[pspp] / src / sfm-read.c

/* PSPP - computes sample statistics.
   Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
   Written by Ben Pfaff <blp@gnu.org>.

   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 2 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, write to the Free Software
   Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
   02111-1307, USA. */

/* AIX requires this to be the first thing in the file.  */
#include <config.h>
#if __GNUC__
#define alloca __builtin_alloca
#else
#if HAVE_ALLOCA_H
#include <alloca.h>
#else
#ifdef _AIX
#pragma alloca
#else
#ifndef alloca                  /* predefined by HP cc +Olibcalls */
char *alloca ();
#endif
#endif
#endif
#endif

#include <assert.h>
#include <stdlib.h>
#include <ctype.h>
#include <errno.h>
#include <float.h>
#include "alloc.h"
#include "error.h"
#include "file-handle.h"
#include "filename.h"
#include "format.h"
#include "getline.h"
#include "hash.h"
#include "magic.h"
#include "misc.h"
#include "sfm.h"
#include "sfmP.h"
#include "value-labels.h"
#include "str.h"
#include "var.h"

#include "debug-print.h"

/* PORTME: This file may require substantial revision for those
   systems that don't meet the typical 32-bit integer/64-bit double
   model.  It's kinda hard to tell without having one of them on my
   desk.  */

/* sfm's file_handle extension. */
struct sfm_fhuser_ext
  {
    FILE *file;                 /* Actual file. */
    int opened;                 /* Reference count. */

    struct dictionary *dict;    /* File's dictionary. */

    int reverse_endian;         /* 1=file has endianness opposite us. */
    int case_size;              /* Number of `values's per case. */
    long ncases;                /* Number of cases, -1 if unknown. */
    int compressed;             /* 1=compressed, 0=not compressed. */
    double bias;                /* Compression bias, usually 100.0. */
    int weight_index;           /* 0-based index of weighting variable, or -1. */

    /* File's special constants. */
    flt64 sysmis;
    flt64 highest;
    flt64 lowest;

    /* Uncompression buffer. */
    flt64 *buf;                 /* Buffer data. */
    flt64 *ptr;                 /* Current location in buffer. */
    flt64 *end;                 /* End of buffer data. */

    /* Compression instruction octet. */
    unsigned char x[sizeof (flt64)];
    /* Current instruction octet. */
    unsigned char *y;           /* Location in current instruction octet. */
  };

static struct fh_ext_class sfm_r_class;

#if GLOBAL_DEBUGGING
void dump_dictionary (struct dictionary * dict);
#endif
\f
/* Utilities. */

/* bswap_int32(): Reverse the byte order of 32-bit integer *X. */
static inline void
bswap_int32 (int32 *x)
{
  unsigned char *y = (unsigned char *) x;
  unsigned char t;

  t = y[0];
  y[0] = y[3];
  y[3] = t;

  t = y[1];
  y[1] = y[2];
  y[2] = t;
}

/* Reverse the byte order of 64-bit floating point *X. */
static inline void
bswap_flt64 (flt64 *x)
{
  unsigned char *y = (unsigned char *) x;
  unsigned char t;

  t = y[0];
  y[0] = y[7];
  y[7] = t;

  t = y[1];
  y[1] = y[6];
  y[6] = t;

  t = y[2];
  y[2] = y[5];
  y[5] = t;

  t = y[3];
  y[3] = y[4];
  y[4] = t;
}

static void
corrupt_msg (int class, const char *format,...)
  __attribute__ ((format (printf, 2, 3)));

/* Displays a corrupt sysfile error. */
static void
corrupt_msg (int class, const char *format,...)
{
  char buf[1024];
  
  {
    va_list args;

    va_start (args, format);
    vsnprintf (buf, 1024, format, args);
    va_end (args);
  }
  
  {
    struct error e;

    e.class = class;
    getl_location (&e.where.filename, &e.where.line_number);
    e.title = _("corrupt system file: ");
    e.text = buf;

    err_vmsg (&e);
  }
}

/* Closes a system file after we're done with it. */
static void
sfm_close (struct file_handle * h)
{
  struct sfm_fhuser_ext *ext = h->ext;

  ext->opened--;
  assert (ext->opened == 0);
  if (EOF == fn_close (h->fn, ext->file))
    msg (ME, _("%s: Closing system file: %s."), h->fn, strerror (errno));
  free (ext->buf);
  free (h->ext);
}

/* Closes a system file if we're done with it. */
void
sfm_maybe_close (struct file_handle *h)
{
  struct sfm_fhuser_ext *ext = h->ext;

  if (ext->opened == 1)
    fh_close_handle (h);
  else
    ext->opened--;
}
\f
/* Dictionary reader. */

static void *bufread (struct file_handle * handle, void *buf, size_t nbytes,
                      size_t minalloc);

static int read_header (struct file_handle * h, struct sfm_read_info * inf);
static int parse_format_spec (struct file_handle * h, int32 s,
                              struct fmt_spec * v, struct variable *vv);
static int read_value_labels (struct file_handle * h, struct variable ** var_by_index);
static int read_variables (struct file_handle * h, struct variable *** var_by_index);
static int read_machine_int32_info (struct file_handle * h, int size, int count);
static int read_machine_flt64_info (struct file_handle * h, int size, int count);
static int read_documents (struct file_handle * h);

/* Displays the message X with corrupt_msg, then jumps to the lossage
   label. */
#define lose(X)                                 \
        do                                      \
          {                                     \
            corrupt_msg X;                      \
            goto lossage;                       \
          }                                     \
        while (0)

/* Calls bufread with the specified arguments, and jumps to lossage if
   the read fails. */
#define assertive_bufread(a,b,c,d)              \
        do                                      \
          {                                     \
            if (!bufread (a,b,c,d))             \
              goto lossage;                     \
          }                                     \
        while (0)

/* Reads the dictionary from file with handle H, and returns it in a
   dictionary structure.  This dictionary may be modified in order to
   rename, reorder, and delete variables, etc.  */
struct dictionary *
sfm_read_dictionary (struct file_handle * h, struct sfm_read_info * inf)
{
  /* The file handle extension record. */
  struct sfm_fhuser_ext *ext;

  /* Allows for quick reference to variables according to indexes
     relative to position within a case. */
  struct variable **var_by_index = NULL;

  /* Check whether the file is already open. */
  if (h->class == &sfm_r_class)
    {
      ext = h->ext;
      ext->opened++;
      return ext->dict;
    }
  else if (h->class != NULL)
    {
      msg (ME, _("Cannot read file %s as system file: already opened for %s."),
           fh_handle_name (h), h->class->name);
      return NULL;
    }

  msg (VM (1), _("%s: Opening system-file handle %s for reading."),
       fh_handle_filename (h), fh_handle_name (h));
  
  /* Open the physical disk file. */
  ext = xmalloc (sizeof (struct sfm_fhuser_ext));
  ext->file = fn_open (h->norm_fn, "rb");
  if (ext->file == NULL)
    {
      msg (ME, _("An error occurred while opening \"%s\" for reading "
           "as a system file: %s."), h->fn, strerror (errno));
      err_cond_fail ();
      free (ext);
      return NULL;
    }

  /* Initialize the sfm_fhuser_ext structure. */
  h->class = &sfm_r_class;
  h->ext = ext;
  ext->dict = NULL;
  ext->buf = ext->ptr = ext->end = NULL;
  ext->y = ext->x + sizeof ext->x;
  ext->opened = 1;

  /* Default special constants. */
  ext->sysmis = -FLT64_MAX;
  ext->highest = FLT64_MAX;
  ext->lowest = second_lowest_flt64;

  /* Read the header. */
  if (!read_header (h, inf))
    goto lossage;

  /* Read about the variables. */
  if (!read_variables (h, &var_by_index))
    goto lossage;

  /* Handle weighting. */
  if (ext->weight_index != -1)
    {
      struct variable *wv = var_by_index[ext->weight_index];

      if (wv == NULL)
        lose ((ME, _("%s: Weighting variable may not be a continuation of "
               "a long string variable."), h->fn));
      else if (wv->type == ALPHA)
        lose ((ME, _("%s: Weighting variable may not be a string variable."),
               h->fn));

      strcpy (ext->dict->weight_var, wv->name);
    }
  else
    ext->dict->weight_var[0] = 0;

  /* Read records of types 3, 4, 6, and 7. */
  for (;;)
    {
      int32 rec_type;

      assertive_bufread (h, &rec_type, sizeof rec_type, 0);
      if (ext->reverse_endian)
        bswap_int32 (&rec_type);

      switch (rec_type)
        {
        case 3:
          if (!read_value_labels (h, var_by_index))
            goto lossage;
          break;

        case 4:
          lose ((ME, _("%s: Orphaned variable index record (type 4).  Type 4 "
                 "records must always immediately follow type 3 records."),
                 h->fn));

        case 6:
          if (!read_documents (h))
            goto lossage;
          break;

        case 7:
          {
            struct
              {
                int32 subtype P;
                int32 size P;
                int32 count P;
              }
            data;

            int skip = 0;

            assertive_bufread (h, &data, sizeof data, 0);
            if (ext->reverse_endian)
              {
                bswap_int32 (&data.subtype);
                bswap_int32 (&data.size);
                bswap_int32 (&data.count);
              }

            /*if(data.size != sizeof(int32) && data.size != sizeof(flt64))
               lose((ME, "%s: Element size in record type 7, subtype %d, is "
               "not either the size of IN (%d) or OBS (%d); actual value "
               "is %d.",
               h->fn, data.subtype, sizeof(int32), sizeof(flt64),
               data.size)); */

            switch (data.subtype)
              {
              case 3:
                if (!read_machine_int32_info (h, data.size, data.count))
                  goto lossage;
                break;

              case 4:
                if (!read_machine_flt64_info (h, data.size, data.count))
                  goto lossage;
                break;

              case 5:
              case 6:
              case 11: /* ?? Used by SPSS 8.0. */
                skip = 1;
                break;

              default:
                msg (MW, _("%s: Unrecognized record type 7, subtype %d "
                     "encountered in system file."), h->fn, data.subtype);
                skip = 1;
              }

            if (skip)
              {
                void *x = bufread (h, NULL, data.size * data.count, 0);
                if (x == NULL)
                  goto lossage;
                free (x);
              }
          }
          break;

        case 999:
          {
            int32 filler;

            assertive_bufread (h, &filler, sizeof filler, 0);
            goto break_out_of_loop;
          }

        default:
          lose ((ME, _("%s: Unrecognized record type %d."), h->fn, rec_type));
        }
    }

break_out_of_loop:
  /* Come here on successful completion. */
  msg (VM (2), _("Read system-file dictionary successfully."));
    
#if DEBUGGING
  dump_dictionary (ext->dict);
#endif
  free (var_by_index);
  return ext->dict;

lossage:
  /* Come here on unsuccessful completion. */
  msg (VM (1), _("Error reading system-file header."));
  
  free (var_by_index);
  fn_close (h->fn, ext->file);
  if (ext && ext->dict)
    free_dictionary (ext->dict);
  free (ext);
  h->class = NULL;
  h->ext = NULL;
  return NULL;
}

/* Read record type 7, subtype 3. */
static int
read_machine_int32_info (struct file_handle * h, int size, int count)
{
  struct sfm_fhuser_ext *ext = h->ext;

  int32 data[8];
  int file_bigendian;

  int i;

  if (size != sizeof (int32) || count != 8)
    lose ((ME, _("%s: Bad size (%d) or count (%d) field on record type 7, "
           "subtype 3.  Expected size %d, count 8."),
           h->fn, size, count, sizeof (int32)));

  assertive_bufread (h, data, sizeof data, 0);
  if (ext->reverse_endian)
    for (i = 0; i < 8; i++)
      bswap_int32 (&data[i]);

  /* PORTME: Check floating-point representation. */
#ifdef FPREP_IEEE754
  if (data[4] != 1)
    lose ((ME, _("%s: Floating-point representation in system file is not "
                 "IEEE-754.  PSPP cannot convert between floating-point "
                 "formats."), h->fn));
#endif

  /* PORTME: Check recorded file endianness against intuited file
     endianness. */
#ifdef WORDS_BIGENDIAN
  file_bigendian = 1;
#else
  file_bigendian = 0;
#endif
  if (ext->reverse_endian)
    file_bigendian ^= 1;
  if (file_bigendian ^ (data[6] == 1))
    lose ((ME, _("%s: File-indicated endianness (%s) does not match endianness "
           "intuited from file header (%s)."),
           h->fn, file_bigendian ? _("big-endian") : _("little-endian"),
           data[6] == 1 ? _("big-endian") : (data[6] == 2 ? _("little-endian")
                                          : _("unknown"))));

  /* PORTME: Character representation code. */
  if (data[7] != 2 && data[7] != 3)
    lose ((ME, _("%s: File-indicated character representation code (%s) is not "
           "ASCII."), h->fn,
       data[7] == 1 ? "EBCDIC" : (data[7] == 4 ? _("DEC Kanji") : _("Unknown"))));

  return 1;

lossage:
  return 0;
}

/* Read record type 7, subtype 4. */
static int
read_machine_flt64_info (struct file_handle * h, int size, int count)
{
  struct sfm_fhuser_ext *ext = h->ext;

  flt64 data[3];

  int i;

  if (size != sizeof (flt64) || count != 3)
    lose ((ME, _("%s: Bad size (%d) or count (%d) field on record type 7, "
           "subtype 4.  Expected size %d, count 8."),
           h->fn, size, count, sizeof (flt64)));

  assertive_bufread (h, data, sizeof data, 0);
  if (ext->reverse_endian)
    for (i = 0; i < 3; i++)
      bswap_flt64 (&data[i]);

  if (data[0] != SYSMIS || data[1] != FLT64_MAX
      || data[2] != second_lowest_flt64)
    {
      ext->sysmis = data[0];
      ext->highest = data[1];
      ext->lowest = data[2];
      msg (MW, _("%s: File-indicated value is different from internal value "
                 "for at least one of the three system values.  SYSMIS: "
                 "indicated %g, expected %g; HIGHEST: %g, %g; LOWEST: "
                 "%g, %g."),
           h->fn, (double) data[0], (double) SYSMIS,
           (double) data[1], (double) FLT64_MAX,
           (double) data[2], (double) second_lowest_flt64);
    }
  
  return 1;

lossage:
  return 0;
}

static int
read_header (struct file_handle * h, struct sfm_read_info * inf)
{
  struct sfm_fhuser_ext *ext = h->ext;  /* File extension strcut. */
  struct sysfile_header hdr;            /* Disk buffer. */
  struct dictionary *dict;              /* File dictionary. */
  char prod_name[sizeof hdr.prod_name + 1];     /* Buffer for product name. */
  int skip_amt;                 /* Amount of product name to omit. */
  int i;

  /* Create the dictionary. */
  dict = ext->dict = xmalloc (sizeof *dict);
  dict->var = NULL;
  dict->name_tab = NULL;
  dict->nvar = 0;
  dict->N = 0;
  dict->nval = -1;              /* Unknown. */
  dict->n_splits = 0;
  dict->splits = NULL;
  dict->weight_var[0] = 0;
  dict->weight_index = -1;
  dict->filter_var[0] = 0;
  dict->label = NULL;
  dict->n_documents = 0;
  dict->documents = NULL;

  /* Read header, check magic. */
  assertive_bufread (h, &hdr, sizeof hdr, 0);
  if (0 != strncmp ("$FL2", hdr.rec_type, 4))
    lose ((ME, _("%s: Bad magic.  Proper system files begin with "
                 "the four characters `$FL2'. This file will not be read."),
           h->fn));

  /* Check eye-catcher string. */
  memcpy (prod_name, hdr.prod_name, sizeof hdr.prod_name);
  for (i = 0; i < 60; i++)
    if (!isprint ((unsigned char) prod_name[i]))
      prod_name[i] = ' ';
  for (i = 59; i >= 0; i--)
    if (!isgraph ((unsigned char) prod_name[i]))
      {
        prod_name[i] = '\0';
        break;
      }
  prod_name[60] = '\0';
  
  {
#define N_PREFIXES 2
    static const char *prefix[N_PREFIXES] =
      {
        "@(#) SPSS DATA FILE",
        "SPSS SYSTEM FILE.",
      };

    int i;

    for (i = 0; i < N_PREFIXES; i++)
      if (!strncmp (prefix[i], hdr.prod_name, strlen (prefix[i])))
        {
          skip_amt = strlen (prefix[i]);
          break;
        }
  }
  
  /* Check endianness. */
  /* PORTME: endianness. */
  if (hdr.layout_code == 2)
    ext->reverse_endian = 0;
  else
    {
      bswap_int32 (&hdr.layout_code);
      if (hdr.layout_code != 2)
        lose ((ME, _("%s: File layout code has unexpected value %d.  Value "
               "should be 2, in big-endian or little-endian format."),
               h->fn, hdr.layout_code));

      ext->reverse_endian = 1;
      bswap_int32 (&hdr.case_size);
      bswap_int32 (&hdr.compressed);
      bswap_int32 (&hdr.weight_index);
      bswap_int32 (&hdr.ncases);
      bswap_flt64 (&hdr.bias);
    }

  /* Copy basic info and verify correctness. */
  ext->case_size = hdr.case_size;
  if (hdr.case_size <= 0 || ext->case_size > (INT_MAX
                                              / (int) sizeof (union value) / 2))
    lose ((ME, _("%s: Number of elements per case (%d) is not between 1 "
           "and %d."), h->fn, hdr.case_size, INT_MAX / sizeof (union value) / 2));

  ext->compressed = hdr.compressed;

  ext->weight_index = hdr.weight_index - 1;
  if (hdr.weight_index < 0 || hdr.weight_index > hdr.case_size)
    lose ((ME, _("%s: Index of weighting variable (%d) is not between 0 "
           "and number of elements per case (%d)."),
           h->fn, hdr.weight_index, ext->case_size));

  ext->ncases = hdr.ncases;
  if (ext->ncases < -1 || ext->ncases > INT_MAX / 2)
    lose ((ME, _("%s: Number of cases in file (%ld) is not between -1 and "
           "%d."), h->fn, (long) ext->ncases, INT_MAX / 2));

  ext->bias = hdr.bias;
  if (ext->bias != 100.0)
    corrupt_msg (MW, _("%s: Compression bias (%g) is not the usual "
                 "value of 100."), h->fn, ext->bias);

  /* Make a file label only on the condition that the given label is
     not all spaces or nulls. */
  {
    int i;

    dict->label = NULL;
    for (i = sizeof hdr.file_label - 1; i >= 0; i--)
      if (!isspace ((unsigned char) hdr.file_label[i])
          && hdr.file_label[i] != 0)
        {
          dict->label = xmalloc (i + 2);
          memcpy (dict->label, hdr.file_label, i + 1);
          dict->label[i + 1] = 0;
          break;
        }
  }

  if (inf)
    {
      char *cp;

      memcpy (inf->creation_date, hdr.creation_date, 9);
      inf->creation_date[9] = 0;

      memcpy (inf->creation_time, hdr.creation_time, 8);
      inf->creation_time[8] = 0;

#ifdef WORDS_BIGENDIAN
      inf->bigendian = !ext->reverse_endian;
#else
      inf->bigendian = ext->reverse_endian;
#endif

      inf->compressed = hdr.compressed;

      inf->ncases = hdr.ncases;

      for (cp = &prod_name[skip_amt]; cp < &prod_name[60]; cp++)
        if (isgraph ((unsigned char) *cp))
          break;
      strcpy (inf->product, cp);
    }

  return 1;

lossage:
  return 0;
}

/* Reads most of the dictionary from file H; also fills in the
   associated VAR_BY_INDEX array. 

   Note: the dictionary returned by this function has an invalid NVAL
   element, also the VAR[] array does not have the FV and LV elements
   set, however the NV elements *are* set.  This is because the caller
   will probably modify the dictionary before reading it in from the
   file.  Also, the get.* elements are set to appropriate values to
   allow the file to be read.  */
static int
read_variables (struct file_handle * h, struct variable *** var_by_index)
{
  int i;

  struct sfm_fhuser_ext *ext = h->ext;  /* File extension record. */
  struct dictionary *dict = ext->dict;  /* Dictionary being constructed. */
  struct sysfile_variable sv;           /* Disk buffer. */
  int long_string_count = 0;    /* # of long string continuation
                                   records still expected. */
  int next_value = 0;           /* Index to next `value' structure. */

  /* Allocate variables. */
  dict->var = xmalloc (sizeof *dict->var * ext->case_size);
  *var_by_index = xmalloc (sizeof **var_by_index * ext->case_size);

  /* Read in the entry for each variable and use the info to
     initialize the dictionary. */
  for (i = 0; i < ext->case_size; i++)
    {
      struct variable *vv;
      int j;

      assertive_bufread (h, &sv, sizeof sv, 0);

      if (ext->reverse_endian)
        {
          bswap_int32 (&sv.rec_type);
          bswap_int32 (&sv.type);
          bswap_int32 (&sv.has_var_label);
          bswap_int32 (&sv.n_missing_values);
          bswap_int32 (&sv.print);
          bswap_int32 (&sv.write);
        }

      if (sv.rec_type != 2)
        lose ((ME, _("%s: position %d: Bad record type (%d); "
               "the expected value was 2."), h->fn, i, sv.rec_type));

      /* If there was a long string previously, make sure that the
         continuations are present; otherwise make sure there aren't
         any. */
      if (long_string_count)
        {
          if (sv.type != -1)
            lose ((ME, _("%s: position %d: String variable does not have "
                         "proper number of continuation records."), h->fn, i));

          (*var_by_index)[i] = NULL;
          long_string_count--;
          continue;
        }
      else if (sv.type == -1)
        lose ((ME, _("%s: position %d: Superfluous long string continuation "
               "record."), h->fn, i));

      /* Check fields for validity. */
      if (sv.type < 0 || sv.type > 255)
        lose ((ME, _("%s: position %d: Bad variable type code %d."),
               h->fn, i, sv.type));
      if (sv.has_var_label != 0 && sv.has_var_label != 1)
        lose ((ME, _("%s: position %d: Variable label indicator field is not "
               "0 or 1."), h->fn, i));
      if (sv.n_missing_values < -3 || sv.n_missing_values > 3
          || sv.n_missing_values == -1)
        lose ((ME, _("%s: position %d: Missing value indicator field is not "
                     "-3, -2, 0, 1, 2, or 3."), h->fn, i));

      /* Construct internal variable structure, initialize critical bits. */
      vv = (*var_by_index)[i] = dict->var[dict->nvar++] = xmalloc (sizeof *vv);
      vv->index = dict->nvar - 1;
      vv->foo = -1;
      vv->label = NULL;

      /* Copy first character of variable name. */
      if (!isalpha ((unsigned char) sv.name[0])
          && sv.name[0] != '@' && sv.name[0] != '#')
        lose ((ME, _("%s: position %d: Variable name begins with invalid "
               "character."), h->fn, i));
      if (islower ((unsigned char) sv.name[0]))
        msg (MW, _("%s: position %d: Variable name begins with lowercase letter "
             "%c."), h->fn, i, sv.name[0]);
      if (sv.name[0] == '#')
        msg (MW, _("%s: position %d: Variable name begins with octothorpe "
                   "(`#').  Scratch variables should not appear in system "
                   "files."), h->fn, i);
      vv->name[0] = toupper ((unsigned char) (sv.name[0]));

      /* Copy remaining characters of variable name. */
      for (j = 1; j < 8; j++)
        {
          int c = (unsigned char) sv.name[j];

          if (isspace (c))
            break;
          else if (islower (c))
            {
              msg (MW, _("%s: position %d: Variable name character %d is "
                   "lowercase letter %c."), h->fn, i, j + 1, sv.name[j]);
              vv->name[j] = toupper ((unsigned char) (c));
            }
          else if (isalnum (c) || c == '.' || c == '@'
                   || c == '#' || c == '$' || c == '_')
            vv->name[j] = c;
          else
            lose ((ME, _("%s: position %d: character `\\%03o' (%c) is not valid in a "
                   "variable name."), h->fn, i, c, c));
        }
      vv->name[j] = 0;

      /* Set type, width, and `left' fields and allocate `value'
         indices. */
      if (sv.type == 0)
        {
          vv->type = NUMERIC;
          vv->width = 0;
          vv->get.nv = 1;
          vv->get.fv = next_value++;
          vv->nv = 1;
        }
      else
        {
          vv->type = ALPHA;
          vv->width = sv.type;
          vv->nv = DIV_RND_UP (vv->width, MAX_SHORT_STRING);
          vv->get.nv = DIV_RND_UP (vv->width, sizeof (flt64));
          vv->get.fv = next_value;
          next_value += vv->get.nv;
          long_string_count = vv->get.nv - 1;
        }
      vv->left = (vv->name[0] == '#');
      vv->val_labs = val_labs_create (vv->width);

      /* Get variable label, if any. */
      if (sv.has_var_label == 1)
        {
          /* Disk buffer. */
          int32 len;

          /* Read length of label. */
          assertive_bufread (h, &len, sizeof len, 0);
          if (ext->reverse_endian)
            bswap_int32 (&len);

          /* Check len. */
          if (len < 0 || len > 255)
            lose ((ME, _("%s: Variable %s indicates variable label of invalid "
                   "length %d."), h->fn, vv->name, len));

          /* Read label into variable structure. */
          vv->label = bufread (h, NULL, ROUND_UP (len, sizeof (int32)), len + 1);
          if (vv->label == NULL)
            goto lossage;
          vv->label[len] = '\0';
        }

      /* Set missing values. */
      if (sv.n_missing_values != 0)
        {
          flt64 mv[3];

          if (vv->width > MAX_SHORT_STRING)
            lose ((ME, _("%s: Long string variable %s may not have missing "
                   "values."), h->fn, vv->name));

          assertive_bufread (h, mv, sizeof *mv * abs (sv.n_missing_values), 0);

          if (ext->reverse_endian && vv->type == NUMERIC)
            for (j = 0; j < abs (sv.n_missing_values); j++)
              bswap_flt64 (&mv[j]);

          if (sv.n_missing_values > 0)
            {
              vv->miss_type = sv.n_missing_values;
              if (vv->type == NUMERIC)
                for (j = 0; j < sv.n_missing_values; j++)
                  vv->missing[j].f = mv[j];
              else
                for (j = 0; j < sv.n_missing_values; j++)
                  memcpy (vv->missing[j].s, &mv[j], vv->width);
            }
          else
            {
              int x = 0;

              if (vv->type == ALPHA)
                lose ((ME, _("%s: String variable %s may not have missing "
                       "values specified as a range."), h->fn, vv->name));

              if (mv[0] == ext->lowest)
                {
                  vv->miss_type = MISSING_LOW;
                  vv->missing[x++].f = mv[1];
                }
              else if (mv[1] == ext->highest)
                {
                  vv->miss_type = MISSING_HIGH;
                  vv->missing[x++].f = mv[0];
                }
              else
                {
                  vv->miss_type = MISSING_RANGE;
                  vv->missing[x++].f = mv[0];
                  vv->missing[x++].f = mv[1];
                }

              if (sv.n_missing_values == -3)
                {
                  vv->miss_type += 3;
                  vv->missing[x++].f = mv[2];
                }
            }
        }
      else
        vv->miss_type = MISSING_NONE;

      if (!parse_format_spec (h, sv.print, &vv->print, vv)
          || !parse_format_spec (h, sv.write, &vv->write, vv))
        goto lossage;
    }

  /* Some consistency checks. */
  if (long_string_count != 0)
    lose ((ME, _("%s: Long string continuation records omitted at end of "
           "dictionary."), h->fn));
  if (next_value != ext->case_size)
    lose ((ME, _("%s: System file header indicates %d variable positions but "
           "%d were read from file."), h->fn, ext->case_size, next_value));
  dict->var = xrealloc (dict->var, sizeof *dict->var * dict->nvar);

  /* Construct hash table of dictionary in order to speed up
     later processing and to check for duplicate varnames. */
  dict->name_tab = hsh_create (8, compare_variables, hash_variable,
                               NULL, NULL);
  for (i = 0; i < dict->nvar; i++)
    if (NULL != hsh_insert (dict->name_tab, dict->var[i]))
      lose ((ME, _("%s: Duplicate variable name `%s' within system file."),
             h->fn, dict->var[i]->name));

  return 1;

lossage:
  for (i = 0; i < dict->nvar; i++)
    {
      free (dict->var[i]->label);
      free (dict->var[i]);
    }
  free (dict->var);
  if (dict->name_tab)
    hsh_destroy (dict->name_tab);
  free (dict);
  ext->dict = NULL;

  return 0;
}

/* Translates the format spec from sysfile format to internal
   format. */
static int
parse_format_spec (struct file_handle *h, int32 s, struct fmt_spec *v, struct variable *vv)
{
  if ((size_t) ((s >> 16) & 0xff)
      >= sizeof translate_fmt / sizeof *translate_fmt)
    lose ((ME, _("%s: Bad format specifier byte (%d)."),
           h->fn, (s >> 16) & 0xff));
  
  v->type = translate_fmt[(s >> 16) & 0xff];
  v->w = (s >> 8) & 0xff;
  v->d = s & 0xff;

  /* FIXME?  Should verify the resulting specifier more thoroughly. */

  if (v->type == -1)
    lose ((ME, _("%s: Bad format specifier byte (%d)."),
           h->fn, (s >> 16) & 0xff));
  if ((vv->type == ALPHA) ^ ((formats[v->type].cat & FCAT_STRING) != 0))
    lose ((ME, _("%s: %s variable %s has %s format specifier %s."),
           h->fn, vv->type == ALPHA ? _("String") : _("Numeric"),
           vv->name,
           formats[v->type].cat & FCAT_STRING ? _("string") : _("numeric"),
           formats[v->type].name));
  return 1;

lossage:
  return 0;
}

/* Reads value labels from sysfile H and inserts them into the
   associated dictionary. */
int
read_value_labels (struct file_handle * h, struct variable ** var_by_index)
{
  struct sfm_fhuser_ext *ext = h->ext;  /* File extension record. */

  struct label 
    {
      unsigned char raw_value[8]; /* Value as uninterpreted bytes. */
      union value value;        /* Value. */
      char *label;              /* Null-terminated label string. */
    };

  struct label *labels = NULL;
  int32 n_labels;               /* Number of labels. */

  struct variable **var = NULL; /* Associated variables. */
  int32 n_vars;                 /* Number of associated variables. */

  int i;

  /* First step: read the contents of the type 3 record and record its
     contents.  Note that we can't do much with the data since we
     don't know yet whether it is of numeric or string type. */

  /* Read number of labels. */
  assertive_bufread (h, &n_labels, sizeof n_labels, 0);
  if (ext->reverse_endian)
    bswap_int32 (&n_labels);

  /* Allocate memory. */
  labels = xmalloc (n_labels * sizeof *labels);
  for (i = 0; i < n_labels; i++)
    labels[i].label = NULL;

  /* Read each value/label tuple into labels[]. */
  for (i = 0; i < n_labels; i++)
    {
      struct label *label = labels + i;
      unsigned char label_len;
      size_t padded_len;

      /* Read value. */
      assertive_bufread (h, label->raw_value, sizeof label->raw_value, 0);

      /* Read label length. */
      assertive_bufread (h, &label_len, sizeof label_len, 0);
      padded_len = ROUND_UP (label_len + 1, sizeof (flt64));

      /* Read label, padding. */
      label->label = xmalloc (padded_len + 1);
      assertive_bufread (h, label->label, padded_len - 1, 0);
      label->label[label_len] = 0;
    }

  /* Second step: Read the type 4 record that has the list of
     variables to which the value labels are to be applied. */

  /* Read record type of type 4 record. */
  {
    int32 rec_type;
    
    assertive_bufread (h, &rec_type, sizeof rec_type, 0);
    if (ext->reverse_endian)
      bswap_int32 (&rec_type);
    
    if (rec_type != 4)
      lose ((ME, _("%s: Variable index record (type 4) does not immediately "
             "follow value label record (type 3) as it should."), h->fn));
  }

  /* Read number of variables associated with value label from type 4
     record. */
  assertive_bufread (h, &n_vars, sizeof n_vars, 0);
  if (ext->reverse_endian)
    bswap_int32 (&n_vars);
  if (n_vars < 1 || n_vars > ext->dict->nvar)
    lose ((ME, _("%s: Number of variables associated with a value label (%d) "
           "is not between 1 and the number of variables (%d)."),
           h->fn, n_vars, ext->dict->nvar));

  /* Read the list of variables. */
  var = xmalloc (n_vars * sizeof *var);
  for (i = 0; i < n_vars; i++)
    {
      int32 var_index;
      struct variable *v;

      /* Read variable index, check range. */
      assertive_bufread (h, &var_index, sizeof var_index, 0);
      if (ext->reverse_endian)
        bswap_int32 (&var_index);
      if (var_index < 1 || var_index > ext->case_size)
        lose ((ME, _("%s: Variable index associated with value label (%d) is "
               "not between 1 and the number of values (%d)."),
               h->fn, var_index, ext->case_size));

      /* Make sure it's a real variable. */
      v = var_by_index[var_index - 1];
      if (v == NULL)
        lose ((ME, _("%s: Variable index associated with value label (%d) "
                     "refers to a continuation of a string variable, not to "
                     "an actual variable."), h->fn, var_index));
      if (v->type == ALPHA && v->width > MAX_SHORT_STRING)
        lose ((ME, _("%s: Value labels are not allowed on long string "
                     "variables (%s)."), h->fn, v->name));

      /* Add it to the list of variables. */
      var[i] = v;
    }

  /* Type check the variables. */
  for (i = 1; i < n_vars; i++)
    if (var[i]->type != var[0]->type)
      lose ((ME, _("%s: Variables associated with value label are not all of "
             "identical type.  Variable %s has %s type, but variable %s has "
             "%s type."), h->fn,
             var[0]->name, var[0]->type == ALPHA ? _("string") : _("numeric"),
             var[i]->name, var[i]->type == ALPHA ? _("string") : _("numeric")));

  /* Fill in labels[].value, now that we know the desired type. */
  for (i = 0; i < n_labels; i++) 
    {
      struct label *label = labels + i;
      
      if (var[0]->type == ALPHA)
        {
          const int copy_len = min (sizeof (label->raw_value),
                                    sizeof (label->label));
          memcpy (label->value.s, label->raw_value, copy_len);
        } else {
          flt64 f;
          assert (sizeof f == sizeof label->raw_value);
          memcpy (&f, label->raw_value, sizeof f);
          if (ext->reverse_endian)
            bswap_flt64 (&f);
          label->value.f = f;
        }
    }
  
  /* Assign the value_label's to each variable. */
  for (i = 0; i < n_vars; i++)
    {
      struct variable *v = var[i];
      int j;

      /* Add each label to the variable. */
      for (j = 0; j < n_labels; j++)
        {
          struct label *label = labels + j;
          if (!val_labs_replace (v->val_labs, label->value, label->label))
            continue;

          if (var[0]->type == NUMERIC)
            msg (MW, _("%s: File contains duplicate label for value %g for "
                 "variable %s."), h->fn, label->value.f, v->name);
          else
            msg (MW, _("%s: File contains duplicate label for value `%.*s' "
                 "for variable %s."),
                 h->fn, v->width, label->value.s, v->name);
        }
    }

  for (i = 0; i < n_labels; i++)
    free (labels[i].label);
  free (var);
  return 1;

lossage:
  if (labels) 
    {
      for (i = 0; i < n_labels; i++)
        free (labels[i].label);
      free (labels); 
    }
  free (var);
  return 0;
}

/* Reads NBYTES bytes from the file represented by H.  If BUF is
   non-NULL, uses that as the buffer; otherwise allocates at least
   MINALLOC bytes.  Returns a pointer to the buffer on success, NULL
   on failure. */
static void *
bufread (struct file_handle * h, void *buf, size_t nbytes, size_t minalloc)
{
  struct sfm_fhuser_ext *ext = h->ext;

  if (buf == NULL)
    buf = xmalloc (max (nbytes, minalloc));
  if (1 != fread (buf, nbytes, 1, ext->file))
    {
      if (ferror (ext->file))
        msg (ME, _("%s: Reading system file: %s."), h->fn, strerror (errno));
      else
        corrupt_msg (ME, _("%s: Unexpected end of file."), h->fn);
      return NULL;
    }
  return buf;
}

/* Reads a document record, type 6, from system file H, and sets up
   the documents and n_documents fields in the associated
   dictionary. */
static int
read_documents (struct file_handle * h)
{
  struct sfm_fhuser_ext *ext = h->ext;
  struct dictionary *dict = ext->dict;
  int32 n_lines;

  if (dict->documents != NULL)
    lose ((ME, _("%s: System file contains multiple type 6 (document) records."),
           h->fn));

  assertive_bufread (h, &n_lines, sizeof n_lines, 0);
  dict->n_documents = n_lines;
  if (dict->n_documents <= 0)
    lose ((ME, _("%s: Number of document lines (%ld) must be greater than 0."),
           h->fn, (long) dict->n_documents));

  dict->documents = bufread (h, NULL, 80 * n_lines, 0);
  if (dict->documents == NULL)
    return 0;
  return 1;

lossage:
  return 0;
}

#if GLOBAL_DEBUGGING
#include "debug-print.h"
/* Displays dictionary DICT on stdout. */
void
dump_dictionary (struct dictionary * dict)
{
  int i;

  debug_printf ((_("dictionary:\n")));
  for (i = 0; i < dict->nvar; i++)
    {
      char print[32];
      struct variable *v = dict->var[i];
      int n, j;

      debug_printf (("   var %s", v->name));
      /*debug_printf (("(indices:%d,%d)", v->index, v->foo));*/
      debug_printf (("(type:%s,%d)", (v->type == NUMERIC ? _("num")
                                 : (v->type == ALPHA ? _("str") : "!!!")),
                     v->width));
      debug_printf (("(fv:%d,%d)", v->fv, v->nv));
      /*debug_printf (("(get.fv:%d,%d)", v->get.fv, v->get.nv));*/
      debug_printf (("(left:%s)(miss:", v->left ? _("left") : _("right")));
              
      switch (v->miss_type)
        {
        case MISSING_NONE:
          n = 0;
          debug_printf ((_("none")));
          break;
        case MISSING_1:
          n = 1;
          debug_printf ((_("one")));
          break;
        case MISSING_2:
          n = 2;
          debug_printf ((_("two")));
          break;
        case MISSING_3:
          n = 3;
          debug_printf ((_("three")));
          break;
        case MISSING_RANGE:
          n = 2;
          debug_printf ((_("range")));
          break;
        case MISSING_LOW:
          n = 1;
          debug_printf ((_("low")));
          break;
        case MISSING_HIGH:
          n = 1;
          debug_printf ((_("high")));
          break;
        case MISSING_RANGE_1:
          n = 3;
          debug_printf ((_("range+1")));
          break;
        case MISSING_LOW_1:
          n = 2;
          debug_printf ((_("low+1")));
          break;
        case MISSING_HIGH_1:
          n = 2;
          debug_printf ((_("high+1")));
          break;
        default:
          assert (0);
        }
      for (j = 0; j < n; j++)
        if (v->type == NUMERIC)
          debug_printf ((",%g", v->missing[j].f));
        else
          debug_printf ((",\"%.*s\"", v->width, v->missing[j].s));
      strcpy (print, fmt_to_string (&v->print));
      debug_printf ((")(fmt:%s,%s)(lbl:%s)\n",
                     print, fmt_to_string (&v->write),
                     v->label ? v->label : "nolabel"));
    }
}
#endif
\f
/* Data reader. */

/* Reads compressed data into H->BUF and sets other pointers
   appropriately.  Returns nonzero only if both no errors occur and
   data was read. */
static int
buffer_input (struct file_handle * h)
{
  struct sfm_fhuser_ext *ext = h->ext;
  size_t amt;

  if (ext->buf == NULL)
    ext->buf = xmalloc (sizeof *ext->buf * 128);
  amt = fread (ext->buf, sizeof *ext->buf, 128, ext->file);
  if (ferror (ext->file))
    {
      msg (ME, _("%s: Error reading file: %s."), h->fn, strerror (errno));
      return 0;
    }
  ext->ptr = ext->buf;
  ext->end = &ext->buf[amt];
  return amt;
}

/* Reads a single case consisting of compressed data from system file
   H into the array TEMP[] according to dictionary DICT, and returns
   nonzero only if successful. */
/* Data in system files is compressed in the following manner:
   data values are grouped into sets of eight; each of the eight has
   one instruction byte, which are output together in an octet; each
   byte gives a value for that byte or indicates that the value can be
   found following the instructions. */
static int
read_compressed_data (struct file_handle * h, flt64 * temp)
{
  struct sfm_fhuser_ext *ext = h->ext;

  const unsigned char *p_end = ext->x + sizeof (flt64);
  unsigned char *p = ext->y;

  const flt64 *temp_beg = temp;
  const flt64 *temp_end = &temp[ext->case_size];

  for (;;)
    {
      for (; p < p_end; p++)
        switch (*p)
          {
          case 0:
            /* Code 0 is ignored. */
            continue;
          case 252:
            /* Code 252 is end of file. */
            if (temp_beg != temp)
              lose ((ME, _("%s: Compressed data is corrupted.  Data ends "
                     "partway through a case."), h->fn));
            goto lossage;
          case 253:
            /* Code 253 indicates that the value is stored explicitly
               following the instruction bytes. */
            if (ext->ptr == NULL || ext->ptr >= ext->end)
              if (!buffer_input (h))
                {
                  lose ((ME, _("%s: Unexpected end of file."), h->fn));
                  goto lossage;
                }
            memcpy (temp++, ext->ptr++, sizeof *temp);
            if (temp >= temp_end)
              goto winnage;
            break;
          case 254:
            /* Code 254 indicates a string that is all blanks. */
            memset (temp++, ' ', sizeof *temp);
            if (temp >= temp_end)
              goto winnage;
            break;
          case 255:
            /* Code 255 indicates the system-missing value. */
            *temp = ext->sysmis;
            if (ext->reverse_endian)
              bswap_flt64 (temp);
            temp++;
            if (temp >= temp_end)
              goto winnage;
            break;
          default:
            /* Codes 1 through 251 inclusive are taken to indicate a
               value of (BYTE - BIAS), where BYTE is the byte's value
               and BIAS is the compression bias (generally 100.0). */
            *temp = *p - ext->bias;
            if (ext->reverse_endian)
              bswap_flt64 (temp);
            temp++;
            if (temp >= temp_end)
              goto winnage;
            break;
          }

      /* We have reached the end of this instruction octet.  Read
         another. */
      if (ext->ptr == NULL || ext->ptr >= ext->end)
        if (!buffer_input (h))
          {
            if (temp_beg != temp)
              lose ((ME, _("%s: Unexpected end of file."), h->fn));
            goto lossage;
          }
      memcpy (ext->x, ext->ptr++, sizeof *temp);
      p = ext->x;
    }

  /* Not reached. */
  assert (0);

winnage:
  /* We have filled up an entire record.  Update state and return
     successfully. */
  ext->y = ++p;
  return 1;

lossage:
  /* We have been unsuccessful at filling a record, either through i/o
     error or through an end-of-file indication.  Update state and
     return unsuccessfully. */
  return 0;
}

/* Reads one case from system file H into the value array PERM
   according to the instructions given in associated dictionary DICT,
   which must have the get.* elements appropriately set.  Returns
   nonzero only if successful.  */
int
sfm_read_case (struct file_handle * h, union value * perm, struct dictionary * dict)
{
  struct sfm_fhuser_ext *ext = h->ext;

  size_t nbytes;
  flt64 *temp;

  int i;

  /* Make sure the caller remembered to finish polishing the
     dictionary returned by sfm_read_dictionary(). */
  assert (dict->nval > 0);

  /* The first concern is to obtain a full case relative to the data
     file.  (Cases in the data file have no particular relationship to
     cases in the active file.) */
  nbytes = sizeof *temp * ext->case_size;
  temp = local_alloc (nbytes);

  if (ext->compressed == 0)
    {
      size_t amt = fread (temp, 1, nbytes, ext->file);

      if (amt != nbytes)
        {
          if (ferror (ext->file))
            msg (ME, _("%s: Reading system file: %s."), h->fn, strerror (errno));
          else if (amt != 0)
            msg (ME, _("%s: Partial record at end of system file."), h->fn);
          goto lossage;
        }
    }
  else if (!read_compressed_data (h, temp))
    goto lossage;

  /* Translate a case in data file format to a case in active file
     format. */
  for (i = 0; i < dict->nvar; i++)
    {
      struct variable *v = dict->var[i];

      if (v->get.fv == -1)
        continue;
      
      if (v->type == NUMERIC)
        {
          flt64 src = temp[v->get.fv];
          if (ext->reverse_endian)
            bswap_flt64 (&src);
          perm[v->fv].f = src == ext->sysmis ? SYSMIS : src;
        }
      else
        memcpy (&perm[v->fv].s, &temp[v->get.fv], v->width);
    }

  local_free (temp);
  return 1;

lossage:
  local_free (temp);
  return 0;
}

static struct fh_ext_class sfm_r_class =
{
  3,
  N_("reading as a system file"),
  sfm_close,
};