines
     how the rest of the header data is to be interpreted. */
  dict = dict_create (encoding
                      ? encoding
                      : choose_encoding (r, &header, extensions[EXT_INTEGER],
                                         extensions[EXT_ENCODING]));
  r->encoding = dict_get_encoding (dict);

  /* These records don't use variables at all. */
  if (document != NULL)
    parse_document (dict, document);

  if (extensions[EXT_INTEGER] != NULL)
    parse_machine_integer_info (r, extensions[EXT_INTEGER], info);

  if (extensions[EXT_FLOAT] != NULL)
    parse_machine_float_info (r, extensions[EXT_FLOAT]);

  if (extensions[EXT_FILE_ATTRS] != NULL)
    parse_data_file_attributes (r, extensions[EXT_FILE_ATTRS], dict);

  parse_header (r, &header, info, dict);

  /* Parse the variable records, the basis of almost everything else. */
  parse_variable_records (r, dict, vars, n_vars);

  /* Parse value labels and the weight variable immediately after the variable
     records.  These records use indexes into var_recs[], so we must parse them
     before those indexes become invalidated by very long string variables. */
  for (i = 0; i < n_labels; i++)
    parse_value_labels (r, dict, vars, n_vars, &labels[i]);
  if (header.weight_idx != 0)
    {
      struct variable *weight_var;

      weight_var = lookup_var_by_index (r, 76, vars, n_vars,
                                        header.weight_idx);
      if (var_is_numeric (weight_var))
        dict_set_weight (dict, weight_var);
      else
        sys_error (r, -1, _("Weighting variable must be numeric "
                            "(not string variable `%s')."),
                   var_get_name (weight_var));
    }

  if (extensions[EXT_DISPLAY] != NULL)
    parse_display_parameters (r, extensions[EXT_DISPLAY], dict);

  /* The following records use short names, so they need to be parsed before
     parse_long_var_name_map() changes short names to long names. */
  if (extensions[EXT_MRSETS] != NULL)
    parse_mrsets (r, extensions[EXT_MRSETS], dict);

  if (extensions[EXT_MRSETS2] != NULL)
    parse_mrsets (r, extensions[EXT_MRSETS2], dict);

  if (extensions[EXT_LONG_STRINGS] != NULL)
    parse_long_string_map (r, extensions[EXT_LONG_STRINGS], dict);

  /* Now rename variables to their long names. */
  parse_long_var_name_map (r, extensions[EXT_LONG_NAMES], dict);

  /* The following records use long names, so they need to follow renaming. */
  if (extensions[EXT_VAR_ATTRS] != NULL)
    parse_variable_attributes (r, extensions[EXT_VAR_ATTRS], dict);

  if (extensions[EXT_LONG_LABELS] != NULL)
    parse_long_string_value_labels (r, extensions[EXT_LONG_LABELS], dict);

  /* Warn if the actual amount of data per case differs from the
     amount that the header claims.  SPSS version 13 gets this
     wrong when very long strings are involved, so don't warn in
     that case. */
  if (header.nominal_case_size != -1 && header.nominal_case_size != n_vars
      && info->version_major != 13)
    sys_warn (r, -1, _("File header claims %d variable positions but "
                       "%zu were read from file."),
              header.nominal_case_size, n_vars);

  /* Create an index of dictionary variable widths for
     sfm_read_case to use.  We cannot use the `struct variable's
     from the dictionary we created, because the caller owns the
     dictionary and may destroy or modify its variables. */
  sfm_dictionary_to_sfm_vars (dict, &r->sfm_vars, &r->sfm_var_cnt);
  pool_register (r->pool, free, r->sfm_vars);
  r->proto = caseproto_ref_pool (dict_get_proto (dict), r->pool);

  *dictp = dict;
  if (infop != info)
    {
      sfm_read_info_destroy (info);
      free (info);
    }

  return casereader_create_sequential
    (NULL, r->proto,
     r->case_cnt == -1 ? CASENUMBER_MAX: r->case_cnt,
                                       &sys_file_casereader_class, r);

error:
  if (infop != info)
    {
      sfm_read_info_destroy (info);
      free (info);
    }

  close_reader (r);
  dict_destroy (dict);
  *dictp = NULL;
  return NULL;
}

/* Closes a system file after we're done with it.
   Returns true if an I/O error has occurred on READER, false
   otherwise. */
static bool
close_reader (struct sfm_reader *r)
{
  bool error;

  if (r == NULL)
    return true;

  if (r->file)
    {
      if (fn_close (fh_get_file_name (r->fh), r->file) == EOF)
        {
          msg (ME, _("Error closing system file `%s': %s."),
               fh_get_file_name (r->fh), strerror (errno));
          r->error = true;
        }
      r->file = NULL;
    }

  fh_unlock (r->lock);
  fh_unref (r->fh);

  error = r->error;
  pool_destroy (r->pool);

  return !error;
}

/* Destroys READER. */
static void
sys_file_casereader_destroy (struct casereader *reader UNUSED, void *r_)
{
  struct sfm_reader *r = r_;
  close_reader (r);
}

/* Returns true if FILE is an SPSS system file,
   false otherwise. */
bool
sfm_detect (FILE *file)
{
  char magic[5];

  if (fread (magic, 4, 1, file) != 1)
    return false;
  magic[4] = '\0';

  return !strcmp (ASCII_MAGIC, magic) || !strcmp (EBCDIC_MAGIC, magic);
}
\f
/* Reads the global header of the system file.  Initializes *HEADER and *INFO,
   except for the string fields in *INFO, which parse_header() will initialize
   later once the file's encoding is known. */
static void
read_header (struct sfm_reader *r, struct sfm_read_info *info,
             struct sfm_header_record *header)
{
  uint8_t raw_layout_code[4];
  uint8_t raw_bias[8];

  read_string (r, header->magic, sizeof header->magic);
  read_string (r, header->eye_catcher, sizeof header->eye_catcher);

  if (strcmp (ASCII_MAGIC, header->magic)
      && strcmp (EBCDIC_MAGIC, header->magic))
    sys_error (r, 0, _("This is not an SPSS system file."));

  /* Identify integer format. */
  read_bytes (r, raw_layout_code, sizeof raw_layout_code);
  if ((!integer_identify (2, raw_layout_code, sizeof raw_layout_code,
                          &r->integer_format)
       && !integer_identify (3, raw_layout_code, sizeof raw_layout_code,
                             &r->integer_format))
      || (r->integer_format != INTEGER_MSB_FIRST
          && r->integer_format != INTEGER_LSB_FIRST))
    sys_error (r, 64, _("This is not an SPSS system file."));

  header->nominal_case_size = read_int (r);
  if (header->nominal_case_size < 0
      || header->nominal_case_size > INT_MAX / 16)
    header->nominal_case_size = -1;

  r->compressed = read_int (r) != 0;

  header->weight_idx = read_int (r);

  r->case_cnt = read_int (r);
  if ( r->case_cnt > INT_MAX / 2)
    r->case_cnt = -1;

  /* Identify floating-point format and obtain compression bias. */
  read_bytes (r, raw_bias, sizeof raw_bias);
  if (float_identify (100.0, raw_bias, sizeof raw_bias, &r->float_format) == 0)
    {
      uint8_t zero_bias[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };

      if (memcmp (raw_bias, zero_bias, 8))
        sys_warn (r, r->pos - 8,
                  _("Compression bias is not the usual "
                    "value of 100, or system file uses unrecognized "
                    "floating-point format."));
      else
        {
          /* Some software is known to write all-zeros to this
             field.  Such software also writes floating-point
             numbers in the format that we expect by default
             (it seems that all software most likely does, in
             reality), so don't warn in this case. */
        }

      if (r->integer_format == INTEGER_MSB_FIRST)
        r->float_format = FLOAT_IEEE_DOUBLE_BE;
      else
        r->float_format = FLOAT_IEEE_DOUBLE_LE;
    }
  float_convert (r->float_format, raw_bias, FLOAT_NATIVE_DOUBLE, &r->bias);

  read_string (r, header->creation_date, sizeof header->creation_date);
  read_string (r, header->creation_time, sizeof header->creation_time);
  read_string (r, header->file_label, sizeof header->file_label);
  skip_bytes (r, 3);

  info->integer_format = r->integer_format;
  info->float_format = r->float_format;
  info->compressed = r->compressed;
  info->case_cnt = r->case_cnt;
}

/* Reads a variable (type 2) record from R into RECORD. */
static void
read_variable_record (struct sfm_reader *r, struct sfm_var_record *record)
{
  int has_variable_label;

  memset (record, 0, sizeof *record);

  record->pos = r->pos;
  record->width = read_int (r);
  has_variable_label = read_int (r);
  record->missing_value_code = read_int (r);
  record->print_format = read_int (r);
  record->write_format = read_int (r);
  read_bytes (r, record->name, sizeof record->name);

  if (has_variable_label == 1)
    {
      enum { MAX_LABEL_LEN = 255 };
      size_t len, read_len;

      len = read_int (r);

      /* Read up to MAX_LABEL_LEN bytes of label. */
      read_len = MIN (MAX_LABEL_LEN, len);
      record->label = pool_malloc (r->pool, read_len + 1);
      read_string (r, record->label, read_len + 1);

      /* Skip unread label bytes. */
      skip_bytes (r, len - read_len);

      /* Skip label padding up to multiple of 4 bytes. */
      skip_bytes (r, ROUND_UP (len, 4) - len);
    }
  else if (has_variable_label != 0)
    sys_error (r, record->pos,
               _("Variable label indicator field is not 0 or 1."));

  /* Set missing values. */
  if (record->missing_value_code != 0)
    {
      int code = record->missing_value_code;
      if (record->width == 0)
        {
          if (code < -3 || code > 3 || code == -1)
            sys_error (r, record->pos,
                       _("Numeric missing value indicator field is not "
                         "-3, -2, 0, 1, 2, or 3."));
        }
      else
        {
          if (code < 1 || code > 3)
            sys_error (r, record->pos,
                       _("String missing value indicator field is not "
                         "0, 1, 2, or 3."));
        }

      read_bytes (r, record->missing, 8 * abs (code));
    }
}

/* Reads value labels from R into RECORD. */
static void
read_value_label_record (struct sfm_reader *r,
                         struct sfm_value_label_record *record,
                         size_t n_vars)
{
  size_t i;

  /* Read type 3 record. */
  record->pos = r->pos;
  record->n_labels = read_int (r);
  if (record->n_labels > SIZE_MAX / sizeof *record->labels)
    sys_error (r, r->pos - 4, _("Invalid number of labels %zu."),
               record->n_labels);
  record->labels = pool_nmalloc (r->pool, record->n_labels,
                                 sizeof *record->labels);
  for (i = 0; i < record->n_labels; i++)
    {
      struct sfm_value_label *label = &record->labels[i];
      unsigned char label_len;
      size_t padded_len;

      read_bytes (r, label->value, sizeof label->value);

      /* Read label length. */
      read_bytes (r, &label_len, sizeof label_len);
      padded_len = ROUND_UP (label_len + 1, 8);

      /* Read label, padding. */
      label->label = pool_malloc (r->pool, padded_len + 1);
      read_bytes (r, label->label, padded_len - 1);
      label->label[label_len] = '\0';
    }

  /* Read record type of type 4 record. */
  if (read_int (r) != 4)
    sys_error (r, r->pos - 4,
               _("Variable index record (type 4) does not immediately "
                 "follow value label record (type 3) as it should."));

  /* Read number of variables associated with value label from type 4
     record. */
  record->n_vars = read_int (r);
  if (record->n_vars < 1 || record->n_vars > n_vars)
    sys_error (r, r->pos - 4,
               _("Number of variables associated with a value label (%zu) "
                 "is not between 1 and the number of variables (%zu)."),
               record->n_vars, n_vars);
  record->vars = pool_nmalloc (r->pool, record->n_vars, sizeof *record->vars);
  for (i = 0; i < record->n_vars; i++)
    record->vars[i] = read_int (r);
}

/* Reads a document record from R and returns it. */
static struct sfm_document_record *
read_document_record (struct sfm_reader *r)
{
  struct sfm_document_record *record;
  int n_lines;

  record = pool_malloc (r->pool, sizeof *record);
  record->pos = r->pos;

  n_lines = read_int (r);
  if (n_lines <= 0 || n_lines >= INT_MAX / DOC_LINE_LENGTH)
    sys_error (r, record->pos,
               _("Number of document lines (%d) "
                 "must be greater than 0 and less than %d."),
               n_lines, INT_MAX / DOC_LINE_LENGTH);

  record->n_lines = n_lines;
  record->documents = pool_malloc (r->pool, DOC_LINE_LENGTH * n_lines);
  read_bytes (r, record->documents, DOC_LINE_LENGTH * n_lines);

  return record;
}

static void
read_extension_record_header (struct sfm_reader *r, int subtype,
                              struct sfm_extension_record *record)
{
  record->pos = r->pos;
  record->size = read_int (r);
  record->count = read_int (r);

  /* Check that SIZE * COUNT + 1 doesn't overflow.  Adding 1
     allows an extra byte for a null terminator, used by some
     extension processing routines. */
  if (record->size != 0
      && size_overflow_p (xsum (1, xtimes (record->count, record->size))))
    sys_error (r, record->pos, "Record type 7 subtype %d too large.", subtype);
}

/* Reads an extension record from R into RECORD. */
static struct sfm_extension_record *
read_extension_record (struct sfm_reader *r, int subtype)
{
  struct extension_record_type
    {
      int subtype;
      int size;
      int count;
    };

  static const struct extension_record_type types[] =
    {
      /* Implemented record types. */
      { EXT_INTEGER,      4, 8 },
      { EXT_FLOAT,        8, 3 },
      { EXT_MRSETS,       1, 0 },
      { EXT_DISPLAY,      4, 0 },
      { EXT_LONG_NAMES,   1, 0 },
      { EXT_LONG_STRINGS, 1, 0 },
      { EXT_NCASES,       8, 2 },
      { EXT_FILE_ATTRS,   1, 0 },
      { EXT_VAR_ATTRS,    1, 0 },
      { EXT_MRSETS2,      1, 0 },
      { EXT_ENCODING,     1, 0 },
      { EXT_LONG_LABELS,  1, 0 },

      /* Ignored record types. */
      { EXT_VAR_SETS,     0, 0 },
      { EXT_DATE,         0, 0 },
      { EXT_DATA_ENTRY,   0, 0 },
      { EXT_DATAVIEW,     0, 0 },
    };

  const struct extension_record_type *type;
  struct sfm_extension_record *record;
  size_t n_bytes;

  record = pool_malloc (r->pool, sizeof *record);
  read_extension_record_header (r, subtype, record);
  n_bytes = record->count * record->size;

  for (type = types; type < &types[sizeof types / sizeof *types]; type++)
    if (subtype == type->subtype)
      {
        if (type->size > 0 && record->size != type->size)
          sys_warn (r, record->pos,
                    _("Record type 7, subtype %d has bad size %zu "
                      "(expected %d)."), subtype, record->size, type->size);
        else if (type->count > 0 && record->count != type->count)
          sys_warn (r, record->pos,
                    _("Record type 7, subtype %d has bad count %zu "
                      "(expected %d)."), subtype, record->count, type->count);
        else if (type->count == 0 && type->size == 0)
          {
            /* Ignore this record. */
          }
        else
          {
            char *data = pool_malloc (r->pool, n_bytes + 1);
            data[n_bytes] = '\0';

            record->data = data;
            read_bytes (r, record->data, n_bytes);
            return record;
          }

        goto skip;
      }

  sys_warn (r, record->pos,
            _("Unrecognized record type 7, subtype %d.  Please send a "
              "copy of this file, and the syntax which created it to %s."),
            subtype, PACKAGE_BUGREPORT);

skip:
  skip_bytes (r, n_bytes);
  return NULL;
}

static void
skip_extension_record (struct sfm_reader *r, int subtype)
{
  struct sfm_extension_record record;

  read_extension_record_header (r, subtype, &record);
  skip_bytes (r, record.count * record.size);
}

static void
parse_header (struct sfm_reader *r, const struct sfm_header_record *header,
              struct sfm_read_info *info, struct dictionary *dict)
{
  const char *dict_encoding = dict_get_encoding (dict);
  struct substring product;
  struct substring label;

  /* Convert file label to UTF-8 and put it into DICT. */
  label = recode_substring_pool ("UTF-8", dict_encoding,
                                 ss_cstr (header->file_label), r->pool);
  ss_trim (&label, ss_cstr (" "));
  label.string[label.length] = '\0';
  dict_set_label (dict, label.string);

  /* Put creation date and time in UTF-8 into INFO. */
  info->creation_date = recode_string ("UTF-8", dict_encoding,
                                       header->creation_date, -1);
  info->creation_time = recode_string ("UTF-8", dict_encoding,
                                       header->creation_time, -1);

  /* Put product name into INFO, dropping eye-catcher string if present. */
  product = recode_substring_pool ("UTF-8", dict_encoding,
                                   ss_cstr (header->eye_catcher), r->pool);
  ss_match_string (&product, ss_cstr ("@(#) SPSS DATA FILE"));
  ss_trim (&product, ss_cstr (" "));
  info->product = ss_xstrdup (product);
}

/* Reads a variable (type 2) record from R and adds the
   corresponding variable to DICT.
   Also skips past additional variable records for long string
   variables. */
static void
parse_variable_records (struct sfm_reader *r, struct dictionary *dict,
                        struct sfm_var_record *var_recs, size_t n_var_recs)
{
  const char *dict_encoding = dict_get_encoding (dict);
  struct sfm_var_record *rec;
  int n_warnings = 0;

  for (rec = var_recs; rec < &var_recs[n_var_recs]; )
    {
      struct variable *var;
      size_t n_values;
      char *name;
      size_t i;

      name = recode_string_pool ("UTF-8", dict_encoding,
                                 rec->name, 8, r->pool);
      name[strcspn (name, " ")] = '\0';

      if (!dict_id_is_valid (dict, name, false)
          || name[0] == '$' || name[0] == '#')
        sys_error (r, rec->pos, _("Invalid variable name `%s'."), name);

      if (rec->width < 0 || rec->width > 255)
        sys_error (r, rec->pos,
                   _("Bad width %d for variable %s."), rec->width, name);

      var = rec->var = dict_create_var (dict, name, rec->width);
      if (var == NULL)
        sys_error (r, rec->pos, _("Duplicate variable name `%s'."), name);

      /* Set the short name the same as the long name. */
      var_set_short_name (var, 0, name);

      /* Get variable label, if any. */
      if (rec->label)
        {
          char *utf8_label;

          utf8_label = recode_string_pool ("UTF-8", dict_encoding,
                                           rec->label, -1, r->pool);
          var_set_label (var, utf8_label, false);
        }

      /* Set missing values. */
      if (rec->missing_value_code != 0)
        {
          int width = var_get_width (var);
          struct missing_values mv;

          mv_init_pool (r->pool, &mv, width);
          if (var_is_numeric (var))
            {
              bool has_range = rec->missing_value_code < 0;
              int n_discrete = (has_range
                                ? rec->missing_value_code == -3
                                : rec->missing_value_code);
              int ofs = 0;

              if (has_range)
                {
                  double low = parse_float (r, rec->missing, 0);
                  double high = parse_float (r, rec->missing, 8);

                  /* Deal with SPSS 21 change in representation. */
                  if (low == SYSMIS)
                    low = LOWEST;

                  mv_add_range (&mv, low, high);
                  ofs += 16;
                }

              for (i = 0; i < n_discrete; i++)
                {
                  mv_add_num (&mv, parse_float (r, rec->missing, ofs));
                  ofs += 8;
                }
            }
          else
            {
              union value value;

              value_init_pool (r->pool, &value, width);
              value_set_missing (&value, width);
              for (i = 0; i < rec->missing_value_code; i++)
                {
                  uint8_t *s = value_str_rw (&value, width);
                  memcpy (s, rec->missing + 8 * i, MIN (width, 8));
                  mv_add_str (&mv, s);
                }
            }
          var_set_missing_values (var, &mv);
        }

      /* Set formats. */
      parse_format_spec (r, rec->pos + 12, rec->print_format,
                         PRINT_FORMAT, var, &n_warnings);
      parse_format_spec (r, rec->pos + 16, rec->write_format,
                         WRITE_FORMAT, var, &n_warnings);

      /* Account for values.
         Skip long string continuation records, if any. */
      n_values = rec->width == 0 ? 1 : DIV_RND_UP (rec->width, 8);
      for (i = 1; i < n_values; i++)
        if (i + (rec - var_recs) >= n_var_recs || rec[i].width != -1)
          sys_error (r, rec->pos, _("Missing string continuation record."));
      rec += n_values;
    }
}

/* Translates the format spec from sysfile format to internal
   format. */
static void
parse_format_spec (struct sfm_reader *r, off_t pos, unsigned int format,
                   enum which_format which, struct variable *v,
                   int *n_warnings)
{
  const int max_warnings = 8;
  uint8_t raw_type = format >> 16;
  uint8_t w = format >> 8;
  uint8_t d = format;
  struct fmt_spec f;
  bool ok;

  f.w = w;
  f.d = d;

  msg_disable ();
  ok = (fmt_from_io (raw_type, &f.type)
        && fmt_check_output (&f)
        && fmt_check_width_compat (&f, var_get_width (v)));
  msg_enable ();

  if (ok)
    {
      if (which == PRINT_FORMAT)
        var_set_print_format (v, &f);
      else
        var_set_write_format (v, &f);
    }
  else if (format == 0)
    {
      /* Actually observed in the wild.  No point in warning about it. */
    }
  else if (++*n_warnings <= max_warnings)
    {
      if (which == PRINT_FORMAT)
        sys_warn (r, pos, _("Variable %s with width %d has invalid print "
                            "format 0x%x."),
                  var_get_name (v), var_get_width (v), format);
      else
        sys_warn (r, pos, _("Variable %s with width %d has invalid write "
                            "format 0x%x."),
                  var_get_name (v), var_get_width (v), format);

      if (*n_warnings == max_warnings)
        sys_warn (r, -1, _("Suppressing further invalid format warnings."));
    }
}

static void
parse_document (struct dictionary *dict, struct sfm_document_record *record)
{
  const char *p;

  for (p = record->documents;
       p < record->documents + DOC_LINE_LENGTH * record->n_lines;
       p += DOC_LINE_LENGTH)
    {
      struct substring line;

      line = recode_substring_pool ("UTF-8", dict_get_encoding (dict),
                                    ss_buffer (p, DOC_LINE_LENGTH), NULL);
      ss_rtrim (&line, ss_cstr (" "));
      line.string[line.length] = '\0';

      dict_add_document_line (dict, line.string, false);

      ss_dealloc (&line);
    }
}

/* Parses record type 7, subtype 3. */
static void
parse_machine_integer_info (struct sfm_reader *r,
                            const struct sfm_extension_record *record,
                            struct sfm_read_info *info)
{
  int float_representation, expected_float_format;
  int integer_representation, expected_integer_format;

  /* Save version info. */
  info->version_major = parse_int (r, record->data, 0);
  info->version_minor = parse_int (r, record->data, 4);
  info->version_revision = parse_int (r, record->data, 8);

  /* Check floating point format. */
  float_representation = parse_int (r, record->data, 16);
  if (r->float_format == FLOAT_IEEE_DOUBLE_BE
      || r->float_format == FLOAT_IEEE_DOUBLE_LE)
    expected_float_format = 1;
  else if (r->float_format == FLOAT_Z_LONG)
    expected_float_format = 2;
  else if (r->float_format == FLOAT_VAX_G || r->float_format == FLOAT_VAX_D)
    expected_float_format = 3;
  else
    NOT_REACHED ();
  if (float_representation != expected_float_format)
    sys_error (r, record->pos, _("Floating-point representation indicated by "
                 "system file (%d) differs from expected (%d)."),
               float_representation, expected_float_format);

  /* Check integer format. */
  integer_representation = parse_int (r, record->data, 24);
  if (r->integer_format == INTEGER_MSB_FIRST)
    expected_integer_format = 1;
  else if (r->integer_format == INTEGER_LSB_FIRST)
    expected_integer_format = 2;
  else
    NOT_REACHED ();
  if (integer_representation != expected_integer_format)
    sys_warn (r, record->pos,
              _("Integer format indicated by system file (%d) "
                "differs from expected (%d)."),
              integer_representation, expected_integer_format);

}

static const char *
choose_encoding (struct sfm_reader *r,
                 const struct sfm_header_record *header,
                 const struct sfm_extension_record *ext_integer,
                 const struct sfm_extension_record *ext_encoding)
{
  /* The EXT_ENCODING record is a more reliable way to determine dictionary
     encoding. */
  if (ext_encoding)
    return ext_encoding->data;

  /* But EXT_INTEGER is better than nothing as a fallback. */
  if (ext_integer)
    {
      int codepage = parse_int (r, ext_integer->data, 7 * 4);
      const char *encoding;

      switch (codepage)
        {
        case 1:
          return "EBCDIC-US";

        case 2:
        case 3:
          /* These ostensibly mean "7-bit ASCII" and "8-bit ASCII"[sic]
             respectively.  However, there are known to be many files in the wild
             with character code 2, yet have data which are clearly not ASCII.
             Therefore we ignore these values. */
          break;

        case 4:
          return "MS_KANJI";

        default:
          encoding = sys_get_encoding_from_codepage (codepage);
          if (encoding != NULL)
            return encoding;
          break;
        }
    }

  /* If the file magic number is EBCDIC then its character data is too. */
  if (!strcmp (header->magic, EBCDIC_MAGIC))
    return "EBCDIC-US";

  return locale_charset ();
}

/* Parses record type 7, subtype 4. */
static void
parse_machine_float_info (struct sfm_reader *r,
                          const struct sfm_extension_record *record)
{
  double sysmis = parse_float (r, record->data, 0);
  double highest = parse_float (r, record->data, 8);
  double lowest = parse_float (r, record->data, 16);

  if (sysmis != SYSMIS)
    sys_warn (r, record->pos,
              _("File specifies unexpected value %g (%a) as %s, "
                "instead of %g (%a)."),
              sysmis, sysmis, "SYSMIS", SYSMIS, SYSMIS);

  if (highest != HIGHEST)
    sys_warn (r, record->pos,
              _("File specifies unexpected value %g (%a) as %s, "
                "instead of %g (%a)."),
              highest, highest, "HIGHEST", HIGHEST, HIGHEST);

  /* SPSS before version 21 used a unique value just bigger than SYSMIS as
     LOWEST.  SPSS 21 uses SYSMIS for LOWEST, which is OK because LOWEST only
     appears in a context (missing values) where SYSMIS cannot. */
  if (lowest != LOWEST && lowest != SYSMIS)
    sys_warn (r, record->pos,
              _("File specifies unexpected value %g (%a) as %s, "
                "instead of %g (%a) or %g (%a)."),
              lowest, lowest, "LOWEST", LOWEST, LOWEST, SYSMIS, SYSMIS);
}

/* Parses record type 7, subtype 7 or 19. */
static void
parse_mrsets (struct sfm_reader *r, const struct sfm_extension_record *record,
              struct dictionary *dict)
{
  struct text_record *text;
  struct mrset *mrset;

  text = open_text_record (r, record, false);
  for (;;)
    {
      const char *counted = NULL;
      const char *name;
      const char *label;
      struct stringi_set var_names;
      size_t allocated_vars;
      char delimiter;
      int width;

      mrset = xzalloc (sizeof *mrset);

      name = text_get_token (text, ss_cstr ("="), NULL);
      if (name == NULL)
        break;
      mrset->name = recode_string ("UTF-8", r->encoding, name, -1);

      if (mrset->name[0] != '$')
        {
          sys_warn (r, record->pos,
                    _("`%s' does not begin with `$' at offset %zu "
                      "in MRSETS record."), mrset->name, text_pos (text));
          break;
        }

      if (text_match (text, 'C'))
        {
          mrset->type = MRSET_MC;
          if (!text_match (text, ' '))
            {
              sys_warn (r, record->pos,
                        _("Missing space following `%c' at offset %zu "
                          "in MRSETS record."), 'C', text_pos (text));
              break;
            }
        }
      else if (text_match (text, 'D'))
        {
          mrset->type = MRSET_MD;
          mrset->cat_source = MRSET_VARLABELS;
        }
      else if (text_match (text, 'E'))
        {
          char *number;

          mrset->type = MRSET_MD;
          mrset->cat_source = MRSET_COUNTEDVALUES;
          if (!text_match (text, ' '))
            {
              sys_warn (r, record->pos,
                        _("Missing space following `%c' at offset %zu "
                          "in MRSETS record."), 'E',  text_pos (text));
              break;
            }

          number = text_get_token (text, ss_cstr (" "), NULL);
          if (!strcmp (number, "11"))
            mrset->label_from_var_label = true;
          else if (strcmp (number, "1"))
            sys_warn (r, record->pos,
                      _("Unexpected label source value `%s' following `E' "
                        "at offset %zu in MRSETS record."),
                      number, text_pos (text));
        }
      else
        {
          sys_warn (r, record->pos,
                    _("Missing `C', `D', or `E' at offset %zu "
                      "in MRSETS record."),
                    text_pos (text));
          break;
        }

      if (mrset->type == MRSET_MD)
        {
          counted = text_parse_counted_string (r, text);
          if (counted == NULL)
            break;
        }

      label = text_parse_counted_string (r, text);
      if (label == NULL)
        break;
      if (label[0] != '\0')
        mrset->label = recode_string ("UTF-8", r->encoding, label, -1);

      stringi_set_init (&var_names);
      allocated_vars = 0;
      width = INT_MAX;
      do
        {
          const char *raw_var_name;
          struct variable *var;
          char *var_name;

          raw_var_name = text_get_token (text, ss_cstr (" \n"), &delimiter);
          if (raw_var_name == NULL)
            {
              sys_warn (r, record->pos,
                        _("Missing new-line parsing variable names "
                          "at offset %zu in MRSETS record."),
                        text_pos (text));
              break;
            }
          var_name = recode_string ("UTF-8", r->encoding, raw_var_name, -1);

          var = dict_lookup_var (dict, var_name);
          if (var == NULL)
            {
              free (var_name);
              continue;
            }
          if (!stringi_set_insert (&var_names, var_name))
            {
              sys_warn (r, record->pos,
                        _("Duplicate variable name %s "
                          "at offset %zu in MRSETS record."),
                        var_name, text_pos (text));
              free (var_name);
              continue;
            }
          free (var_name);

          if (mrset->label == NULL && mrset->label_from_var_label
              && var_has_label (var))
            mrset->label = xstrdup (var_get_label (var));

          if (mrset->n_vars
              && var_get_type (var) != var_get_type (mrset->vars[0]))
            {
              sys_warn (r, record->pos,
                        _("MRSET %s contains both string and "
                          "numeric variables."), name);
              continue;
            }
          width = MIN (width, var_get_width (var));

          if (mrset->n_vars >= allocated_vars)
            mrset->vars = x2nrealloc (mrset->vars, &allocated_vars,
                                      sizeof *mrset->vars);
          mrset->vars[mrset->n_vars++] = var;
        }
      while (delimiter != '\n');

      if (mrset->n_vars < 2)
        {
          sys_warn (r, record->pos,
                    _("MRSET %s has only %zu variables."), mrset->name,
                    mrset->n_vars);
          mrset_destroy (mrset);
          stringi_set_destroy (&var_names);
          continue;
        }

      if (mrset->type == MRSET_MD)
        {
          mrset->width = width;
          value_init (&mrset->counted, width);
          if (width == 0)
            mrset->counted.f = c_strtod (counted, NULL);
          else
            value_copy_str_rpad (&mrset->counted, width,
                                 (const uint8_t *) counted, ' ');
        }

      dict_add_mrset (dict, mrset);
      mrset = NULL;
      stringi_set_destroy (&var_names);
    }
  mrset_destroy (mrset);
  close_text_record (r, text);
}

/* Read record type 7, subtype 11, which specifies how variables
   should be displayed in GUI environments. */
static void
parse_display_parameters (struct sfm_reader *r,
                         const struct sfm_extension_record *record,
                         struct dictionary *dict)
{
  bool includes_width;
  bool warned = false;
  size_t n_vars;
  size_t ofs;
  size_t i;

  n_vars = dict_get_var_cnt (dict);
  if (record->count == 3 * n_vars)
    includes_width = true;
  else if (record->count == 2 * n_vars)
    includes_width = false;
  else
    {
      sys_warn (r, record->pos,
                _("Extension 11 has bad count %zu (for %zu variables)."),
                record->count, n_vars);
      return;
    }

  ofs = 0;
  for (i = 0; i < n_vars; ++i)
    {
      struct variable *v = dict_get_var (dict, i);
      int measure, width, align;

      measure = parse_int (r, record->data, ofs);
      ofs += 4;

      if (includes_width)
        {
          width = parse_int (r, record->data, ofs);
          ofs += 4;
        }
      else
        width = 0;

      align = parse_int (r, record->data, ofs);
      ofs += 4;

      /* SPSS sometimes seems to set variables' measure to zero. */
      if (0 == measure)
        measure = 1;

      if (measure < 1 || measure > 3 || align < 0 || align > 2)
        {
          if (!warned)
            sys_warn (r, record->pos,
                      _("Invalid variable display parameters for variable "
                        "%zu (%s).  Default parameters substituted."),
                      i, var_get_name (v));
          warned = true;
          continue;
        }

      var_set_measure (v, (measure == 1 ? MEASURE_NOMINAL
                           : measure == 2 ? MEASURE_ORDINAL
                           : MEASURE_SCALE));
      var_set_alignment (v, (align == 0 ? ALIGN_LEFT
                             : align == 1 ? ALIGN_RIGHT
                             : ALIGN_CENTRE));

      /* Older versions (SPSS 9.0) sometimes set the display
         width to zero.  This causes confusion in the GUI, so
         only set the width if it is nonzero. */
      if (width > 0)
        var_set_display_width (v, width);
    }
}

static void
rename_var_and_save_short_names (struct dictionary *dict, struct variable *var,
                                 const char *new_name)
{
  size_t n_short_names;
  char **short_names;
  size_t i;

  /* Renaming a variable may clear its short names, but we
     want to retain them, so we save them and re-set them
     afterward. */
  n_short_names = var_get_short_name_cnt (var);
  short_names = xnmalloc (n_short_names, sizeof *short_names);
  for (i = 0; i < n_short_names; i++)
    {
      const char *s = var_get_short_name (var, i);
      short_names[i] = s != NULL ? xstrdup (s) : NULL;
    }

  /* Set long name. */
  dict_rename_var (dict, var, new_name);

  /* Restore short names. */
  for (i = 0; i < n_short_names; i++)
    {
      var_set_short_name (var, i, short_names[i]);
      free (short_names[i]);
    }
  free (short_names);
}

/* Parses record type 7, subtype 13, which gives the long name that corresponds
   to each short name.  Modifies variable names in DICT accordingly.  */
static void
parse_long_var_name_map (struct sfm_reader *r,
                         const struct sfm_extension_record *record,
                         struct dictionary *dict)
{
  struct text_record *text;
  struct variable *var;
  char *long_name;

  if (record == NULL)
    {
      /* There are no long variable names.  Use the short variable names,
         converted to lowercase, as the long variable names. */
      size_t i;

      for (i = 0; i < dict_get_var_cnt (dict); i++)
        {
          struct variable *var = dict_get_var (dict, i);
          char *new_name;

          new_name = utf8_to_lower (var_get_name (var));
          rename_var_and_save_short_names (dict, var, new_name);
          free (new_name);
        }

      return;
    }

  /* Rename each of the variables, one by one.  (In a correctly constructed
     system file, this cannot create any intermediate duplicate variable names,
     because all of the new variable names are longer than any of the old
     variable names and thus there cannot be any overlaps.) */
  text = open_text_record (r, record, true);
  while (read_variable_to_value_pair (r, dict, text, &var, &long_name))
    {
      /* Validate long name. */
      if (!dict_id_is_valid (dict, long_name, false))
        {
          sys_warn (r, record->pos,
                    _("Long variable mapping from %s to invalid "
                      "variable name `%s'."),
                    var_get_name (var), long_name);
          continue;
        }

      /* Identify any duplicates. */
      if (utf8_strcasecmp (var_get_short_name (var, 0), long_name)
          && dict_lookup_var (dict, long_name) != NULL)
        {
          sys_warn (r, record->pos,
                    _("Duplicate long variable name `%s'."), long_name);
          continue;
        }

      rename_var_and_save_short_names (dict, var, long_name);
    }
  close_text_record (r, text);
}

/* Reads record type 7, subtype 14, which gives the real length
   of each very long string.  Rearranges DICT accordingly. */
static void
parse_long_string_map (struct sfm_reader *r,
                       const struct sfm_extension_record *record,
                       struct dictionary *dict)
{
  struct text_record *text;
  struct variable *var;
  char *length_s;

  text = open_text_record (r, record, true);
  while (read_variable_to_value_pair (r, dict, text, &var, &length_s))
    {
      size_t idx = var_get_dict_index (var);
      long int length;
      int segment_cnt;
      int i;

      /* Get length. */
      length = strtol (length_s, NULL, 10);
      if (length < 1 || length > MAX_STRING)
        {
          sys_warn (r, record->pos,
                    _("%s listed as string of invalid length %s "
                      "in very long string record."),
                    var_get_name (var), length_s);
          continue;
        }

      /* Check segments. */
      segment_cnt = sfm_width_to_segments (length);
      if (segment_cnt == 1)
        {
          sys_warn (r, record->pos,
                    _("%s listed in very long string record with width %s, "
                      "which requires only one segment."),
                    var_get_name (var), length_s);
          continue;
        }
      if (idx + segment_cnt > dict_get_var_cnt (dict))
        sys_error (r, record->pos,
                   _("Very long string %s overflows dictionary."),
                   var_get_name (var));

      /* Get the short names from the segments and check their
         lengths. */
      for (i = 0; i < segment_cnt; i++)
        {
          struct variable *seg = dict_get_var (dict, idx + i);
          int alloc_width = sfm_segment_alloc_width (length, i);
          int width = var_get_width (seg);

          if (i > 0)
            var_set_short_name (var, i, var_get_short_name (seg, 0));
          if (ROUND_UP (width, 8) != ROUND_UP (alloc_width, 8))
            sys_error (r, record->pos,
                       _("Very long string with width %ld has segment %d "
                         "of width %d (expected %d)."),
                       length, i, width, alloc_width);
        }
      dict_delete_consecutive_vars (dict, idx + 1, segment_cnt - 1);
      var_set_width (var, length);
    }
  close_text_record (r, text);
  dict_compact_values (dict);
}

static void
parse_value_labels (struct sfm_reader *r, struct dictionary *dict,
                    const struct sfm_var_record *var_recs, size_t n_var_recs,
                    const struct sfm_value_label_record *record)
{
  struct variable **vars;
  char **utf8_labels;
  size_t i;

  utf8_labels = pool_nmalloc (r->pool, sizeof *utf8_labels, record->n_labels);
  for (i = 0; i < record->n_labels; i++)
    utf8_labels[i] = recode_string_pool ("UTF-8", dict_get_encoding (dict),
                                         record->labels[i].label, -1,
                                         r->pool);

  vars = pool_nmalloc (r->pool, record->n_vars, sizeof *vars);
  for (i = 0; i < record->n_vars; i++)
    vars[i] = lookup_var_by_index (r, record->pos,
                                   var_recs, n_var_recs, record->vars[i]);

  for (i = 1; i < record->n_vars; i++)
    if (var_get_type (vars[i]) != var_get_type (vars[0]))
      sys_error (r, record->pos,
                 _("Variables associated with value label are not all of "
                   "identical type.  Variable %s is %s, but variable "
                   "%s is %s."),
                 var_get_name (vars[0]),
                 var_is_numeric (vars[0]) ? _("numeric") : _("string"),
                 var_get_name (vars[i]),
                 var_is_numeric (vars[i]) ? _("numeric") : _("string"));

  for (i = 0; i < record->n_vars; i++)
    {
      struct variable *var = vars[i];
      int width;
      size_t j;

      width = var_get_width (var);
      if (width > 8)
        sys_error (r, record->pos,
                   _("Value labels may not be added to long string "
                     "variables (e.g. %s) using records types 3 and 4."),
                   var_get_name (var));

      for (j = 0; j < record->n_labels; j++)
        {
          struct sfm_value_label *label = &record->labels[j];
          union value value;

          value_init (&value, width);
          if (width == 0)
            value.f = parse_float (r, label->value, 0);
          else
            memcpy (value_str_rw (&value, width), label->value, width);

          if (!var_add_value_label (var, &value, utf8_labels[j]))
            {
              if (var_is_numeric (var))
                sys_warn (r, record->pos,
                          _("Duplicate value label for %g on %s."),
                          value.f, var_get_name (var));
              else
                sys_warn (r, record->pos,
                          _("Duplicate value label for `%.*s' on %s."),
                          width, value_str (&value, width),
                          var_get_name (var));
            }

          value_destroy (&value, width);
        }
    }

  pool_free (r->pool, vars);
  for (i = 0; i < record->n_labels; i++)
    pool_free (r->pool, utf8_labels[i]);
  pool_free (r->pool, utf8_labels);
}

static struct variable *
lookup_var_by_index (struct sfm_reader *r, off_t offset,
                     const struct sfm_var_record *var_recs, size_t n_var_recs,
                     int idx)
{
  const struct sfm_var_record *rec;

  if (idx < 1 || idx > n_var_recs)
    {
      sys_error (r, offset,
                 _("Variable index %d not in valid range 1...%zu."),
                 idx, n_var_recs);
      return NULL;
    }

  rec = &var_recs[idx - 1];
  if (rec->var == NULL)
    {
      sys_error (r, offset,
                 _("Variable index %d refers to long string continuation."),
                 idx);
      return NULL;
    }

  return rec->var;
}

/* Parses a set of custom attributes from TEXT into ATTRS.
   ATTRS may be a null pointer, in which case the attributes are
   read but discarded. */
static void
parse_attributes (struct sfm_reader *r, struct text_record *text,
                  struct attrset *attrs)
{
  do
    {
      struct attribute *attr;
      char *key;
      int index;

      /* Parse the key. */
      key = text_get_token (text, ss_cstr ("("), NULL);
      if (key == NULL)
        return;

      attr = attribute_create (key);
      for (index = 1; ; index++)
        {
          /* Parse the value. */
          char *value;
          size_t length;

          value = text_get_token (text, ss_cstr ("\n"), NULL);
          if (value == NULL)
            {
              text_warn (r, text, _("Error parsing attribute value %s[%d]."),
                         key, index);
              break;
            }              

          length = strlen (value);
          if (length >= 2 && value[0] == '\'' && value[length - 1] == '\'') 
            {
              value[length - 1] = '\0';
              attribute_add_value (attr, value + 1); 
            }
          else 
            {
              text_warn (r, text,
                         _("Attribute value %s[%d] is not quoted: %s."),
                         key, index, value);
              attribute_add_value (attr, value); 
            }

          /* Was this the last value for this attribute? */
          if (text_match (text, ')'))
            break;
        }
      if (attrs != NULL)
        attrset_add (attrs, attr);
      else
        attribute_destroy (attr);
    }
  while (!text_match (text, '/'));
}

/* Reads record type 7, subtype 17, which lists custom
   attributes on the data file.  */
static void
parse_data_file_attributes (struct sfm_reader *r,
                            const struct sfm_extension_record *record,
                            struct dictionary *dict)
{
  struct text_record *text = open_text_record (r, record, true);
  parse_attributes (r, text, dict_get_attributes (dict));
  close_text_record (r, text);
}

/* Parses record type 7, subtype 18, which lists custom
   attributes on individual variables.  */
static void
parse_variable_attributes (struct sfm_reader *r,
                           const struct sfm_extension_record *record,
                           struct dictionary *dict)
{
  struct text_record *text;
  struct variable *var;

  text = open_text_record (r, record, true);
  while (text_read_variable_name (r, dict, text, ss_cstr (":"), &var))
    parse_attributes (r, text, var != NULL ? var_get_attributes (var) : NULL);
  close_text_record (r, text);
}

static void
check_overflow (struct sfm_reader *r,
                const struct sfm_extension_record *record,
                size_t ofs, size_t length)
{
  size_t end = record->size * record->count;
  if (length >= end || ofs + length > end)
    sys_error (r, record->pos + end,
               _("Long string value label record ends unexpectedly."));
}

static void
parse_long_string_value_labels (struct sfm_reader *r,
                                const struct sfm_extension_record *record,
                                struct dictionary *dict)
{
  const char *dict_encoding = dict_get_encoding (dict);
  size_t end = record->size * record->count;
  size_t ofs = 0;

  while (ofs < end)
    {
      char *var_name;
      size_t n_labels, i;
      struct variable *var;
      union value value;
      int var_name_len;
      int width;

      /* Parse variable name length. */
      check_overflow (r, record, ofs, 4);
      var_name_len = parse_int (r, record->data, ofs);
      ofs += 4;

      /* Parse variable name, width, and number of labels. */
      check_overflow (r, record, ofs, var_name_len + 8);
      var_name = recode_string_pool ("UTF-8", dict_encoding,
                                     (const char *) record->data + ofs,
                                     var_name_len, r->pool);
      width = parse_int (r, record->data, ofs + var_name_len);
      n_labels = parse_int (r, record->data, ofs + var_name_len + 4);
      ofs += var_name_len + 8;

      /* Look up 'var' and validate. */
      var = dict_lookup_var (dict, var_name);
      if (var == NULL)
        sys_warn (r, record->pos + ofs,
                  _("Ignoring long string value record for "
                    "unknown variable %s."), var_name);
      else if (var_is_numeric (var))
        {
          sys_warn (r, record->pos + ofs,
                    _("Ignoring long string value record for "
                      "numeric variable %s."), var_name);
          var = NULL;
        }
      else if (width != var_get_width (var))
        {
          sys_warn (r, record->pos + ofs,
                    _("Ignoring long string value record for variable %s "
                      "because the record's width (%d) does not match the "
                      "variable's width (%d)."),
                    var_name, width, var_get_width (var));
          var = NULL;
        }

      /* Parse values. */
      value_init_pool (r->pool, &value, width);
      for (i = 0; i < n_labels; i++)
        {
          size_t value_length, label_length;
          bool skip = var == NULL;

          /* Parse value length. */
          check_overflow (r, record, ofs, 4);
          value_length = parse_int (r, record->data, ofs);
          ofs += 4;

          /* Parse value. */
          check_overflow (r, record, ofs, value_length);
          if (!skip)
            {
              if (value_length == width)
                memcpy (value_str_rw (&value, width),
                        (const uint8_t *) record->data + ofs, width);
              else
                {
                  sys_warn (r, record->pos + ofs,
                            _("Ignoring long string value %zu for variable "
                              "%s, with width %d, that has bad value "
                              "width %zu."),
                            i, var_get_name (var), width, value_length);
                  skip = true;
                }
            }
          ofs += value_length;

          /* Parse label length. */
          check_overflow (r, record, ofs, 4);
          label_length = parse_int (r, record->data, ofs);
          ofs += 4;

          /* Parse label. */
          check_overflow (r, record, ofs, label_length);
          if (!skip)
            {
              char *label;

              label = recode_string_pool ("UTF-8", dict_encoding,
                                          (const char *) record->data + ofs,
                                          label_length, r->pool);
              if (!var_add_value_label (var, &value, label))
                sys_warn (r, record->pos + ofs,
                          _("Duplicate value label for `%.*s' on %s."),
                          width, value_str (&value, width),
                          var_get_name (var));
              pool_free (r->pool, label);
            }
          ofs += label_length;
        }
    }
}
\f
/* Case reader. */

static void partial_record (struct sfm_reader *r)
     NO_RETURN;

static void read_error (struct casereader *, const struct sfm_reader *);

static bool read_case_number (struct sfm_reader *, double *);
static bool read_case_string (struct sfm_reader *, uint8_t *, size_t);
static int read_opcode (struct sfm_reader *);
static bool read_compressed_number (struct sfm_reader *, double *);
static bool read_compressed_string (struct sfm_reader *, uint8_t *);
static bool read_whole_strings (struct sfm_reader *, uint8_t *, size_t);
static bool skip_whole_strings (struct sfm_reader *, size_t);

/* Reads and returns one case from READER's file.  Returns a null
   pointer if not successful. */
static struct ccase *
sys_file_casereader_read (struct casereader *reader, void *r_)
{
  struct sfm_reader *r = r_;
  struct ccase *volatile c;
  int i;

  if (r->error)
    return NULL;

  c = case_create (r->proto);
  if (setjmp (r->bail_out))
    {
      casereader_force_error (reader);
      case_unref (c);
      return NULL;
    }

  for (i = 0; i < r->sfm_var_cnt; i++)
    {
      struct sfm_var *sv = &r->sfm_vars[i];
      union value *v = case_data_rw_idx (c, sv->case_index);

      if (sv->var_width == 0)
        {
          if (!read_case_number (r, &v->f))
            goto eof;
        }
      else
        {
          uint8_t *s = value_str_rw (v, sv->var_width);
          if (!read_case_string (r, s + sv->offset, sv->segment_width))
            goto eof;
          if (!skip_whole_strings (r, ROUND_DOWN (sv->padding, 8)))
            partial_record (r);
        }
    }
  return c;

eof:
  if (i != 0)
    partial_record (r);
  if (r->case_cnt != -1)
    read_error (reader, r);
  case_unref (c);
  return NULL;
}

/* Issues an error that R ends in a partial record. */
static void
partial_record (struct sfm_reader *r)
{
  sys_error (r, r->pos, _("File ends in partial case."));
}

/* Issues an error that an unspecified error occurred SFM, and
   marks R tainted. */
static void
read_error (struct casereader *r, const struct sfm_reader *sfm)
{
  msg (ME, _("Error reading case from file %s."), fh_get_name (sfm->fh));
  casereader_force_error (r);
}

/* Reads a number from R and stores its value in *D.
   If R is compressed, reads a compressed number;
   otherwise, reads a number in the regular way.
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
read_case_number (struct sfm_reader *r, double *d)
{
  if (!r->compressed)
    {
      uint8_t number[8];
      if (!try_read_bytes (r, number, sizeof number))
        return false;
      float_convert (r->float_format, number, FLOAT_NATIVE_DOUBLE, d);
      return true;
    }
  else
    return read_compressed_number (r, d);
}

/* Reads LENGTH string bytes from R into S.
   Always reads a multiple of 8 bytes; if LENGTH is not a
   multiple of 8, then extra bytes are read and discarded without
   being written to S.
   Reads compressed strings if S is compressed.
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
read_case_string (struct sfm_reader *r, uint8_t *s, size_t length)
{
  size_t whole = ROUND_DOWN (length, 8);
  size_t partial = length % 8;

  if (whole)
    {
      if (!read_whole_strings (r, s, whole))
        return false;
    }

  if (partial)
    {
      uint8_t bounce[8];
      if (!read_whole_strings (r, bounce, sizeof bounce))
        {
          if (whole)
            partial_record (r);
          return false;
        }
      memcpy (s + whole, bounce, partial);
    }

  return true;
}

/* Reads and returns the next compression opcode from R. */
static int
read_opcode (struct sfm_reader *r)
{
  assert (r->compressed);
  for (;;)
    {
      int opcode;
      if (r->opcode_idx >= sizeof r->opcodes)
        {
          if (!try_read_bytes (r, r->opcodes, sizeof r->opcodes))
            return -1;
          r->opcode_idx = 0;
        }
      opcode = r->opcodes[r->opcode_idx++];

      if (opcode != 0)
        return opcode;
    }
}

/* Reads a compressed number from R and stores its value in D.
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
read_compressed_number (struct sfm_reader *r, double *d)
{
  int opcode = read_opcode (r);
  switch (opcode)
    {
    case -1:
    case 252:
      return false;

    case 253:
      *d = read_float (r);
      break;

    case 254:
      float_convert (r->float_format, "        ", FLOAT_NATIVE_DOUBLE, d);
      if (!r->corruption_warning)
        {
          r->corruption_warning = true;
          sys_warn (r, r->pos,
                    _("Possible compressed data corruption: "
                      "compressed spaces appear in numeric field."));
        }
      break;

    case 255:
      *d = SYSMIS;
      break;

    default:
      *d = opcode - r->bias;
      break;
    }

  return true;
}

/* Reads a compressed 8-byte string segment from R and stores it
   in DST.
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
read_compressed_string (struct sfm_reader *r, uint8_t *dst)
{
  int opcode = read_opcode (r);
  switch (opcode)
    {
    case -1:
    case 252:
      return false;

    case 253:
      read_bytes (r, dst, 8);
      break;

    case 254:
      memset (dst, ' ', 8);
      break;

    default:
      {
        double value = opcode - r->bias;
        float_convert (FLOAT_NATIVE_DOUBLE, &value, r->float_format, dst);
        if (value == 0.0)
          {
            /* This has actually been seen "in the wild".  The submitter of the
               file that showed that the contents decoded as spaces, but they
               were at the end of the field so it's possible that the null
               bytes just acted as null terminators. */
          }
        else if (!r->corruption_warning)
          {
            r->corruption_warning = true;
            sys_warn (r, r->pos,
                      _("Possible compressed data corruption: "
                        "string contains compressed integer (opcode %d)."),
                      opcode);
          }
      }
      break;
    }

  return true;
}

/* Reads LENGTH string bytes from R into S.
   LENGTH must be a multiple of 8.
   Reads compressed strings if S is compressed.
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
read_whole_strings (struct sfm_reader *r, uint8_t *s, size_t length)
{
  assert (length % 8 == 0);
  if (!r->compressed)
    return try_read_bytes (r, s, length);
  else
    {
      size_t ofs;
      for (ofs = 0; ofs < length; ofs += 8)
        if (!read_compressed_string (r, s + ofs))
          {
            if (ofs != 0)
              partial_record (r);
            return false;
          }
      return true;
    }
}

/* Skips LENGTH string bytes from R.
   LENGTH must be a multiple of 8.
   (LENGTH is also limited to 1024, but that's only because the
   current caller never needs more than that many bytes.)
   Returns true if successful, false if end of file is
   reached immediately. */
static bool
skip_whole_strings (struct sfm_reader *r, size_t length)
{
  uint8_t buffer[1024];
  assert (length < sizeof buffer);
  return read_whole_strings (r, buffer, length);
}
\f
/* Helpers for reading records that contain structured text
   strings. */

/* Maximum number of warnings to issue for a single text
   record. */
#define MAX_TEXT_WARNINGS 5

/* State. */
struct text_record
  {
    struct substring buffer;    /* Record contents. */
    off_t start;                /* Starting offset in file. */
    size_t pos;                 /* Current position in buffer. */
    int n_warnings;             /* Number of warnings issued or suppressed. */
    bool recoded;               /* Recoded into UTF-8? */
  };

static struct text_record *
open_text_record (struct sfm_reader *r,
                  const struct sfm_extension_record *record,
                  bool recode_to_utf8)
{
  struct text_record *text;
  struct substring raw;

  text = pool_alloc (r->pool, sizeof *text);
  raw = ss_buffer (record->data, record->size * record->count);
  text->start = record->pos;
  text->buffer = (recode_to_utf8
                  ? recode_substring_pool ("UTF-8", r->encoding, raw, r->pool)
                  : raw);
  text->pos = 0;
  text->n_warnings = 0;
  text->recoded = recode_to_utf8;

  return text;
}

/* Closes TEXT, frees its storage, and issues a final warning
   about suppressed warnings if necesary. */
static void
close_text_record (struct sfm_reader *r, struct text_record *text)
{
  if (text->n_warnings > MAX_TEXT_WARNINGS)
    sys_warn (r, -1, _("Suppressed %d additional related warnings."),
              text->n_warnings - MAX_TEXT_WARNINGS);
  if (text->recoded)
    pool_free (r->pool, ss_data (text->buffer));
}

/* Reads a variable=value pair from TEXT.
   Looks up the variable in DICT and stores it into *VAR.
   Stores a null-terminated value into *VALUE. */
static bool
read_variable_to_value_pair (struct sfm_reader *r, struct dictionary *dict,
                             struct text_record *text,
                             struct variable **var, char **value)
{
  for (;;)
    {
      if (!text_read_short_name (r, dict, text, ss_cstr ("="), var))
        return false;
      
      *value = text_get_token (text, ss_buffer ("\t\0", 2), NULL);
      if (*value == NULL)
        return false;

      text->pos += ss_span (ss_substr (text->buffer, text->pos, SIZE_MAX),
                            ss_buffer ("\t\0", 2));

      if (*var != NULL)
        return true;
    }
}

static bool
text_read_variable_name (struct sfm_reader *r, struct dictionary *dict,
                         struct text_record *text, struct substring delimiters,
                         struct variable **var)
{
  char *name;

  name = text_get_token (text, delimiters, NULL);
  if (name == NULL)
    return false;

  *var = dict_lookup_var (dict, name);
  if (*var != NULL)
    return true;

  text_warn (r, text, _("Dictionary record refers to unknown variable %s."),
             name);
  return false;
}


static bool
text_read_short_name (struct sfm_reader *r, struct dictionary *dict,
                      struct text_record *text, struct substring delimiters,
                      struct variable **var)
{
  char *short_name = text_get_token (text, delimiters, NULL);
  if (short_name == NULL)
    return false;

  *var = dict_lookup_var (dict, short_name);
  if (*var == NULL)
    text_warn (r, text, _("Dictionary record refers to unknown variable %s."),
               short_name);
  return true;
}

/* Displays a warning for the current file position, limiting the
   number to MAX_TEXT_WARNINGS for TEXT. */
static void
text_warn (struct sfm_reader *r, struct text_record *text,
           const char *format, ...)
{
  if (text->n_warnings++ < MAX_TEXT_WARNINGS) 
    {
      va_list args;

      va_start (args, format);
      sys_msg (r, text->start + text->pos, MW, format, args);
      va_end (args);
    }
}

static char *
text_get_token (struct text_record *text, struct substring delimiters,
                char *delimiter)
{
  struct substring token;
  char *end;

  if (!ss_tokenize (text->buffer, delimiters, &text->pos, &token))
    return NULL;

  end = &ss_data (token)[ss_length (token)];
  if (delimiter != NULL)
    *delimiter = *end;
  *end = '\0';
  return ss_data (token);
}

/* Reads a integer value expressed in decimal, then a space, then a string that
   consists of exactly as many bytes as specified by the integer, then a space,
   from TEXT.  Returns the string, null-terminated, as a subset of TEXT's
   buffer (so the caller should not free the string). */
static const char *
text_parse_counted_string (struct sfm_reader *r, struct text_record *text)
{
  size_t start;
  size_t n;
  char *s;

  start = text->pos;
  n = 0;
  while (text->pos < text->buffer.length)
    {
      int c = text->buffer.string[text->pos];
      if (c < '0' || c > '9')
        break;
      n = (n * 10) + (c - '0');
      text->pos++;
    }
  if (text->pos >= text->buffer.length || start == text->pos)
    {
      sys_warn (r, text->start,
                _("Expecting digit at offset %zu in MRSETS record."),
                text->pos);
      return NULL;
    }

  if (!text_match (text, ' '))
    {
      sys_warn (r, text->start,
                _("Expecting space at offset %zu in MRSETS record."),
                text->pos);
      return NULL;
    }

  if (text->pos + n > text->buffer.length)
    {
      sys_warn (r, text->start,
                _("%zu-byte string starting at offset %zu "
                  "exceeds record length %zu."),
                n, text->pos, text->buffer.length);
      return NULL;
    }

  s = &text->buffer.string[text->pos];
  if (s[n] != ' ')
    {
      sys_warn (r, text->start,
                _("Expecting space at offset %zu following %zu-byte string."),
                text->pos + n, n);
      return NULL;
    }
  s[n] = '\0';
  text->pos += n + 1;
  return s;
}

static bool
text_match (struct text_record *text, char c)
{
  if (text->buffer.string[text->pos] == c) 
    {
      text->pos++;
      return true;
    }
  else
    return false;
}

/* Returns the current byte offset (as converted to UTF-8, if it was converted)
   inside the TEXT's string. */
static size_t
text_pos (const struct text_record *text)
{
  return text->pos;
}
\f
/* Messages. */

/* Displays a corruption message. */
static void
sys_msg (struct sfm_reader *r, off_t offset,
         int class, const char *format, va_list args)
{
  struct msg m;
  struct string text;

  ds_init_empty (&text);
  if (offset >= 0)
    ds_put_format (&text, _("`%s' near offset 0x%llx: "),
                   fh_get_file_name (r->fh), (long long int) offset);
  else
    ds_put_format (&text, _("`%s': "), fh_get_file_name (r->fh));
  ds_put_vformat (&text, format, args);

  m.category = msg_class_to_category (class);
  m.severity = msg_class_to_severity (class);
  m.file_name = NULL;
  m.first_line = 0;
  m.last_line = 0;
  m.first_column = 0;
  m.last_column = 0;
  m.text = ds_cstr (&text);

  msg_emit (&m);
}

/* Displays a warning for offset OFFSET in the file. */
static void
sys_warn (struct sfm_reader *r, off_t offset, const char *format, ...)
{
  va_list args;

  va_start (args, format);
  sys_msg (r, offset, MW, format, args);
  va_end (args);
}

/* Displays an error for the current file position,
   marks it as in an error state,
   and aborts reading it using longjmp. */
static void
sys_error (struct sfm_reader *r, off_t offset, const char *format, ...)
{
  va_list args;

  va_start (args, format);
  sys_msg (r, offset, ME, format, args);
  va_end (args);

  r->error = true;
  longjmp (r->bail_out, 1);
}
\f
/* Reads BYTE_CNT bytes into BUF.
   Returns true if exactly BYTE_CNT bytes are successfully read.
   Aborts if an I/O error or a partial read occurs.
   If EOF_IS_OK, then an immediate end-of-file causes false to be
   returned; otherwise, immediate end-of-file causes an abort
   too. */
static inline bool
read_bytes_internal (struct sfm_reader *r, bool eof_is_ok,
                   void *buf, size_t byte_cnt)
{
  size_t bytes_read = fread (buf, 1, byte_cnt, r->file);
  r->pos += bytes_read;
  if (bytes_read == byte_cnt)
    return true;
  else if (ferror (r->file))
    sys_error (r, r->pos, _("System error: %s."), strerror (errno));
  else if (!eof_is_ok || bytes_read != 0)
    sys_error (r, r->pos, _("Unexpected end of file."));
  else
    return false;
}

/* Reads BYTE_CNT into BUF.
   Aborts upon I/O error or if end-of-file is encountered. */
static void
read_bytes (struct sfm_reader *r, void *buf, size_t byte_cnt)
{
  read_bytes_internal (r, false, buf, byte_cnt);
}

/* Reads BYTE_CNT bytes into BUF.
   Returns true if exactly BYTE_CNT bytes are successfully read.
   Returns false if an immediate end-of-file is encountered.
   Aborts if an I/O error or a partial read occurs. */
static bool
try_read_bytes (struct sfm_reader *r, void *buf, size_t byte_cnt)
{
  return read_bytes_internal (r, true, buf, byte_cnt);
}

/* Reads a 32-bit signed integer from R and returns its value in
   host format. */
static int
read_int (struct sfm_reader *r)
{
  uint8_t integer[4];
  read_bytes (r, integer, sizeof integer);
  return integer_get (r->integer_format, integer, sizeof integer);
}

/* Reads a 64-bit floating-point number from R and returns its
   value in host format. */
static double
read_float (struct sfm_reader *r)
{
  uint8_t number[8];
  read_bytes (r, number, sizeof number);
  return float_get_double (r->float_format, number);
}

static int
parse_int (struct sfm_reader *r, const void *data, size_t ofs)
{
  return integer_get (r->integer_format, (const uint8_t *) data + ofs, 4);
}

static double
parse_float (struct sfm_reader *r, const void *data, size_t ofs)
{
  return float_get_double (r->float_format, (const uint8_t *) data + ofs);
}

/* Reads exactly SIZE - 1 bytes into BUFFER
   and stores a null byte into BUFFER[SIZE - 1]. */
static void
read_string (struct sfm_reader *r, char *buffer, size_t size)
{
  assert (size > 0);
  read_bytes (r, buffer, size - 1);
  buffer[size - 1] = '\0';
}

/* Skips BYTES bytes forward in R. */
static void
skip_bytes (struct sfm_reader *r, size_t bytes)
{
  while (bytes > 0)
    {
      char buffer[1024];
      size_t chunk = MIN (sizeof buffer, bytes);
      read_bytes (r, buffer, chunk);
      bytes -= chunk;
    }
}
\f
static const struct casereader_class sys_file_casereader_class =
  {
    sys_file_casereader_read,
    sys_file_casereader_destroy,
    NULL,
    NULL,
  };