1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2007, 2009, 2010, 2011 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include "data/sys-file-writer.h"
20 #include "data/sys-file-private.h"
29 #include "data/attributes.h"
30 #include "data/case.h"
31 #include "data/casewriter-provider.h"
32 #include "data/casewriter.h"
33 #include "data/dictionary.h"
34 #include "data/file-handle-def.h"
35 #include "data/file-name.h"
36 #include "data/format.h"
37 #include "data/make-file.h"
38 #include "data/missing-values.h"
39 #include "data/mrset.h"
40 #include "data/settings.h"
41 #include "data/short-names.h"
42 #include "data/value-labels.h"
43 #include "data/variable.h"
44 #include "libpspp/float-format.h"
45 #include "libpspp/i18n.h"
46 #include "libpspp/integer-format.h"
47 #include "libpspp/message.h"
48 #include "libpspp/misc.h"
49 #include "libpspp/str.h"
50 #include "libpspp/string-array.h"
51 #include "libpspp/version.h"
53 #include "gl/xmemdup0.h"
54 #include "gl/minmax.h"
55 #include "gl/unlocked-io.h"
56 #include "gl/xalloc.h"
59 #define _(msgid) gettext (msgid)
60 #define N_(msgid) (msgid)
62 /* Compression bias used by PSPP. Values between (1 -
63 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
65 #define COMPRESSION_BIAS 100
67 /* System file writer. */
70 struct file_handle *fh; /* File handle. */
71 struct fh_lock *lock; /* Mutual exclusion for file. */
72 FILE *file; /* File stream. */
73 struct replace_file *rf; /* Ticket for replacing output file. */
75 bool compress; /* 1=compressed, 0=not compressed. */
76 casenumber case_cnt; /* Number of cases written so far. */
78 /* Compression buffering.
80 Compressed data is output as groups of 8 1-byte opcodes
81 followed by up to 8 (depending on the opcodes) 8-byte data
82 items. Data items and opcodes arrive at the same time but
83 must be reordered for writing to disk, thus a small amount
85 uint8_t opcodes[8]; /* Buffered opcodes. */
86 int opcode_cnt; /* Number of buffered opcodes. */
87 uint8_t data[8][8]; /* Buffered data. */
88 int data_cnt; /* Number of buffered data items. */
91 struct sfm_var *sfm_vars; /* Variables. */
92 size_t sfm_var_cnt; /* Number of variables. */
93 size_t segment_cnt; /* Number of variables including extra segments
94 for long string variables. */
97 static const struct casewriter_class sys_file_casewriter_class;
99 static void write_header (struct sfm_writer *, const struct dictionary *);
100 static void write_variable (struct sfm_writer *, const struct variable *);
101 static void write_value_labels (struct sfm_writer *,
102 const struct dictionary *);
103 static void write_integer_info_record (struct sfm_writer *,
104 const struct dictionary *);
105 static void write_float_info_record (struct sfm_writer *);
107 static void write_longvar_table (struct sfm_writer *w,
108 const struct dictionary *dict);
110 static void write_encoding_record (struct sfm_writer *w,
111 const struct dictionary *);
113 static void write_vls_length_table (struct sfm_writer *w,
114 const struct dictionary *dict);
116 static void write_long_string_value_labels (struct sfm_writer *,
117 const struct dictionary *);
119 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
122 static void write_variable_display_parameters (struct sfm_writer *w,
123 const struct dictionary *dict);
125 static void write_documents (struct sfm_writer *, const struct dictionary *);
127 static void write_data_file_attributes (struct sfm_writer *,
128 const struct dictionary *);
129 static void write_variable_attributes (struct sfm_writer *,
130 const struct dictionary *);
132 static void write_int (struct sfm_writer *, int32_t);
133 static inline void convert_double_to_output_format (double, uint8_t[8]);
134 static void write_float (struct sfm_writer *, double);
135 static void write_string (struct sfm_writer *, const char *, size_t);
136 static void write_utf8_string (struct sfm_writer *, const char *encoding,
137 const char *string, size_t width);
138 static void write_utf8_record (struct sfm_writer *, const char *encoding,
139 const struct string *content, int subtype);
140 static void write_string_record (struct sfm_writer *,
141 const struct substring content, int subtype);
142 static void write_bytes (struct sfm_writer *, const void *, size_t);
143 static void write_zeros (struct sfm_writer *, size_t);
144 static void write_spaces (struct sfm_writer *, size_t);
145 static void write_value (struct sfm_writer *, const union value *, int width);
147 static void write_case_uncompressed (struct sfm_writer *,
148 const struct ccase *);
149 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
150 static void flush_compressed (struct sfm_writer *);
151 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
152 static void put_cmp_number (struct sfm_writer *, double);
153 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
155 static bool write_error (const struct sfm_writer *);
156 static bool close_writer (struct sfm_writer *);
158 /* Returns default options for writing a system file. */
159 struct sfm_write_options
160 sfm_writer_default_options (void)
162 struct sfm_write_options opts;
163 opts.create_writeable = true;
164 opts.compress = settings_get_scompression ();
169 /* Opens the system file designated by file handle FH for writing
170 cases from dictionary D according to the given OPTS.
172 No reference to D is retained, so it may be modified or
173 destroyed at will after this function returns. D is not
174 modified by this function, except to assign short names. */
176 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
177 struct sfm_write_options opts)
179 struct sfm_writer *w;
184 if (opts.version != 2 && opts.version != 3)
186 msg (ME, _("Unknown system file version %d. Treating as version %d."),
191 /* Create and initialize writer. */
192 w = xmalloc (sizeof *w);
198 w->compress = opts.compress;
201 w->opcode_cnt = w->data_cnt = 0;
203 /* Figure out how to map in-memory case data to on-disk case
204 data. Also count the number of segments. Very long strings
205 occupy multiple segments, otherwise each variable only takes
207 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
210 /* Open file handle as an exclusive writer. */
211 /* TRANSLATORS: this fragment will be interpolated into
212 messages in fh_lock() that identify types of files. */
213 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
217 /* Create the file on disk. */
219 if (opts.create_writeable)
221 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
225 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
226 fh_get_file_name (fh), strerror (errno));
230 /* Write the file header. */
233 /* Write basic variable info. */
234 short_names_assign (d);
235 for (i = 0; i < dict_get_var_cnt (d); i++)
236 write_variable (w, dict_get_var (d, i));
238 write_value_labels (w, d);
240 if (dict_get_document_line_cnt (d) > 0)
241 write_documents (w, d);
243 write_integer_info_record (w, d);
244 write_float_info_record (w);
246 write_mrsets (w, d, true);
248 write_variable_display_parameters (w, d);
250 if (opts.version >= 3)
251 write_longvar_table (w, d);
253 write_vls_length_table (w, d);
255 write_long_string_value_labels (w, d);
257 if (attrset_count (dict_get_attributes (d)))
258 write_data_file_attributes (w, d);
259 write_variable_attributes (w, d);
261 write_mrsets (w, d, false);
263 write_encoding_record (w, d);
265 /* Write end-of-headers record. */
272 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
279 /* Returns value of X truncated to two least-significant digits. */
290 /* Calculates the offset of data for TARGET_VAR from the
291 beginning of each case's data for dictionary D. The return
292 value is in "octs" (8-byte units). */
294 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
300 for (i = 0; i < dict_get_var_cnt (d); i++)
302 struct variable *var = dict_get_var (d, i);
303 if (var == target_var)
305 oct_idx += sfm_width_to_octs (var_get_width (var));
310 /* Write the sysfile_header header to system file W. */
312 write_header (struct sfm_writer *w, const struct dictionary *d)
315 char creation_date[10];
316 char creation_time[9];
317 const char *file_label;
318 struct variable *weight;
322 /* Record-type code. */
323 write_string (w, "$FL2", 4);
325 /* Product identification. */
326 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
327 version, host_system);
328 write_string (w, prod_name, 60);
333 /* Number of `union value's per case. */
334 write_int (w, calc_oct_idx (d, NULL));
337 write_int (w, w->compress);
339 /* Weight variable. */
340 weight = dict_get_weight (d);
341 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
343 /* Number of cases. We don't know this in advance, so we write
344 -1 to indicate an unknown number of cases. Later we can
345 come back and overwrite it with the true value. */
348 /* Compression bias. */
349 write_float (w, COMPRESSION_BIAS);
351 /* Creation date and time. */
352 if (time (&t) == (time_t) -1)
354 strcpy (creation_date, "01 Jan 70");
355 strcpy (creation_time, "00:00:00");
359 static const char *const month_name[12] =
361 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
362 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
364 struct tm *tmp = localtime (&t);
365 int day = rerange (tmp->tm_mday);
366 int mon = rerange (tmp->tm_mon + 1);
367 int year = rerange (tmp->tm_year);
368 int hour = rerange (tmp->tm_hour + 1);
369 int min = rerange (tmp->tm_min + 1);
370 int sec = rerange (tmp->tm_sec + 1);
372 snprintf (creation_date, sizeof creation_date,
373 "%02d %s %02d", day, month_name[mon - 1], year);
374 snprintf (creation_time, sizeof creation_time,
375 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
377 write_string (w, creation_date, 9);
378 write_string (w, creation_time, 8);
381 file_label = dict_get_label (d);
382 if (file_label == NULL)
384 write_utf8_string (w, dict_get_encoding (d), file_label, 64);
390 /* Write format spec FMT to W, after adjusting it to be
391 compatible with the given WIDTH. */
393 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
395 assert (fmt_check_output (&fmt));
396 assert (sfm_width_to_segments (width) == 1);
399 fmt_resize (&fmt, width);
400 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
403 /* Write a string continuation variable record for each 8-byte
404 section beyond the initial 8 bytes, for a variable of the
407 write_variable_continuation_records (struct sfm_writer *w, int width)
411 assert (sfm_width_to_segments (width) == 1);
412 for (position = 8; position < width; position += 8)
414 write_int (w, 2); /* Record type. */
415 write_int (w, -1); /* Width. */
416 write_int (w, 0); /* No variable label. */
417 write_int (w, 0); /* No missing values. */
418 write_int (w, 0); /* Print format. */
419 write_int (w, 0); /* Write format. */
420 write_zeros (w, 8); /* Name. */
424 /* Write the variable record(s) for variable V to system file
427 write_variable (struct sfm_writer *w, const struct variable *v)
429 int width = var_get_width (v);
430 int segment_cnt = sfm_width_to_segments (width);
431 int seg0_width = sfm_segment_alloc_width (width, 0);
432 const char *encoding = var_get_encoding (v);
433 struct missing_values mv;
440 write_int (w, seg0_width);
442 /* Variable has a variable label? */
443 write_int (w, var_has_label (v));
445 /* Number of missing values. If there is a range, then the
446 range counts as 2 missing values and causes the number to be
448 mv_copy (&mv, var_get_missing_values (v));
449 if (mv_get_width (&mv) > 8)
451 if (mv_has_range (&mv))
452 write_int (w, -2 - mv_n_values (&mv));
454 write_int (w, mv_n_values (&mv));
456 /* Print and write formats. */
457 write_format (w, *var_get_print_format (v), seg0_width);
458 write_format (w, *var_get_write_format (v), seg0_width);
461 The full name is in a translation table written
463 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
466 if (var_has_label (v))
468 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
469 size_t label_len = MIN (strlen (label), 255);
470 size_t padded_len = ROUND_UP (label_len, 4);
471 write_int (w, label_len);
472 write_string (w, label, padded_len);
476 /* Write the missing values, if any, range first. */
477 if (mv_has_range (&mv))
480 mv_get_range (&mv, &x, &y);
484 for (i = 0; i < mv_n_values (&mv); i++)
485 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
487 write_variable_continuation_records (w, seg0_width);
489 /* Write additional segments for very long string variables. */
490 for (i = 1; i < segment_cnt; i++)
492 int seg_width = sfm_segment_alloc_width (width, i);
493 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
495 write_int (w, 2); /* Variable record. */
496 write_int (w, seg_width); /* Width. */
497 write_int (w, 0); /* No variable label. */
498 write_int (w, 0); /* No missing values. */
499 write_format (w, fmt, seg_width); /* Print format. */
500 write_format (w, fmt, seg_width); /* Write format. */
501 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
503 write_variable_continuation_records (w, seg_width);
509 /* Writes the value labels to system file W.
511 Value labels for long string variables are written separately,
512 by write_long_string_value_labels. */
514 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
518 struct hmap_node hmap_node;
519 const struct val_labs *val_labs;
521 size_t n_indexes, allocated_indexes;
524 size_t n_sets, allocated_sets;
525 struct label_set **sets;
526 struct hmap same_sets;
530 n_sets = allocated_sets = 0;
532 hmap_init (&same_sets);
535 for (i = 0; i < dict_get_var_cnt (d); i++)
537 struct variable *v = dict_get_var (d, i);
539 if (var_has_value_labels (v) && var_get_width (v) <= 8)
541 const struct val_labs *val_labs = var_get_value_labels (v);
542 unsigned int hash = val_labs_hash (val_labs, 0);
543 struct label_set *set;
545 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
548 if (val_labs_equal (set->val_labs, val_labs))
550 if (set->n_indexes >= set->allocated_indexes)
551 set->indexes = x2nrealloc (set->indexes,
552 &set->allocated_indexes,
553 sizeof *set->indexes);
554 set->indexes[set->n_indexes++] = idx;
559 set = xmalloc (sizeof *set);
560 set->val_labs = val_labs;
561 set->indexes = xmalloc (sizeof *set->indexes);
562 set->indexes[0] = idx;
564 set->allocated_indexes = 1;
565 hmap_insert (&same_sets, &set->hmap_node, hash);
567 if (n_sets >= allocated_sets)
568 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
569 sets[n_sets++] = set;
573 idx += sfm_width_to_octs (var_get_width (v));
576 for (i = 0; i < n_sets; i++)
578 const struct label_set *set = sets[i];
579 const struct val_labs *val_labs = set->val_labs;
580 size_t n_labels = val_labs_count (val_labs);
581 int width = val_labs_get_width (val_labs);
582 const struct val_lab **labels;
585 /* Value label record. */
586 write_int (w, 3); /* Record type. */
587 write_int (w, n_labels);
588 labels = val_labs_sorted (val_labs);
589 for (j = 0; j < n_labels; j++)
591 const struct val_lab *vl = labels[j];
592 char *label = recode_string (dict_get_encoding (d), UTF8,
593 val_lab_get_escaped_label (vl), -1);
594 uint8_t len = MIN (strlen (label), 255);
596 write_value (w, val_lab_get_value (vl), width);
597 write_bytes (w, &len, 1);
598 write_bytes (w, label, len);
599 write_zeros (w, REM_RND_UP (len + 1, 8));
604 /* Value label variable record. */
605 write_int (w, 4); /* Record type. */
606 write_int (w, set->n_indexes);
607 for (j = 0; j < set->n_indexes; j++)
608 write_int (w, set->indexes[j] + 1);
611 for (i = 0; i < n_sets; i++)
613 struct label_set *set = sets[i];
619 hmap_destroy (&same_sets);
622 /* Writes record type 6, document record. */
624 write_documents (struct sfm_writer *w, const struct dictionary *d)
626 const struct string_array *docs = dict_get_documents (d);
627 const char *enc = dict_get_encoding (d);
630 write_int (w, 6); /* Record type. */
631 write_int (w, docs->n);
632 for (i = 0; i < docs->n; i++)
634 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
635 size_t s_len = strlen (s);
636 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
638 write_bytes (w, s, write_len);
639 write_spaces (w, DOC_LINE_LENGTH - write_len);
645 put_attrset (struct string *string, const struct attrset *attrs)
647 const struct attribute *attr;
648 struct attrset_iterator i;
650 for (attr = attrset_first (attrs, &i); attr != NULL;
651 attr = attrset_next (attrs, &i))
653 size_t n_values = attribute_get_n_values (attr);
656 ds_put_cstr (string, attribute_get_name (attr));
657 ds_put_byte (string, '(');
658 for (j = 0; j < n_values; j++)
659 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
660 ds_put_byte (string, ')');
665 write_data_file_attributes (struct sfm_writer *w,
666 const struct dictionary *d)
668 struct string s = DS_EMPTY_INITIALIZER;
669 put_attrset (&s, dict_get_attributes (d));
670 write_utf8_record (w, dict_get_encoding (d), &s, 17);
675 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
677 struct string s = DS_EMPTY_INITIALIZER;
678 size_t n_vars = dict_get_var_cnt (d);
679 size_t n_attrsets = 0;
682 for (i = 0; i < n_vars; i++)
684 struct variable *v = dict_get_var (d, i);
685 struct attrset *attrs = var_get_attributes (v);
686 if (attrset_count (attrs))
689 ds_put_byte (&s, '/');
690 ds_put_format (&s, "%s:", var_get_name (v));
691 put_attrset (&s, attrs);
695 write_utf8_record (w, dict_get_encoding (d), &s, 18);
699 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
700 SPSS before release 14, otherwise writes sets supported only by later
703 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
706 const char *encoding = dict_get_encoding (dict);
707 struct string s = DS_EMPTY_INITIALIZER;
711 n_mrsets = dict_get_n_mrsets (dict);
715 for (i = 0; i < n_mrsets; i++)
717 const struct mrset *mrset = dict_get_mrset (dict, i);
721 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
725 name = recode_string (encoding, "UTF-8", mrset->name, -1);
726 ds_put_format (&s, "%s=", name);
729 if (mrset->type == MRSET_MD)
733 if (mrset->cat_source == MRSET_COUNTEDVALUES)
734 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
736 ds_put_byte (&s, 'D');
738 if (mrset->width == 0)
739 counted = xasprintf ("%.0f", mrset->counted.f);
741 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
743 ds_put_format (&s, "%zu %s", strlen (counted), counted);
747 ds_put_byte (&s, 'C');
748 ds_put_byte (&s, ' ');
750 if (mrset->label && !mrset->label_from_var_label)
752 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
753 ds_put_format (&s, "%zu %s", strlen (label), label);
757 ds_put_cstr (&s, "0 ");
759 for (j = 0; j < mrset->n_vars; j++)
761 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
762 char *short_name = recode_string (encoding, "UTF-8",
763 short_name_utf8, -1);
764 str_lowercase (short_name);
765 ds_put_format (&s, " %s", short_name);
768 ds_put_byte (&s, '\n');
771 if (!ds_is_empty (&s))
772 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
776 /* Write the alignment, width and scale values. */
778 write_variable_display_parameters (struct sfm_writer *w,
779 const struct dictionary *dict)
783 write_int (w, 7); /* Record type. */
784 write_int (w, 11); /* Record subtype. */
785 write_int (w, 4); /* Data item (int32) size. */
786 write_int (w, w->segment_cnt * 3); /* Number of data items. */
788 for (i = 0; i < dict_get_var_cnt (dict); ++i)
790 struct variable *v = dict_get_var (dict, i);
791 int width = var_get_width (v);
792 int segment_cnt = sfm_width_to_segments (width);
793 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
794 : var_get_measure (v) == MEASURE_ORDINAL ? 2
796 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
797 : var_get_alignment (v) == ALIGN_RIGHT ? 1
801 for (i = 0; i < segment_cnt; i++)
803 int width_left = width - sfm_segment_effective_offset (width, i);
804 write_int (w, measure);
805 write_int (w, (i == 0 ? var_get_display_width (v)
806 : var_default_display_width (width_left)));
807 write_int (w, alignment);
812 /* Writes the table of lengths for very long string variables. */
814 write_vls_length_table (struct sfm_writer *w,
815 const struct dictionary *dict)
820 ds_init_empty (&map);
821 for (i = 0; i < dict_get_var_cnt (dict); ++i)
823 const struct variable *v = dict_get_var (dict, i);
824 if (sfm_width_to_segments (var_get_width (v)) > 1)
825 ds_put_format (&map, "%s=%05d%c\t",
826 var_get_short_name (v, 0), var_get_width (v), 0);
828 if (!ds_is_empty (&map))
829 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
835 write_long_string_value_labels (struct sfm_writer *w,
836 const struct dictionary *dict)
838 size_t n_vars = dict_get_var_cnt (dict);
842 /* Figure out the size in advance. */
844 for (i = 0; i < n_vars; i++)
846 struct variable *var = dict_get_var (dict, i);
847 const struct val_labs *val_labs = var_get_value_labels (var);
848 const char *encoding = var_get_encoding (var);
849 int width = var_get_width (var);
850 const struct val_lab *val_lab;
852 if (val_labs_count (val_labs) == 0 || width < 9)
856 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
857 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
858 val_lab = val_labs_next (val_labs, val_lab))
861 size += recode_string_len (encoding, "UTF-8",
862 val_lab_get_escaped_label (val_lab), -1);
868 write_int (w, 7); /* Record type. */
869 write_int (w, 21); /* Record subtype */
870 write_int (w, 1); /* Data item (byte) size. */
871 write_int (w, size); /* Number of data items. */
873 start = ftello (w->file);
874 for (i = 0; i < n_vars; i++)
876 struct variable *var = dict_get_var (dict, i);
877 const struct val_labs *val_labs = var_get_value_labels (var);
878 const char *encoding = var_get_encoding (var);
879 int width = var_get_width (var);
880 const struct val_lab *val_lab;
883 if (val_labs_count (val_labs) == 0 || width < 9)
886 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
887 write_int (w, strlen (var_name));
888 write_bytes (w, var_name, strlen (var_name));
891 write_int (w, width);
892 write_int (w, val_labs_count (val_labs));
893 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
894 val_lab = val_labs_next (val_labs, val_lab))
899 write_int (w, width);
900 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
903 label = recode_string (var_get_encoding (var), "UTF-8",
904 val_lab_get_escaped_label (val_lab), -1);
905 len = strlen (label);
907 write_bytes (w, label, len);
911 assert (ftello (w->file) == start + size);
915 write_encoding_record (struct sfm_writer *w,
916 const struct dictionary *d)
918 /* IANA says "...character set names may be up to 40 characters taken
919 from the printable characters of US-ASCII," so character set names
920 don't need to be recoded to be in UTF-8.
922 We convert encoding names to uppercase because SPSS writes encoding
923 names in uppercase. */
924 char *encoding = xstrdup (dict_get_encoding (d));
925 str_uppercase (encoding);
926 write_string_record (w, ss_cstr (encoding), 20);
930 /* Writes the long variable name table. */
932 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
937 ds_init_empty (&map);
938 for (i = 0; i < dict_get_var_cnt (dict); i++)
940 struct variable *v = dict_get_var (dict, i);
942 ds_put_byte (&map, '\t');
943 ds_put_format (&map, "%s=%s",
944 var_get_short_name (v, 0), var_get_name (v));
946 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
950 /* Write integer information record. */
952 write_integer_info_record (struct sfm_writer *w,
953 const struct dictionary *d)
955 int version_component[3];
959 /* Parse the version string. */
960 memset (version_component, 0, sizeof version_component);
961 sscanf (bare_version, "%d.%d.%d",
962 &version_component[0], &version_component[1], &version_component[2]);
964 /* Figure out the floating-point format. */
965 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
966 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
968 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
970 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
975 /* Choose codepage. */
976 codepage = sys_get_codepage_from_encoding (dict_get_encoding (d));
979 /* Default to "7-bit ASCII" if the codepage number is unknown, because
980 many files use this codepage number regardless of their actual
986 write_int (w, 7); /* Record type. */
987 write_int (w, 3); /* Record subtype. */
988 write_int (w, 4); /* Data item (int32) size. */
989 write_int (w, 8); /* Number of data items. */
990 write_int (w, version_component[0]);
991 write_int (w, version_component[1]);
992 write_int (w, version_component[2]);
993 write_int (w, -1); /* Machine code. */
994 write_int (w, float_format);
995 write_int (w, 1); /* Compression code. */
996 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
997 write_int (w, codepage);
1000 /* Write floating-point information record. */
1002 write_float_info_record (struct sfm_writer *w)
1004 write_int (w, 7); /* Record type. */
1005 write_int (w, 4); /* Record subtype. */
1006 write_int (w, 8); /* Data item (flt64) size. */
1007 write_int (w, 3); /* Number of data items. */
1008 write_float (w, SYSMIS); /* System-missing value. */
1009 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1010 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1013 /* Writes case C to system file W. */
1015 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1018 struct sfm_writer *w = w_;
1020 if (ferror (w->file))
1022 casewriter_force_error (writer);
1030 write_case_uncompressed (w, c);
1032 write_case_compressed (w, c);
1037 /* Destroys system file writer W. */
1039 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1041 struct sfm_writer *w = w_;
1042 if (!close_writer (w))
1043 casewriter_force_error (writer);
1046 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1048 write_error (const struct sfm_writer *writer)
1050 return ferror (writer->file);
1053 /* Closes a system file after we're done with it.
1054 Returns true if successful, false if an I/O error occurred. */
1056 close_writer (struct sfm_writer *w)
1064 if (w->file != NULL)
1067 if (w->opcode_cnt > 0)
1068 flush_compressed (w);
1071 ok = !write_error (w);
1073 /* Seek back to the beginning and update the number of cases.
1074 This is just a courtesy to later readers, so there's no need
1075 to check return values or report errors. */
1076 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1078 write_int (w, w->case_cnt);
1082 if (fclose (w->file) == EOF)
1086 msg (ME, _("An I/O error occurred writing system file `%s'."),
1087 fh_get_file_name (w->fh));
1089 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1093 fh_unlock (w->lock);
1102 /* System file writer casewriter class. */
1103 static const struct casewriter_class sys_file_casewriter_class =
1105 sys_file_casewriter_write,
1106 sys_file_casewriter_destroy,
1110 /* Writes case C to system file W, without compressing it. */
1112 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1116 for (i = 0; i < w->sfm_var_cnt; i++)
1118 struct sfm_var *v = &w->sfm_vars[i];
1120 if (v->var_width == 0)
1121 write_float (w, case_num_idx (c, v->case_index));
1124 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1126 write_spaces (w, v->padding);
1131 /* Writes case C to system file W, with compression. */
1133 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1137 for (i = 0; i < w->sfm_var_cnt; i++)
1139 struct sfm_var *v = &w->sfm_vars[i];
1141 if (v->var_width == 0)
1143 double d = case_num_idx (c, v->case_index);
1145 put_cmp_opcode (w, 255);
1146 else if (d >= 1 - COMPRESSION_BIAS
1147 && d <= 251 - COMPRESSION_BIAS
1149 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1152 put_cmp_opcode (w, 253);
1153 put_cmp_number (w, d);
1158 int offset = v->offset;
1161 /* This code properly deals with a width that is not a
1162 multiple of 8, by ensuring that the final partial
1163 oct (8 byte unit) is treated as padded with spaces
1165 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1167 const void *data = case_str_idx (c, v->case_index) + offset;
1168 int chunk_size = MIN (width, 8);
1169 if (!memcmp (data, " ", chunk_size))
1170 put_cmp_opcode (w, 254);
1173 put_cmp_opcode (w, 253);
1174 put_cmp_string (w, data, chunk_size);
1178 /* This code deals properly with padding that is not a
1179 multiple of 8 bytes, by discarding the remainder,
1180 which was already effectively padded with spaces in
1181 the previous loop. (Note that v->width + v->padding
1182 is always a multiple of 8.) */
1183 for (padding = v->padding / 8; padding > 0; padding--)
1184 put_cmp_opcode (w, 254);
1189 /* Flushes buffered compressed opcodes and data to W.
1190 The compression buffer must not be empty. */
1192 flush_compressed (struct sfm_writer *w)
1194 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1196 write_bytes (w, w->opcodes, w->opcode_cnt);
1197 write_zeros (w, 8 - w->opcode_cnt);
1199 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1201 w->opcode_cnt = w->data_cnt = 0;
1204 /* Appends OPCODE to the buffered set of compression opcodes in
1205 W. Flushes the compression buffer beforehand if necessary. */
1207 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1209 if (w->opcode_cnt >= 8)
1210 flush_compressed (w);
1212 w->opcodes[w->opcode_cnt++] = opcode;
1215 /* Appends NUMBER to the buffered compression data in W. The
1216 buffer must not be full; the way to assure that is to call
1217 this function only just after a call to put_cmp_opcode, which
1218 will flush the buffer as necessary. */
1220 put_cmp_number (struct sfm_writer *w, double number)
1222 assert (w->opcode_cnt > 0);
1223 assert (w->data_cnt < 8);
1225 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1228 /* Appends SIZE bytes of DATA to the buffered compression data in
1229 W, followed by enough spaces to pad the output data to exactly
1230 8 bytes (thus, SIZE must be no greater than 8). The buffer
1231 must not be full; the way to assure that is to call this
1232 function only just after a call to put_cmp_opcode, which will
1233 flush the buffer as necessary. */
1235 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1237 assert (w->opcode_cnt > 0);
1238 assert (w->data_cnt < 8);
1241 memset (w->data[w->data_cnt], ' ', 8);
1242 memcpy (w->data[w->data_cnt], data, size);
1246 /* Writes 32-bit integer X to the output file for writer W. */
1248 write_int (struct sfm_writer *w, int32_t x)
1250 write_bytes (w, &x, sizeof x);
1253 /* Converts NATIVE to the 64-bit format used in output files in
1256 convert_double_to_output_format (double native, uint8_t output[8])
1258 /* If "double" is not a 64-bit type, then convert it to a
1259 64-bit type. Otherwise just copy it. */
1260 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1261 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1263 memcpy (output, &native, sizeof native);
1266 /* Writes floating-point number X to the output file for writer
1269 write_float (struct sfm_writer *w, double x)
1272 convert_double_to_output_format (x, output);
1273 write_bytes (w, output, sizeof output);
1276 /* Writes contents of VALUE with the given WIDTH to W, padding
1277 with zeros to a multiple of 8 bytes.
1278 To avoid a branch, and because we don't actually need to
1279 support it, WIDTH must be no bigger than 8. */
1281 write_value (struct sfm_writer *w, const union value *value, int width)
1283 assert (width <= 8);
1285 write_float (w, value->f);
1288 write_bytes (w, value_str (value, width), width);
1289 write_zeros (w, 8 - width);
1293 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1294 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1295 is padded on the right with spaces. */
1297 write_string (struct sfm_writer *w, const char *string, size_t width)
1299 size_t data_bytes = MIN (strlen (string), width);
1300 size_t pad_bytes = width - data_bytes;
1301 write_bytes (w, string, data_bytes);
1302 while (pad_bytes-- > 0)
1303 putc (' ', w->file);
1306 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1307 recoded version in a field of the given WIDTH to W. The string is truncated
1308 or padded on the right with spaces to exactly WIDTH bytes. */
1310 write_utf8_string (struct sfm_writer *w, const char *encoding,
1311 const char *string, size_t width)
1313 char *s = recode_string (encoding, "UTF-8", string, -1);
1314 write_string (w, s, width);
1318 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1319 from UTF-8 encoded into ENCODING. */
1321 write_utf8_record (struct sfm_writer *w, const char *encoding,
1322 const struct string *content, int subtype)
1326 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1327 write_string_record (w, s, subtype);
1331 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1334 write_string_record (struct sfm_writer *w,
1335 const struct substring content, int subtype)
1338 write_int (w, subtype);
1340 write_int (w, ss_length (content));
1341 write_bytes (w, ss_data (content), ss_length (content));
1344 /* Writes SIZE bytes of DATA to W's output file. */
1346 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1348 fwrite (data, 1, size, w->file);
1351 /* Writes N zeros to W's output file. */
1353 write_zeros (struct sfm_writer *w, size_t n)
1359 /* Writes N spaces to W's output file. */
1361 write_spaces (struct sfm_writer *w, size_t n)
1364 putc (' ', w->file);