1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-2000, 2006-2013 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. */
77 uint8_t space; /* ' ' in the file's character encoding. */
79 /* Compression buffering.
81 Compressed data is output as groups of 8 1-byte opcodes
82 followed by up to 8 (depending on the opcodes) 8-byte data
83 items. Data items and opcodes arrive at the same time but
84 must be reordered for writing to disk, thus a small amount
86 uint8_t opcodes[8]; /* Buffered opcodes. */
87 int opcode_cnt; /* Number of buffered opcodes. */
88 uint8_t data[8][8]; /* Buffered data. */
89 int data_cnt; /* Number of buffered data items. */
92 struct sfm_var *sfm_vars; /* Variables. */
93 size_t sfm_var_cnt; /* Number of variables. */
94 size_t segment_cnt; /* Number of variables including extra segments
95 for long string variables. */
98 static const struct casewriter_class sys_file_casewriter_class;
100 static void write_header (struct sfm_writer *, const struct dictionary *);
101 static void write_variable (struct sfm_writer *, const struct variable *);
102 static void write_value_labels (struct sfm_writer *,
103 const struct dictionary *);
104 static void write_integer_info_record (struct sfm_writer *,
105 const struct dictionary *);
106 static void write_float_info_record (struct sfm_writer *);
108 static void write_longvar_table (struct sfm_writer *w,
109 const struct dictionary *dict);
111 static void write_encoding_record (struct sfm_writer *w,
112 const struct dictionary *);
114 static void write_vls_length_table (struct sfm_writer *w,
115 const struct dictionary *dict);
117 static void write_long_string_value_labels (struct sfm_writer *,
118 const struct dictionary *);
120 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
123 static void write_variable_display_parameters (struct sfm_writer *w,
124 const struct dictionary *dict);
126 static void write_documents (struct sfm_writer *, const struct dictionary *);
128 static void write_data_file_attributes (struct sfm_writer *,
129 const struct dictionary *);
130 static void write_variable_attributes (struct sfm_writer *,
131 const struct dictionary *);
133 static void write_int (struct sfm_writer *, int32_t);
134 static inline void convert_double_to_output_format (double, uint8_t[8]);
135 static void write_float (struct sfm_writer *, double);
136 static void write_string (struct sfm_writer *, const char *, size_t);
137 static void write_utf8_string (struct sfm_writer *, const char *encoding,
138 const char *string, size_t width);
139 static void write_utf8_record (struct sfm_writer *, const char *encoding,
140 const struct string *content, int subtype);
141 static void write_string_record (struct sfm_writer *,
142 const struct substring content, int subtype);
143 static void write_bytes (struct sfm_writer *, const void *, size_t);
144 static void write_zeros (struct sfm_writer *, size_t);
145 static void write_spaces (struct sfm_writer *, size_t);
146 static void write_value (struct sfm_writer *, const union value *, int width);
148 static void write_case_uncompressed (struct sfm_writer *,
149 const struct ccase *);
150 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
151 static void flush_compressed (struct sfm_writer *);
152 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
153 static void put_cmp_number (struct sfm_writer *, double);
154 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
156 static bool write_error (const struct sfm_writer *);
157 static bool close_writer (struct sfm_writer *);
159 /* Returns default options for writing a system file. */
160 struct sfm_write_options
161 sfm_writer_default_options (void)
163 struct sfm_write_options opts;
164 opts.create_writeable = true;
165 opts.compress = settings_get_scompression ();
170 /* Opens the system file designated by file handle FH for writing
171 cases from dictionary D according to the given OPTS.
173 No reference to D is retained, so it may be modified or
174 destroyed at will after this function returns. D is not
175 modified by this function, except to assign short names. */
177 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
178 struct sfm_write_options opts)
180 struct encoding_info encoding_info;
181 struct sfm_writer *w;
186 if (opts.version != 2 && opts.version != 3)
188 msg (ME, _("Unknown system file version %d. Treating as version %d."),
193 /* Create and initialize writer. */
194 w = xmalloc (sizeof *w);
200 w->compress = opts.compress;
203 w->opcode_cnt = w->data_cnt = 0;
205 /* Figure out how to map in-memory case data to on-disk case
206 data. Also count the number of segments. Very long strings
207 occupy multiple segments, otherwise each variable only takes
209 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
212 /* Open file handle as an exclusive writer. */
213 /* TRANSLATORS: this fragment will be interpolated into
214 messages in fh_lock() that identify types of files. */
215 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
219 /* Create the file on disk. */
221 if (opts.create_writeable)
223 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
227 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
228 fh_get_file_name (fh), strerror (errno));
232 get_encoding_info (&encoding_info, dict_get_encoding (d));
233 w->space = encoding_info.space[0];
235 /* Write the file header. */
238 /* Write basic variable info. */
239 short_names_assign (d);
240 for (i = 0; i < dict_get_var_cnt (d); i++)
241 write_variable (w, dict_get_var (d, i));
243 write_value_labels (w, d);
245 if (dict_get_document_line_cnt (d) > 0)
246 write_documents (w, d);
248 write_integer_info_record (w, d);
249 write_float_info_record (w);
251 write_mrsets (w, d, true);
253 write_variable_display_parameters (w, d);
255 if (opts.version >= 3)
256 write_longvar_table (w, d);
258 write_vls_length_table (w, d);
260 write_long_string_value_labels (w, d);
262 if (opts.version >= 3)
264 if (attrset_count (dict_get_attributes (d)))
265 write_data_file_attributes (w, d);
266 write_variable_attributes (w, d);
269 write_mrsets (w, d, false);
271 write_encoding_record (w, d);
273 /* Write end-of-headers record. */
280 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
287 /* Returns value of X truncated to two least-significant digits. */
298 /* Calculates the offset of data for TARGET_VAR from the
299 beginning of each case's data for dictionary D. The return
300 value is in "octs" (8-byte units). */
302 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
308 for (i = 0; i < dict_get_var_cnt (d); i++)
310 struct variable *var = dict_get_var (d, i);
311 if (var == target_var)
313 oct_idx += sfm_width_to_octs (var_get_width (var));
318 /* Write the sysfile_header header to system file W. */
320 write_header (struct sfm_writer *w, const struct dictionary *d)
322 const char *dict_encoding = dict_get_encoding (d);
324 char creation_date[10];
325 char creation_time[9];
326 const char *file_label;
327 struct variable *weight;
331 /* Record-type code. */
332 if (is_encoding_ebcdic_compatible (dict_encoding))
333 write_string (w, EBCDIC_MAGIC, 4);
335 write_string (w, ASCII_MAGIC, 4);
337 /* Product identification. */
338 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
339 version, host_system);
340 write_utf8_string (w, dict_encoding, prod_name, 60);
345 /* Number of `union value's per case. */
346 write_int (w, calc_oct_idx (d, NULL));
349 write_int (w, w->compress);
351 /* Weight variable. */
352 weight = dict_get_weight (d);
353 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
355 /* Number of cases. We don't know this in advance, so we write
356 -1 to indicate an unknown number of cases. Later we can
357 come back and overwrite it with the true value. */
360 /* Compression bias. */
361 write_float (w, COMPRESSION_BIAS);
363 /* Creation date and time. */
364 if (time (&t) == (time_t) -1)
366 strcpy (creation_date, "01 Jan 70");
367 strcpy (creation_time, "00:00:00");
371 static const char *const month_name[12] =
373 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
374 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
376 struct tm *tmp = localtime (&t);
377 int day = rerange (tmp->tm_mday);
378 int mon = rerange (tmp->tm_mon + 1);
379 int year = rerange (tmp->tm_year);
380 int hour = rerange (tmp->tm_hour + 1);
381 int min = rerange (tmp->tm_min + 1);
382 int sec = rerange (tmp->tm_sec + 1);
384 snprintf (creation_date, sizeof creation_date,
385 "%02d %s %02d", day, month_name[mon - 1], year);
386 snprintf (creation_time, sizeof creation_time,
387 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
389 write_utf8_string (w, dict_encoding, creation_date, 9);
390 write_utf8_string (w, dict_encoding, creation_time, 8);
393 file_label = dict_get_label (d);
394 if (file_label == NULL)
396 write_utf8_string (w, dict_encoding, file_label, 64);
402 /* Write format spec FMT to W, after adjusting it to be
403 compatible with the given WIDTH. */
405 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
407 assert (fmt_check_output (&fmt));
408 assert (sfm_width_to_segments (width) == 1);
411 fmt_resize (&fmt, width);
412 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
415 /* Write a string continuation variable record for each 8-byte
416 section beyond the initial 8 bytes, for a variable of the
419 write_variable_continuation_records (struct sfm_writer *w, int width)
423 assert (sfm_width_to_segments (width) == 1);
424 for (position = 8; position < width; position += 8)
426 write_int (w, 2); /* Record type. */
427 write_int (w, -1); /* Width. */
428 write_int (w, 0); /* No variable label. */
429 write_int (w, 0); /* No missing values. */
430 write_int (w, 0); /* Print format. */
431 write_int (w, 0); /* Write format. */
432 write_zeros (w, 8); /* Name. */
436 /* Write the variable record(s) for variable V to system file
439 write_variable (struct sfm_writer *w, const struct variable *v)
441 int width = var_get_width (v);
442 int segment_cnt = sfm_width_to_segments (width);
443 int seg0_width = sfm_segment_alloc_width (width, 0);
444 const char *encoding = var_get_encoding (v);
445 struct missing_values mv;
452 write_int (w, seg0_width);
454 /* Variable has a variable label? */
455 write_int (w, var_has_label (v));
457 /* Number of missing values. If there is a range, then the
458 range counts as 2 missing values and causes the number to be
460 mv_copy (&mv, var_get_missing_values (v));
461 if (mv_get_width (&mv) > 8)
463 if (mv_has_range (&mv))
464 write_int (w, -2 - mv_n_values (&mv));
466 write_int (w, mv_n_values (&mv));
468 /* Print and write formats. */
469 write_format (w, *var_get_print_format (v), seg0_width);
470 write_format (w, *var_get_write_format (v), seg0_width);
473 The full name is in a translation table written
475 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
478 if (var_has_label (v))
480 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
481 size_t label_len = MIN (strlen (label), 255);
482 size_t padded_len = ROUND_UP (label_len, 4);
483 write_int (w, label_len);
484 write_string (w, label, padded_len);
488 /* Write the missing values, if any, range first. */
489 if (mv_has_range (&mv))
492 mv_get_range (&mv, &x, &y);
496 for (i = 0; i < mv_n_values (&mv); i++)
497 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
499 write_variable_continuation_records (w, seg0_width);
501 /* Write additional segments for very long string variables. */
502 for (i = 1; i < segment_cnt; i++)
504 int seg_width = sfm_segment_alloc_width (width, i);
505 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
507 write_int (w, 2); /* Variable record. */
508 write_int (w, seg_width); /* Width. */
509 write_int (w, 0); /* No variable label. */
510 write_int (w, 0); /* No missing values. */
511 write_format (w, fmt, seg_width); /* Print format. */
512 write_format (w, fmt, seg_width); /* Write format. */
513 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
515 write_variable_continuation_records (w, seg_width);
521 /* Writes the value labels to system file W.
523 Value labels for long string variables are written separately,
524 by write_long_string_value_labels. */
526 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
530 struct hmap_node hmap_node;
531 const struct val_labs *val_labs;
533 size_t n_indexes, allocated_indexes;
536 size_t n_sets, allocated_sets;
537 struct label_set **sets;
538 struct hmap same_sets;
542 n_sets = allocated_sets = 0;
544 hmap_init (&same_sets);
547 for (i = 0; i < dict_get_var_cnt (d); i++)
549 struct variable *v = dict_get_var (d, i);
551 if (var_has_value_labels (v) && var_get_width (v) <= 8)
553 const struct val_labs *val_labs = var_get_value_labels (v);
554 unsigned int hash = val_labs_hash (val_labs, 0);
555 struct label_set *set;
557 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
560 if (val_labs_equal (set->val_labs, val_labs))
562 if (set->n_indexes >= set->allocated_indexes)
563 set->indexes = x2nrealloc (set->indexes,
564 &set->allocated_indexes,
565 sizeof *set->indexes);
566 set->indexes[set->n_indexes++] = idx;
571 set = xmalloc (sizeof *set);
572 set->val_labs = val_labs;
573 set->indexes = xmalloc (sizeof *set->indexes);
574 set->indexes[0] = idx;
576 set->allocated_indexes = 1;
577 hmap_insert (&same_sets, &set->hmap_node, hash);
579 if (n_sets >= allocated_sets)
580 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
581 sets[n_sets++] = set;
585 idx += sfm_width_to_octs (var_get_width (v));
588 for (i = 0; i < n_sets; i++)
590 const struct label_set *set = sets[i];
591 const struct val_labs *val_labs = set->val_labs;
592 size_t n_labels = val_labs_count (val_labs);
593 int width = val_labs_get_width (val_labs);
594 const struct val_lab **labels;
597 /* Value label record. */
598 write_int (w, 3); /* Record type. */
599 write_int (w, n_labels);
600 labels = val_labs_sorted (val_labs);
601 for (j = 0; j < n_labels; j++)
603 const struct val_lab *vl = labels[j];
604 char *label = recode_string (dict_get_encoding (d), UTF8,
605 val_lab_get_escaped_label (vl), -1);
606 uint8_t len = MIN (strlen (label), 255);
608 write_value (w, val_lab_get_value (vl), width);
609 write_bytes (w, &len, 1);
610 write_bytes (w, label, len);
611 write_zeros (w, REM_RND_UP (len + 1, 8));
616 /* Value label variable record. */
617 write_int (w, 4); /* Record type. */
618 write_int (w, set->n_indexes);
619 for (j = 0; j < set->n_indexes; j++)
620 write_int (w, set->indexes[j] + 1);
623 for (i = 0; i < n_sets; i++)
625 struct label_set *set = sets[i];
631 hmap_destroy (&same_sets);
634 /* Writes record type 6, document record. */
636 write_documents (struct sfm_writer *w, const struct dictionary *d)
638 const struct string_array *docs = dict_get_documents (d);
639 const char *enc = dict_get_encoding (d);
642 write_int (w, 6); /* Record type. */
643 write_int (w, docs->n);
644 for (i = 0; i < docs->n; i++)
646 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
647 size_t s_len = strlen (s);
648 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
650 write_bytes (w, s, write_len);
651 write_spaces (w, DOC_LINE_LENGTH - write_len);
657 put_attrset (struct string *string, const struct attrset *attrs)
659 const struct attribute *attr;
660 struct attrset_iterator i;
662 for (attr = attrset_first (attrs, &i); attr != NULL;
663 attr = attrset_next (attrs, &i))
665 size_t n_values = attribute_get_n_values (attr);
668 ds_put_cstr (string, attribute_get_name (attr));
669 ds_put_byte (string, '(');
670 for (j = 0; j < n_values; j++)
671 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
672 ds_put_byte (string, ')');
677 write_data_file_attributes (struct sfm_writer *w,
678 const struct dictionary *d)
680 struct string s = DS_EMPTY_INITIALIZER;
681 put_attrset (&s, dict_get_attributes (d));
682 write_utf8_record (w, dict_get_encoding (d), &s, 17);
687 add_role_attribute (enum var_role role, struct attrset *attrs)
689 struct attribute *attr;
719 attrset_delete (attrs, "$@Role");
721 attr = attribute_create ("$@Role");
722 attribute_add_value (attr, s);
723 attrset_add (attrs, attr);
727 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
729 struct string s = DS_EMPTY_INITIALIZER;
730 size_t n_vars = dict_get_var_cnt (d);
731 size_t n_attrsets = 0;
734 for (i = 0; i < n_vars; i++)
736 struct variable *v = dict_get_var (d, i);
737 struct attrset attrs;
739 attrset_clone (&attrs, var_get_attributes (v));
741 add_role_attribute (var_get_role (v), &attrs);
743 ds_put_byte (&s, '/');
744 ds_put_format (&s, "%s:", var_get_name (v));
745 put_attrset (&s, &attrs);
746 attrset_destroy (&attrs);
749 write_utf8_record (w, dict_get_encoding (d), &s, 18);
753 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
754 SPSS before release 14, otherwise writes sets supported only by later
757 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
760 const char *encoding = dict_get_encoding (dict);
761 struct string s = DS_EMPTY_INITIALIZER;
765 if (is_encoding_ebcdic_compatible (encoding))
771 n_mrsets = dict_get_n_mrsets (dict);
775 for (i = 0; i < n_mrsets; i++)
777 const struct mrset *mrset = dict_get_mrset (dict, i);
781 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
785 name = recode_string (encoding, "UTF-8", mrset->name, -1);
786 ds_put_format (&s, "%s=", name);
789 if (mrset->type == MRSET_MD)
793 if (mrset->cat_source == MRSET_COUNTEDVALUES)
794 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
796 ds_put_byte (&s, 'D');
798 if (mrset->width == 0)
799 counted = xasprintf ("%.0f", mrset->counted.f);
801 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
803 ds_put_format (&s, "%zu %s", strlen (counted), counted);
807 ds_put_byte (&s, 'C');
808 ds_put_byte (&s, ' ');
810 if (mrset->label && !mrset->label_from_var_label)
812 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
813 ds_put_format (&s, "%zu %s", strlen (label), label);
817 ds_put_cstr (&s, "0 ");
819 for (j = 0; j < mrset->n_vars; j++)
821 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
822 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
823 char *short_name = recode_string (encoding, "UTF-8",
824 lower_name_utf8, -1);
825 ds_put_format (&s, " %s", short_name);
827 free (lower_name_utf8);
829 ds_put_byte (&s, '\n');
832 if (!ds_is_empty (&s))
833 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
837 /* Write the alignment, width and scale values. */
839 write_variable_display_parameters (struct sfm_writer *w,
840 const struct dictionary *dict)
844 write_int (w, 7); /* Record type. */
845 write_int (w, 11); /* Record subtype. */
846 write_int (w, 4); /* Data item (int32) size. */
847 write_int (w, w->segment_cnt * 3); /* Number of data items. */
849 for (i = 0; i < dict_get_var_cnt (dict); ++i)
851 struct variable *v = dict_get_var (dict, i);
852 int width = var_get_width (v);
853 int segment_cnt = sfm_width_to_segments (width);
854 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
855 : var_get_measure (v) == MEASURE_ORDINAL ? 2
857 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
858 : var_get_alignment (v) == ALIGN_RIGHT ? 1
862 for (i = 0; i < segment_cnt; i++)
864 int width_left = width - sfm_segment_effective_offset (width, i);
865 write_int (w, measure);
866 write_int (w, (i == 0 ? var_get_display_width (v)
867 : var_default_display_width (width_left)));
868 write_int (w, alignment);
873 /* Writes the table of lengths for very long string variables. */
875 write_vls_length_table (struct sfm_writer *w,
876 const struct dictionary *dict)
881 ds_init_empty (&map);
882 for (i = 0; i < dict_get_var_cnt (dict); ++i)
884 const struct variable *v = dict_get_var (dict, i);
885 if (sfm_width_to_segments (var_get_width (v)) > 1)
886 ds_put_format (&map, "%s=%05d%c\t",
887 var_get_short_name (v, 0), var_get_width (v), 0);
889 if (!ds_is_empty (&map))
890 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
896 write_long_string_value_labels (struct sfm_writer *w,
897 const struct dictionary *dict)
899 size_t n_vars = dict_get_var_cnt (dict);
903 /* Figure out the size in advance. */
905 for (i = 0; i < n_vars; i++)
907 struct variable *var = dict_get_var (dict, i);
908 const struct val_labs *val_labs = var_get_value_labels (var);
909 const char *encoding = var_get_encoding (var);
910 int width = var_get_width (var);
911 const struct val_lab *val_lab;
913 if (val_labs_count (val_labs) == 0 || width < 9)
917 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
918 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
919 val_lab = val_labs_next (val_labs, val_lab))
922 size += recode_string_len (encoding, "UTF-8",
923 val_lab_get_escaped_label (val_lab), -1);
929 write_int (w, 7); /* Record type. */
930 write_int (w, 21); /* Record subtype */
931 write_int (w, 1); /* Data item (byte) size. */
932 write_int (w, size); /* Number of data items. */
934 start = ftello (w->file);
935 for (i = 0; i < n_vars; i++)
937 struct variable *var = dict_get_var (dict, i);
938 const struct val_labs *val_labs = var_get_value_labels (var);
939 const char *encoding = var_get_encoding (var);
940 int width = var_get_width (var);
941 const struct val_lab *val_lab;
944 if (val_labs_count (val_labs) == 0 || width < 9)
947 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
948 write_int (w, strlen (var_name));
949 write_bytes (w, var_name, strlen (var_name));
952 write_int (w, width);
953 write_int (w, val_labs_count (val_labs));
954 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
955 val_lab = val_labs_next (val_labs, val_lab))
960 write_int (w, width);
961 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
964 label = recode_string (var_get_encoding (var), "UTF-8",
965 val_lab_get_escaped_label (val_lab), -1);
966 len = strlen (label);
968 write_bytes (w, label, len);
972 assert (ftello (w->file) == start + size);
976 write_encoding_record (struct sfm_writer *w,
977 const struct dictionary *d)
979 /* IANA says "...character set names may be up to 40 characters taken
980 from the printable characters of US-ASCII," so character set names
981 don't need to be recoded to be in UTF-8.
983 We convert encoding names to uppercase because SPSS writes encoding
984 names in uppercase. */
985 char *encoding = xstrdup (dict_get_encoding (d));
986 str_uppercase (encoding);
987 write_string_record (w, ss_cstr (encoding), 20);
991 /* Writes the long variable name table. */
993 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
998 ds_init_empty (&map);
999 for (i = 0; i < dict_get_var_cnt (dict); i++)
1001 struct variable *v = dict_get_var (dict, i);
1003 ds_put_byte (&map, '\t');
1004 ds_put_format (&map, "%s=%s",
1005 var_get_short_name (v, 0), var_get_name (v));
1007 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1011 /* Write integer information record. */
1013 write_integer_info_record (struct sfm_writer *w,
1014 const struct dictionary *d)
1016 const char *dict_encoding = dict_get_encoding (d);
1017 int version_component[3];
1021 /* Parse the version string. */
1022 memset (version_component, 0, sizeof version_component);
1023 sscanf (bare_version, "%d.%d.%d",
1024 &version_component[0], &version_component[1], &version_component[2]);
1026 /* Figure out the floating-point format. */
1027 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1028 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1030 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1032 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1037 /* Choose codepage. */
1038 codepage = sys_get_codepage_from_encoding (dict_encoding);
1041 /* The codepage is unknown. Choose a default.
1043 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1045 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1046 many files use this codepage number regardless of their actual
1049 if (is_encoding_ascii_compatible (dict_encoding))
1051 else if (is_encoding_ebcdic_compatible (dict_encoding))
1056 write_int (w, 7); /* Record type. */
1057 write_int (w, 3); /* Record subtype. */
1058 write_int (w, 4); /* Data item (int32) size. */
1059 write_int (w, 8); /* Number of data items. */
1060 write_int (w, version_component[0]);
1061 write_int (w, version_component[1]);
1062 write_int (w, version_component[2]);
1063 write_int (w, -1); /* Machine code. */
1064 write_int (w, float_format);
1065 write_int (w, 1); /* Compression code. */
1066 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1067 write_int (w, codepage);
1070 /* Write floating-point information record. */
1072 write_float_info_record (struct sfm_writer *w)
1074 write_int (w, 7); /* Record type. */
1075 write_int (w, 4); /* Record subtype. */
1076 write_int (w, 8); /* Data item (flt64) size. */
1077 write_int (w, 3); /* Number of data items. */
1078 write_float (w, SYSMIS); /* System-missing value. */
1079 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1080 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1083 /* Writes case C to system file W. */
1085 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1088 struct sfm_writer *w = w_;
1090 if (ferror (w->file))
1092 casewriter_force_error (writer);
1100 write_case_uncompressed (w, c);
1102 write_case_compressed (w, c);
1107 /* Destroys system file writer W. */
1109 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1111 struct sfm_writer *w = w_;
1112 if (!close_writer (w))
1113 casewriter_force_error (writer);
1116 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1118 write_error (const struct sfm_writer *writer)
1120 return ferror (writer->file);
1123 /* Closes a system file after we're done with it.
1124 Returns true if successful, false if an I/O error occurred. */
1126 close_writer (struct sfm_writer *w)
1134 if (w->file != NULL)
1137 if (w->opcode_cnt > 0)
1138 flush_compressed (w);
1141 ok = !write_error (w);
1143 /* Seek back to the beginning and update the number of cases.
1144 This is just a courtesy to later readers, so there's no need
1145 to check return values or report errors. */
1146 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1148 write_int (w, w->case_cnt);
1152 if (fclose (w->file) == EOF)
1156 msg (ME, _("An I/O error occurred writing system file `%s'."),
1157 fh_get_file_name (w->fh));
1159 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1163 fh_unlock (w->lock);
1172 /* System file writer casewriter class. */
1173 static const struct casewriter_class sys_file_casewriter_class =
1175 sys_file_casewriter_write,
1176 sys_file_casewriter_destroy,
1180 /* Writes case C to system file W, without compressing it. */
1182 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1186 for (i = 0; i < w->sfm_var_cnt; i++)
1188 struct sfm_var *v = &w->sfm_vars[i];
1190 if (v->var_width == 0)
1191 write_float (w, case_num_idx (c, v->case_index));
1194 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1196 write_spaces (w, v->padding);
1201 /* Writes case C to system file W, with compression. */
1203 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1207 for (i = 0; i < w->sfm_var_cnt; i++)
1209 struct sfm_var *v = &w->sfm_vars[i];
1211 if (v->var_width == 0)
1213 double d = case_num_idx (c, v->case_index);
1215 put_cmp_opcode (w, 255);
1216 else if (d >= 1 - COMPRESSION_BIAS
1217 && d <= 251 - COMPRESSION_BIAS
1219 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1222 put_cmp_opcode (w, 253);
1223 put_cmp_number (w, d);
1228 int offset = v->offset;
1231 /* This code properly deals with a width that is not a
1232 multiple of 8, by ensuring that the final partial
1233 oct (8 byte unit) is treated as padded with spaces
1235 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1237 const void *data = case_str_idx (c, v->case_index) + offset;
1238 int chunk_size = MIN (width, 8);
1239 if (!memcmp (data, " ", chunk_size))
1240 put_cmp_opcode (w, 254);
1243 put_cmp_opcode (w, 253);
1244 put_cmp_string (w, data, chunk_size);
1248 /* This code deals properly with padding that is not a
1249 multiple of 8 bytes, by discarding the remainder,
1250 which was already effectively padded with spaces in
1251 the previous loop. (Note that v->width + v->padding
1252 is always a multiple of 8.) */
1253 for (padding = v->padding / 8; padding > 0; padding--)
1254 put_cmp_opcode (w, 254);
1259 /* Flushes buffered compressed opcodes and data to W.
1260 The compression buffer must not be empty. */
1262 flush_compressed (struct sfm_writer *w)
1264 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1266 write_bytes (w, w->opcodes, w->opcode_cnt);
1267 write_zeros (w, 8 - w->opcode_cnt);
1269 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1271 w->opcode_cnt = w->data_cnt = 0;
1274 /* Appends OPCODE to the buffered set of compression opcodes in
1275 W. Flushes the compression buffer beforehand if necessary. */
1277 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1279 if (w->opcode_cnt >= 8)
1280 flush_compressed (w);
1282 w->opcodes[w->opcode_cnt++] = opcode;
1285 /* Appends NUMBER to the buffered compression data in W. The
1286 buffer must not be full; the way to assure that is to call
1287 this function only just after a call to put_cmp_opcode, which
1288 will flush the buffer as necessary. */
1290 put_cmp_number (struct sfm_writer *w, double number)
1292 assert (w->opcode_cnt > 0);
1293 assert (w->data_cnt < 8);
1295 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1298 /* Appends SIZE bytes of DATA to the buffered compression data in
1299 W, followed by enough spaces to pad the output data to exactly
1300 8 bytes (thus, SIZE must be no greater than 8). The buffer
1301 must not be full; the way to assure that is to call this
1302 function only just after a call to put_cmp_opcode, which will
1303 flush the buffer as necessary. */
1305 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1307 assert (w->opcode_cnt > 0);
1308 assert (w->data_cnt < 8);
1311 memset (w->data[w->data_cnt], w->space, 8);
1312 memcpy (w->data[w->data_cnt], data, size);
1316 /* Writes 32-bit integer X to the output file for writer W. */
1318 write_int (struct sfm_writer *w, int32_t x)
1320 write_bytes (w, &x, sizeof x);
1323 /* Converts NATIVE to the 64-bit format used in output files in
1326 convert_double_to_output_format (double native, uint8_t output[8])
1328 /* If "double" is not a 64-bit type, then convert it to a
1329 64-bit type. Otherwise just copy it. */
1330 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1331 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1333 memcpy (output, &native, sizeof native);
1336 /* Writes floating-point number X to the output file for writer
1339 write_float (struct sfm_writer *w, double x)
1342 convert_double_to_output_format (x, output);
1343 write_bytes (w, output, sizeof output);
1346 /* Writes contents of VALUE with the given WIDTH to W, padding
1347 with zeros to a multiple of 8 bytes.
1348 To avoid a branch, and because we don't actually need to
1349 support it, WIDTH must be no bigger than 8. */
1351 write_value (struct sfm_writer *w, const union value *value, int width)
1353 assert (width <= 8);
1355 write_float (w, value->f);
1358 write_bytes (w, value_str (value, width), width);
1359 write_zeros (w, 8 - width);
1363 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1364 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1365 is padded on the right with spaces. */
1367 write_string (struct sfm_writer *w, const char *string, size_t width)
1369 size_t data_bytes = MIN (strlen (string), width);
1370 size_t pad_bytes = width - data_bytes;
1371 write_bytes (w, string, data_bytes);
1372 while (pad_bytes-- > 0)
1373 putc (w->space, w->file);
1376 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1377 recoded version in a field of the given WIDTH to W. The string is truncated
1378 or padded on the right with spaces to exactly WIDTH bytes. */
1380 write_utf8_string (struct sfm_writer *w, const char *encoding,
1381 const char *string, size_t width)
1383 char *s = recode_string (encoding, "UTF-8", string, -1);
1384 write_string (w, s, width);
1388 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1389 from UTF-8 encoded into ENCODING. */
1391 write_utf8_record (struct sfm_writer *w, const char *encoding,
1392 const struct string *content, int subtype)
1396 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1397 write_string_record (w, s, subtype);
1401 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1404 write_string_record (struct sfm_writer *w,
1405 const struct substring content, int subtype)
1408 write_int (w, subtype);
1410 write_int (w, ss_length (content));
1411 write_bytes (w, ss_data (content), ss_length (content));
1414 /* Writes SIZE bytes of DATA to W's output file. */
1416 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1418 fwrite (data, 1, size, w->file);
1421 /* Writes N zeros to W's output file. */
1423 write_zeros (struct sfm_writer *w, size_t n)
1429 /* Writes N spaces to W's output file. */
1431 write_spaces (struct sfm_writer *w, size_t n)
1434 putc (w->space, w->file);