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 *);
119 static void write_long_string_missing_values (struct sfm_writer *,
120 const struct dictionary *);
122 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
125 static void write_variable_display_parameters (struct sfm_writer *w,
126 const struct dictionary *dict);
128 static void write_documents (struct sfm_writer *, const struct dictionary *);
130 static void write_data_file_attributes (struct sfm_writer *,
131 const struct dictionary *);
132 static void write_variable_attributes (struct sfm_writer *,
133 const struct dictionary *);
135 static void write_int (struct sfm_writer *, int32_t);
136 static inline void convert_double_to_output_format (double, uint8_t[8]);
137 static void write_float (struct sfm_writer *, double);
138 static void write_string (struct sfm_writer *, const char *, size_t);
139 static void write_utf8_string (struct sfm_writer *, const char *encoding,
140 const char *string, size_t width);
141 static void write_utf8_record (struct sfm_writer *, const char *encoding,
142 const struct string *content, int subtype);
143 static void write_string_record (struct sfm_writer *,
144 const struct substring content, int subtype);
145 static void write_bytes (struct sfm_writer *, const void *, size_t);
146 static void write_zeros (struct sfm_writer *, size_t);
147 static void write_spaces (struct sfm_writer *, size_t);
148 static void write_value (struct sfm_writer *, const union value *, int width);
150 static void write_case_uncompressed (struct sfm_writer *,
151 const struct ccase *);
152 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
153 static void flush_compressed (struct sfm_writer *);
154 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
155 static void put_cmp_number (struct sfm_writer *, double);
156 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
158 static bool write_error (const struct sfm_writer *);
159 static bool close_writer (struct sfm_writer *);
161 /* Returns default options for writing a system file. */
162 struct sfm_write_options
163 sfm_writer_default_options (void)
165 struct sfm_write_options opts;
166 opts.create_writeable = true;
167 opts.compress = settings_get_scompression ();
172 /* Opens the system file designated by file handle FH for writing
173 cases from dictionary D according to the given OPTS.
175 No reference to D is retained, so it may be modified or
176 destroyed at will after this function returns. D is not
177 modified by this function, except to assign short names. */
179 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
180 struct sfm_write_options opts)
182 struct encoding_info encoding_info;
183 struct sfm_writer *w;
188 if (opts.version != 2 && opts.version != 3)
190 msg (ME, _("Unknown system file version %d. Treating as version %d."),
195 /* Create and initialize writer. */
196 w = xmalloc (sizeof *w);
202 w->compress = opts.compress;
205 w->opcode_cnt = w->data_cnt = 0;
207 /* Figure out how to map in-memory case data to on-disk case
208 data. Also count the number of segments. Very long strings
209 occupy multiple segments, otherwise each variable only takes
211 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
214 /* Open file handle as an exclusive writer. */
215 /* TRANSLATORS: this fragment will be interpolated into
216 messages in fh_lock() that identify types of files. */
217 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
221 /* Create the file on disk. */
223 if (opts.create_writeable)
225 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
229 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
230 fh_get_file_name (fh), strerror (errno));
234 get_encoding_info (&encoding_info, dict_get_encoding (d));
235 w->space = encoding_info.space[0];
237 /* Write the file header. */
240 /* Write basic variable info. */
241 short_names_assign (d);
242 for (i = 0; i < dict_get_var_cnt (d); i++)
243 write_variable (w, dict_get_var (d, i));
245 write_value_labels (w, d);
247 if (dict_get_document_line_cnt (d) > 0)
248 write_documents (w, d);
250 write_integer_info_record (w, d);
251 write_float_info_record (w);
253 write_mrsets (w, d, true);
255 write_variable_display_parameters (w, d);
257 if (opts.version >= 3)
258 write_longvar_table (w, d);
260 write_vls_length_table (w, d);
262 write_long_string_value_labels (w, d);
263 write_long_string_missing_values (w, d);
265 if (opts.version >= 3)
267 if (attrset_count (dict_get_attributes (d)))
268 write_data_file_attributes (w, d);
269 write_variable_attributes (w, d);
272 write_mrsets (w, d, false);
274 write_encoding_record (w, d);
276 /* Write end-of-headers record. */
283 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
290 /* Returns value of X truncated to two least-significant digits. */
301 /* Calculates the offset of data for TARGET_VAR from the
302 beginning of each case's data for dictionary D. The return
303 value is in "octs" (8-byte units). */
305 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
311 for (i = 0; i < dict_get_var_cnt (d); i++)
313 struct variable *var = dict_get_var (d, i);
314 if (var == target_var)
316 oct_idx += sfm_width_to_octs (var_get_width (var));
321 /* Write the sysfile_header header to system file W. */
323 write_header (struct sfm_writer *w, const struct dictionary *d)
325 const char *dict_encoding = dict_get_encoding (d);
327 char creation_date[10];
328 char creation_time[9];
329 const char *file_label;
330 struct variable *weight;
334 /* Record-type code. */
335 if (is_encoding_ebcdic_compatible (dict_encoding))
336 write_string (w, EBCDIC_MAGIC, 4);
338 write_string (w, ASCII_MAGIC, 4);
340 /* Product identification. */
341 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
342 version, host_system);
343 write_utf8_string (w, dict_encoding, prod_name, 60);
348 /* Number of `union value's per case. */
349 write_int (w, calc_oct_idx (d, NULL));
352 write_int (w, w->compress);
354 /* Weight variable. */
355 weight = dict_get_weight (d);
356 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
358 /* Number of cases. We don't know this in advance, so we write
359 -1 to indicate an unknown number of cases. Later we can
360 come back and overwrite it with the true value. */
363 /* Compression bias. */
364 write_float (w, COMPRESSION_BIAS);
366 /* Creation date and time. */
367 if (time (&t) == (time_t) -1)
369 strcpy (creation_date, "01 Jan 70");
370 strcpy (creation_time, "00:00:00");
374 static const char *const month_name[12] =
376 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
377 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
379 struct tm *tmp = localtime (&t);
380 int day = rerange (tmp->tm_mday);
381 int mon = rerange (tmp->tm_mon + 1);
382 int year = rerange (tmp->tm_year);
383 int hour = rerange (tmp->tm_hour + 1);
384 int min = rerange (tmp->tm_min + 1);
385 int sec = rerange (tmp->tm_sec + 1);
387 snprintf (creation_date, sizeof creation_date,
388 "%02d %s %02d", day, month_name[mon - 1], year);
389 snprintf (creation_time, sizeof creation_time,
390 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
392 write_utf8_string (w, dict_encoding, creation_date, 9);
393 write_utf8_string (w, dict_encoding, creation_time, 8);
396 file_label = dict_get_label (d);
397 if (file_label == NULL)
399 write_utf8_string (w, dict_encoding, file_label, 64);
405 /* Write format spec FMT to W, after adjusting it to be
406 compatible with the given WIDTH. */
408 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
410 assert (fmt_check_output (&fmt));
411 assert (sfm_width_to_segments (width) == 1);
414 fmt_resize (&fmt, width);
415 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
418 /* Write a string continuation variable record for each 8-byte
419 section beyond the initial 8 bytes, for a variable of the
422 write_variable_continuation_records (struct sfm_writer *w, int width)
426 assert (sfm_width_to_segments (width) == 1);
427 for (position = 8; position < width; position += 8)
429 write_int (w, 2); /* Record type. */
430 write_int (w, -1); /* Width. */
431 write_int (w, 0); /* No variable label. */
432 write_int (w, 0); /* No missing values. */
433 write_int (w, 0); /* Print format. */
434 write_int (w, 0); /* Write format. */
435 write_zeros (w, 8); /* Name. */
439 /* Write the variable record(s) for variable V to system file
442 write_variable (struct sfm_writer *w, const struct variable *v)
444 int width = var_get_width (v);
445 int segment_cnt = sfm_width_to_segments (width);
446 int seg0_width = sfm_segment_alloc_width (width, 0);
447 const char *encoding = var_get_encoding (v);
454 write_int (w, seg0_width);
456 /* Variable has a variable label? */
457 write_int (w, var_has_label (v));
459 /* Number of missing values. If there is a range, then the
460 range counts as 2 missing values and causes the number to be
463 Missing values for long string variables are written in a separate
465 if (width <= MAX_SHORT_STRING)
467 const struct missing_values *mv = var_get_missing_values (v);
468 if (mv_has_range (mv))
469 write_int (w, -2 - mv_n_values (mv));
471 write_int (w, mv_n_values (mv));
476 /* Print and write formats. */
477 write_format (w, *var_get_print_format (v), seg0_width);
478 write_format (w, *var_get_write_format (v), seg0_width);
481 The full name is in a translation table written
483 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
486 if (var_has_label (v))
488 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
489 size_t label_len = MIN (strlen (label), 255);
490 size_t padded_len = ROUND_UP (label_len, 4);
491 write_int (w, label_len);
492 write_string (w, label, padded_len);
496 /* Write the missing values, if any, range first. */
497 if (width <= MAX_SHORT_STRING)
499 const struct missing_values *mv = var_get_missing_values (v);
500 if (mv_has_range (mv))
503 mv_get_range (mv, &x, &y);
507 for (i = 0; i < mv_n_values (mv); i++)
508 write_value (w, mv_get_value (mv, i), width);
511 write_variable_continuation_records (w, seg0_width);
513 /* Write additional segments for very long string variables. */
514 for (i = 1; i < segment_cnt; i++)
516 int seg_width = sfm_segment_alloc_width (width, i);
517 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
519 write_int (w, 2); /* Variable record. */
520 write_int (w, seg_width); /* Width. */
521 write_int (w, 0); /* No variable label. */
522 write_int (w, 0); /* No missing values. */
523 write_format (w, fmt, seg_width); /* Print format. */
524 write_format (w, fmt, seg_width); /* Write format. */
525 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
527 write_variable_continuation_records (w, seg_width);
531 /* Writes the value labels to system file W.
533 Value labels for long string variables are written separately,
534 by write_long_string_value_labels. */
536 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
540 struct hmap_node hmap_node;
541 const struct val_labs *val_labs;
543 size_t n_indexes, allocated_indexes;
546 size_t n_sets, allocated_sets;
547 struct label_set **sets;
548 struct hmap same_sets;
552 n_sets = allocated_sets = 0;
554 hmap_init (&same_sets);
557 for (i = 0; i < dict_get_var_cnt (d); i++)
559 struct variable *v = dict_get_var (d, i);
561 if (var_has_value_labels (v) && var_get_width (v) <= 8)
563 const struct val_labs *val_labs = var_get_value_labels (v);
564 unsigned int hash = val_labs_hash (val_labs, 0);
565 struct label_set *set;
567 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
570 if (val_labs_equal (set->val_labs, val_labs))
572 if (set->n_indexes >= set->allocated_indexes)
573 set->indexes = x2nrealloc (set->indexes,
574 &set->allocated_indexes,
575 sizeof *set->indexes);
576 set->indexes[set->n_indexes++] = idx;
581 set = xmalloc (sizeof *set);
582 set->val_labs = val_labs;
583 set->indexes = xmalloc (sizeof *set->indexes);
584 set->indexes[0] = idx;
586 set->allocated_indexes = 1;
587 hmap_insert (&same_sets, &set->hmap_node, hash);
589 if (n_sets >= allocated_sets)
590 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
591 sets[n_sets++] = set;
595 idx += sfm_width_to_octs (var_get_width (v));
598 for (i = 0; i < n_sets; i++)
600 const struct label_set *set = sets[i];
601 const struct val_labs *val_labs = set->val_labs;
602 size_t n_labels = val_labs_count (val_labs);
603 int width = val_labs_get_width (val_labs);
604 const struct val_lab **labels;
607 /* Value label record. */
608 write_int (w, 3); /* Record type. */
609 write_int (w, n_labels);
610 labels = val_labs_sorted (val_labs);
611 for (j = 0; j < n_labels; j++)
613 const struct val_lab *vl = labels[j];
614 char *label = recode_string (dict_get_encoding (d), UTF8,
615 val_lab_get_escaped_label (vl), -1);
616 uint8_t len = MIN (strlen (label), 255);
618 write_value (w, val_lab_get_value (vl), width);
619 write_bytes (w, &len, 1);
620 write_bytes (w, label, len);
621 write_zeros (w, REM_RND_UP (len + 1, 8));
626 /* Value label variable record. */
627 write_int (w, 4); /* Record type. */
628 write_int (w, set->n_indexes);
629 for (j = 0; j < set->n_indexes; j++)
630 write_int (w, set->indexes[j] + 1);
633 for (i = 0; i < n_sets; i++)
635 struct label_set *set = sets[i];
641 hmap_destroy (&same_sets);
644 /* Writes record type 6, document record. */
646 write_documents (struct sfm_writer *w, const struct dictionary *d)
648 const struct string_array *docs = dict_get_documents (d);
649 const char *enc = dict_get_encoding (d);
652 write_int (w, 6); /* Record type. */
653 write_int (w, docs->n);
654 for (i = 0; i < docs->n; i++)
656 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
657 size_t s_len = strlen (s);
658 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
660 write_bytes (w, s, write_len);
661 write_spaces (w, DOC_LINE_LENGTH - write_len);
667 put_attrset (struct string *string, const struct attrset *attrs)
669 const struct attribute *attr;
670 struct attrset_iterator i;
672 for (attr = attrset_first (attrs, &i); attr != NULL;
673 attr = attrset_next (attrs, &i))
675 size_t n_values = attribute_get_n_values (attr);
678 ds_put_cstr (string, attribute_get_name (attr));
679 ds_put_byte (string, '(');
680 for (j = 0; j < n_values; j++)
681 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
682 ds_put_byte (string, ')');
687 write_data_file_attributes (struct sfm_writer *w,
688 const struct dictionary *d)
690 struct string s = DS_EMPTY_INITIALIZER;
691 put_attrset (&s, dict_get_attributes (d));
692 write_utf8_record (w, dict_get_encoding (d), &s, 17);
697 add_role_attribute (enum var_role role, struct attrset *attrs)
699 struct attribute *attr;
729 attrset_delete (attrs, "$@Role");
731 attr = attribute_create ("$@Role");
732 attribute_add_value (attr, s);
733 attrset_add (attrs, attr);
737 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
739 struct string s = DS_EMPTY_INITIALIZER;
740 size_t n_vars = dict_get_var_cnt (d);
741 size_t n_attrsets = 0;
744 for (i = 0; i < n_vars; i++)
746 struct variable *v = dict_get_var (d, i);
747 struct attrset attrs;
749 attrset_clone (&attrs, var_get_attributes (v));
751 add_role_attribute (var_get_role (v), &attrs);
753 ds_put_byte (&s, '/');
754 ds_put_format (&s, "%s:", var_get_name (v));
755 put_attrset (&s, &attrs);
756 attrset_destroy (&attrs);
759 write_utf8_record (w, dict_get_encoding (d), &s, 18);
763 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
764 SPSS before release 14, otherwise writes sets supported only by later
767 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
770 const char *encoding = dict_get_encoding (dict);
771 struct string s = DS_EMPTY_INITIALIZER;
775 if (is_encoding_ebcdic_compatible (encoding))
781 n_mrsets = dict_get_n_mrsets (dict);
785 for (i = 0; i < n_mrsets; i++)
787 const struct mrset *mrset = dict_get_mrset (dict, i);
791 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
795 name = recode_string (encoding, "UTF-8", mrset->name, -1);
796 ds_put_format (&s, "%s=", name);
799 if (mrset->type == MRSET_MD)
803 if (mrset->cat_source == MRSET_COUNTEDVALUES)
804 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
806 ds_put_byte (&s, 'D');
808 if (mrset->width == 0)
809 counted = xasprintf ("%.0f", mrset->counted.f);
811 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
813 ds_put_format (&s, "%zu %s", strlen (counted), counted);
817 ds_put_byte (&s, 'C');
818 ds_put_byte (&s, ' ');
820 if (mrset->label && !mrset->label_from_var_label)
822 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
823 ds_put_format (&s, "%zu %s", strlen (label), label);
827 ds_put_cstr (&s, "0 ");
829 for (j = 0; j < mrset->n_vars; j++)
831 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
832 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
833 char *short_name = recode_string (encoding, "UTF-8",
834 lower_name_utf8, -1);
835 ds_put_format (&s, " %s", short_name);
837 free (lower_name_utf8);
839 ds_put_byte (&s, '\n');
842 if (!ds_is_empty (&s))
843 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
847 /* Write the alignment, width and scale values. */
849 write_variable_display_parameters (struct sfm_writer *w,
850 const struct dictionary *dict)
854 write_int (w, 7); /* Record type. */
855 write_int (w, 11); /* Record subtype. */
856 write_int (w, 4); /* Data item (int32) size. */
857 write_int (w, w->segment_cnt * 3); /* Number of data items. */
859 for (i = 0; i < dict_get_var_cnt (dict); ++i)
861 struct variable *v = dict_get_var (dict, i);
862 int width = var_get_width (v);
863 int segment_cnt = sfm_width_to_segments (width);
864 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
865 : var_get_measure (v) == MEASURE_ORDINAL ? 2
867 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
868 : var_get_alignment (v) == ALIGN_RIGHT ? 1
872 for (i = 0; i < segment_cnt; i++)
874 int width_left = width - sfm_segment_effective_offset (width, i);
875 write_int (w, measure);
876 write_int (w, (i == 0 ? var_get_display_width (v)
877 : var_default_display_width (width_left)));
878 write_int (w, alignment);
883 /* Writes the table of lengths for very long string variables. */
885 write_vls_length_table (struct sfm_writer *w,
886 const struct dictionary *dict)
891 ds_init_empty (&map);
892 for (i = 0; i < dict_get_var_cnt (dict); ++i)
894 const struct variable *v = dict_get_var (dict, i);
895 if (sfm_width_to_segments (var_get_width (v)) > 1)
896 ds_put_format (&map, "%s=%05d%c\t",
897 var_get_short_name (v, 0), var_get_width (v), 0);
899 if (!ds_is_empty (&map))
900 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
905 write_long_string_value_labels (struct sfm_writer *w,
906 const struct dictionary *dict)
908 const char *encoding = dict_get_encoding (dict);
909 size_t n_vars = dict_get_var_cnt (dict);
913 /* Figure out the size in advance. */
915 for (i = 0; i < n_vars; i++)
917 struct variable *var = dict_get_var (dict, i);
918 const struct val_labs *val_labs = var_get_value_labels (var);
919 int width = var_get_width (var);
920 const struct val_lab *val_lab;
922 if (val_labs_count (val_labs) == 0 || width < 9)
926 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
927 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
928 val_lab = val_labs_next (val_labs, val_lab))
931 size += recode_string_len (encoding, "UTF-8",
932 val_lab_get_escaped_label (val_lab), -1);
938 write_int (w, 7); /* Record type. */
939 write_int (w, 21); /* Record subtype */
940 write_int (w, 1); /* Data item (byte) size. */
941 write_int (w, size); /* Number of data items. */
943 start = ftello (w->file);
944 for (i = 0; i < n_vars; i++)
946 struct variable *var = dict_get_var (dict, i);
947 const struct val_labs *val_labs = var_get_value_labels (var);
948 int width = var_get_width (var);
949 const struct val_lab *val_lab;
952 if (val_labs_count (val_labs) == 0 || width < 9)
955 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
956 write_int (w, strlen (var_name));
957 write_bytes (w, var_name, strlen (var_name));
960 write_int (w, width);
961 write_int (w, val_labs_count (val_labs));
962 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
963 val_lab = val_labs_next (val_labs, val_lab))
968 write_int (w, width);
969 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
972 label = recode_string (var_get_encoding (var), "UTF-8",
973 val_lab_get_escaped_label (val_lab), -1);
974 len = strlen (label);
976 write_bytes (w, label, len);
980 assert (ftello (w->file) == start + size);
984 write_long_string_missing_values (struct sfm_writer *w,
985 const struct dictionary *dict)
987 const char *encoding = dict_get_encoding (dict);
988 size_t n_vars = dict_get_var_cnt (dict);
992 /* Figure out the size in advance. */
994 for (i = 0; i < n_vars; i++)
996 struct variable *var = dict_get_var (dict, i);
997 const struct missing_values *mv = var_get_missing_values (var);
998 int width = var_get_width (var);
1000 if (mv_is_empty (mv) || width < 9)
1004 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1006 size += mv_n_values (mv) * (4 + 8);
1011 write_int (w, 7); /* Record type. */
1012 write_int (w, 22); /* Record subtype */
1013 write_int (w, 1); /* Data item (byte) size. */
1014 write_int (w, size); /* Number of data items. */
1016 start = ftello (w->file);
1017 for (i = 0; i < n_vars; i++)
1019 struct variable *var = dict_get_var (dict, i);
1020 const struct missing_values *mv = var_get_missing_values (var);
1021 int width = var_get_width (var);
1022 uint8_t n_missing_values;
1026 if (mv_is_empty (mv) || width < 9)
1029 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1030 write_int (w, strlen (var_name));
1031 write_bytes (w, var_name, strlen (var_name));
1034 n_missing_values = mv_n_values (mv);
1035 write_bytes (w, &n_missing_values, 1);
1037 for (j = 0; j < n_missing_values; j++)
1039 const union value *value = mv_get_value (mv, j);
1042 write_bytes (w, value_str (value, width), 8);
1045 assert (ftello (w->file) == start + size);
1049 write_encoding_record (struct sfm_writer *w,
1050 const struct dictionary *d)
1052 /* IANA says "...character set names may be up to 40 characters taken
1053 from the printable characters of US-ASCII," so character set names
1054 don't need to be recoded to be in UTF-8.
1056 We convert encoding names to uppercase because SPSS writes encoding
1057 names in uppercase. */
1058 char *encoding = xstrdup (dict_get_encoding (d));
1059 str_uppercase (encoding);
1060 write_string_record (w, ss_cstr (encoding), 20);
1064 /* Writes the long variable name table. */
1066 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1071 ds_init_empty (&map);
1072 for (i = 0; i < dict_get_var_cnt (dict); i++)
1074 struct variable *v = dict_get_var (dict, i);
1076 ds_put_byte (&map, '\t');
1077 ds_put_format (&map, "%s=%s",
1078 var_get_short_name (v, 0), var_get_name (v));
1080 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1084 /* Write integer information record. */
1086 write_integer_info_record (struct sfm_writer *w,
1087 const struct dictionary *d)
1089 const char *dict_encoding = dict_get_encoding (d);
1090 int version_component[3];
1094 /* Parse the version string. */
1095 memset (version_component, 0, sizeof version_component);
1096 sscanf (bare_version, "%d.%d.%d",
1097 &version_component[0], &version_component[1], &version_component[2]);
1099 /* Figure out the floating-point format. */
1100 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1101 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1103 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1105 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1110 /* Choose codepage. */
1111 codepage = sys_get_codepage_from_encoding (dict_encoding);
1114 /* The codepage is unknown. Choose a default.
1116 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1118 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1119 many files use this codepage number regardless of their actual
1122 if (is_encoding_ascii_compatible (dict_encoding))
1124 else if (is_encoding_ebcdic_compatible (dict_encoding))
1129 write_int (w, 7); /* Record type. */
1130 write_int (w, 3); /* Record subtype. */
1131 write_int (w, 4); /* Data item (int32) size. */
1132 write_int (w, 8); /* Number of data items. */
1133 write_int (w, version_component[0]);
1134 write_int (w, version_component[1]);
1135 write_int (w, version_component[2]);
1136 write_int (w, -1); /* Machine code. */
1137 write_int (w, float_format);
1138 write_int (w, 1); /* Compression code. */
1139 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1140 write_int (w, codepage);
1143 /* Write floating-point information record. */
1145 write_float_info_record (struct sfm_writer *w)
1147 write_int (w, 7); /* Record type. */
1148 write_int (w, 4); /* Record subtype. */
1149 write_int (w, 8); /* Data item (flt64) size. */
1150 write_int (w, 3); /* Number of data items. */
1151 write_float (w, SYSMIS); /* System-missing value. */
1152 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1153 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1156 /* Writes case C to system file W. */
1158 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1161 struct sfm_writer *w = w_;
1163 if (ferror (w->file))
1165 casewriter_force_error (writer);
1173 write_case_uncompressed (w, c);
1175 write_case_compressed (w, c);
1180 /* Destroys system file writer W. */
1182 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1184 struct sfm_writer *w = w_;
1185 if (!close_writer (w))
1186 casewriter_force_error (writer);
1189 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1191 write_error (const struct sfm_writer *writer)
1193 return ferror (writer->file);
1196 /* Closes a system file after we're done with it.
1197 Returns true if successful, false if an I/O error occurred. */
1199 close_writer (struct sfm_writer *w)
1207 if (w->file != NULL)
1210 if (w->opcode_cnt > 0)
1211 flush_compressed (w);
1214 ok = !write_error (w);
1216 /* Seek back to the beginning and update the number of cases.
1217 This is just a courtesy to later readers, so there's no need
1218 to check return values or report errors. */
1219 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1221 write_int (w, w->case_cnt);
1225 if (fclose (w->file) == EOF)
1229 msg (ME, _("An I/O error occurred writing system file `%s'."),
1230 fh_get_file_name (w->fh));
1232 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1236 fh_unlock (w->lock);
1245 /* System file writer casewriter class. */
1246 static const struct casewriter_class sys_file_casewriter_class =
1248 sys_file_casewriter_write,
1249 sys_file_casewriter_destroy,
1253 /* Writes case C to system file W, without compressing it. */
1255 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1259 for (i = 0; i < w->sfm_var_cnt; i++)
1261 struct sfm_var *v = &w->sfm_vars[i];
1263 if (v->var_width == 0)
1264 write_float (w, case_num_idx (c, v->case_index));
1267 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1269 write_spaces (w, v->padding);
1274 /* Writes case C to system file W, with compression. */
1276 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1280 for (i = 0; i < w->sfm_var_cnt; i++)
1282 struct sfm_var *v = &w->sfm_vars[i];
1284 if (v->var_width == 0)
1286 double d = case_num_idx (c, v->case_index);
1288 put_cmp_opcode (w, 255);
1289 else if (d >= 1 - COMPRESSION_BIAS
1290 && d <= 251 - COMPRESSION_BIAS
1292 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1295 put_cmp_opcode (w, 253);
1296 put_cmp_number (w, d);
1301 int offset = v->offset;
1304 /* This code properly deals with a width that is not a
1305 multiple of 8, by ensuring that the final partial
1306 oct (8 byte unit) is treated as padded with spaces
1308 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1310 const void *data = case_str_idx (c, v->case_index) + offset;
1311 int chunk_size = MIN (width, 8);
1312 if (!memcmp (data, " ", chunk_size))
1313 put_cmp_opcode (w, 254);
1316 put_cmp_opcode (w, 253);
1317 put_cmp_string (w, data, chunk_size);
1321 /* This code deals properly with padding that is not a
1322 multiple of 8 bytes, by discarding the remainder,
1323 which was already effectively padded with spaces in
1324 the previous loop. (Note that v->width + v->padding
1325 is always a multiple of 8.) */
1326 for (padding = v->padding / 8; padding > 0; padding--)
1327 put_cmp_opcode (w, 254);
1332 /* Flushes buffered compressed opcodes and data to W.
1333 The compression buffer must not be empty. */
1335 flush_compressed (struct sfm_writer *w)
1337 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1339 write_bytes (w, w->opcodes, w->opcode_cnt);
1340 write_zeros (w, 8 - w->opcode_cnt);
1342 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1344 w->opcode_cnt = w->data_cnt = 0;
1347 /* Appends OPCODE to the buffered set of compression opcodes in
1348 W. Flushes the compression buffer beforehand if necessary. */
1350 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1352 if (w->opcode_cnt >= 8)
1353 flush_compressed (w);
1355 w->opcodes[w->opcode_cnt++] = opcode;
1358 /* Appends NUMBER to the buffered compression data in W. The
1359 buffer must not be full; the way to assure that is to call
1360 this function only just after a call to put_cmp_opcode, which
1361 will flush the buffer as necessary. */
1363 put_cmp_number (struct sfm_writer *w, double number)
1365 assert (w->opcode_cnt > 0);
1366 assert (w->data_cnt < 8);
1368 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1371 /* Appends SIZE bytes of DATA to the buffered compression data in
1372 W, followed by enough spaces to pad the output data to exactly
1373 8 bytes (thus, SIZE must be no greater than 8). The buffer
1374 must not be full; the way to assure that is to call this
1375 function only just after a call to put_cmp_opcode, which will
1376 flush the buffer as necessary. */
1378 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1380 assert (w->opcode_cnt > 0);
1381 assert (w->data_cnt < 8);
1384 memset (w->data[w->data_cnt], w->space, 8);
1385 memcpy (w->data[w->data_cnt], data, size);
1389 /* Writes 32-bit integer X to the output file for writer W. */
1391 write_int (struct sfm_writer *w, int32_t x)
1393 write_bytes (w, &x, sizeof x);
1396 /* Converts NATIVE to the 64-bit format used in output files in
1399 convert_double_to_output_format (double native, uint8_t output[8])
1401 /* If "double" is not a 64-bit type, then convert it to a
1402 64-bit type. Otherwise just copy it. */
1403 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1404 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1406 memcpy (output, &native, sizeof native);
1409 /* Writes floating-point number X to the output file for writer
1412 write_float (struct sfm_writer *w, double x)
1415 convert_double_to_output_format (x, output);
1416 write_bytes (w, output, sizeof output);
1419 /* Writes contents of VALUE with the given WIDTH to W, padding
1420 with zeros to a multiple of 8 bytes.
1421 To avoid a branch, and because we don't actually need to
1422 support it, WIDTH must be no bigger than 8. */
1424 write_value (struct sfm_writer *w, const union value *value, int width)
1426 assert (width <= 8);
1428 write_float (w, value->f);
1431 write_bytes (w, value_str (value, width), width);
1432 write_zeros (w, 8 - width);
1436 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1437 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1438 is padded on the right with spaces. */
1440 write_string (struct sfm_writer *w, const char *string, size_t width)
1442 size_t data_bytes = MIN (strlen (string), width);
1443 size_t pad_bytes = width - data_bytes;
1444 write_bytes (w, string, data_bytes);
1445 while (pad_bytes-- > 0)
1446 putc (w->space, w->file);
1449 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1450 recoded version in a field of the given WIDTH to W. The string is truncated
1451 or padded on the right with spaces to exactly WIDTH bytes. */
1453 write_utf8_string (struct sfm_writer *w, const char *encoding,
1454 const char *string, size_t width)
1456 char *s = recode_string (encoding, "UTF-8", string, -1);
1457 write_string (w, s, width);
1461 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1462 from UTF-8 encoded into ENCODING. */
1464 write_utf8_record (struct sfm_writer *w, const char *encoding,
1465 const struct string *content, int subtype)
1469 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1470 write_string_record (w, s, subtype);
1474 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1477 write_string_record (struct sfm_writer *w,
1478 const struct substring content, int subtype)
1481 write_int (w, subtype);
1483 write_int (w, ss_length (content));
1484 write_bytes (w, ss_data (content), ss_length (content));
1487 /* Writes SIZE bytes of DATA to W's output file. */
1489 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1491 fwrite (data, 1, size, w->file);
1494 /* Writes N zeros to W's output file. */
1496 write_zeros (struct sfm_writer *w, size_t n)
1502 /* Writes N spaces to W's output file. */
1504 write_spaces (struct sfm_writer *w, size_t n)
1507 putc (w->space, w->file);