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"
30 #include "data/attributes.h"
31 #include "data/case.h"
32 #include "data/casewriter-provider.h"
33 #include "data/casewriter.h"
34 #include "data/dictionary.h"
35 #include "data/file-handle-def.h"
36 #include "data/file-name.h"
37 #include "data/format.h"
38 #include "data/make-file.h"
39 #include "data/missing-values.h"
40 #include "data/mrset.h"
41 #include "data/settings.h"
42 #include "data/short-names.h"
43 #include "data/value-labels.h"
44 #include "data/variable.h"
45 #include "libpspp/float-format.h"
46 #include "libpspp/i18n.h"
47 #include "libpspp/integer-format.h"
48 #include "libpspp/message.h"
49 #include "libpspp/misc.h"
50 #include "libpspp/str.h"
51 #include "libpspp/string-array.h"
52 #include "libpspp/version.h"
54 #include "gl/xmemdup0.h"
55 #include "gl/minmax.h"
56 #include "gl/unlocked-io.h"
57 #include "gl/xalloc.h"
60 #define _(msgid) gettext (msgid)
61 #define N_(msgid) (msgid)
63 /* Compression bias used by PSPP. Values between (1 -
64 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
66 #define COMPRESSION_BIAS 100
68 /* System file writer. */
71 struct file_handle *fh; /* File handle. */
72 struct fh_lock *lock; /* Mutual exclusion for file. */
73 FILE *file; /* File stream. */
74 struct replace_file *rf; /* Ticket for replacing output file. */
76 enum sfm_compression compression;
77 casenumber case_cnt; /* Number of cases written so far. */
78 uint8_t space; /* ' ' in the file's character encoding. */
80 /* Simple compression buffering.
82 Compressed data is output as a series of 8-byte elements, with 1 to 9
83 such elements clustered together. The first element in a cluster is 8
84 1-byte opcodes. Some opcodes call for an additional element in the
85 cluster (hence, if there are eight such opcodes, then the cluster
86 contains a full 9 elements).
88 cbuf[] holds a cluster at a time. */
90 int n_opcodes; /* Number of opcodes in cbuf[0] so far. */
91 int n_elements; /* Number of elements in cbuf[] so far. */
93 /* ZLIB compression. */
94 z_stream zstream; /* ZLIB deflater. */
96 struct zblock *blocks;
97 size_t n_blocks, allocated_blocks;
100 struct sfm_var *sfm_vars; /* Variables. */
101 size_t sfm_var_cnt; /* Number of variables. */
102 size_t segment_cnt; /* Number of variables including extra segments
103 for long string variables. */
108 unsigned int uncompressed_size;
109 unsigned int compressed_size;
112 static const struct casewriter_class sys_file_casewriter_class;
114 static void write_header (struct sfm_writer *, const struct dictionary *);
115 static void write_variable (struct sfm_writer *, const struct variable *);
116 static void write_value_labels (struct sfm_writer *,
117 const struct dictionary *);
118 static void write_integer_info_record (struct sfm_writer *,
119 const struct dictionary *);
120 static void write_float_info_record (struct sfm_writer *);
122 static void write_longvar_table (struct sfm_writer *w,
123 const struct dictionary *dict);
125 static void write_encoding_record (struct sfm_writer *w,
126 const struct dictionary *);
128 static void write_vls_length_table (struct sfm_writer *w,
129 const struct dictionary *dict);
131 static void write_long_string_value_labels (struct sfm_writer *,
132 const struct dictionary *);
133 static void write_long_string_missing_values (struct sfm_writer *,
134 const struct dictionary *);
136 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
139 static void write_variable_display_parameters (struct sfm_writer *w,
140 const struct dictionary *dict);
142 static void write_documents (struct sfm_writer *, const struct dictionary *);
144 static void write_data_file_attributes (struct sfm_writer *,
145 const struct dictionary *);
146 static void write_variable_attributes (struct sfm_writer *,
147 const struct dictionary *);
149 static void write_int (struct sfm_writer *, int32_t);
150 static void write_int64 (struct sfm_writer *, int64_t);
151 static inline void convert_double_to_output_format (double, uint8_t[8]);
152 static void write_float (struct sfm_writer *, double);
153 static void write_string (struct sfm_writer *, const char *, size_t);
154 static void write_utf8_string (struct sfm_writer *, const char *encoding,
155 const char *string, size_t width);
156 static void write_utf8_record (struct sfm_writer *, const char *encoding,
157 const struct string *content, int subtype);
158 static void write_string_record (struct sfm_writer *,
159 const struct substring content, int subtype);
160 static void write_bytes (struct sfm_writer *, const void *, size_t);
161 static void write_zeros (struct sfm_writer *, size_t);
162 static void write_spaces (struct sfm_writer *, size_t);
163 static void write_value (struct sfm_writer *, const union value *, int width);
165 static void write_case_uncompressed (struct sfm_writer *,
166 const struct ccase *);
167 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
168 static void flush_compressed (struct sfm_writer *);
169 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
170 static void put_cmp_number (struct sfm_writer *, double);
171 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
173 static bool start_zstream (struct sfm_writer *);
174 static void finish_zstream (struct sfm_writer *);
175 static void write_ztrailer (struct sfm_writer *);
177 static bool write_error (const struct sfm_writer *);
178 static bool close_writer (struct sfm_writer *);
180 /* Returns default options for writing a system file. */
181 struct sfm_write_options
182 sfm_writer_default_options (void)
184 struct sfm_write_options opts;
185 opts.compression = (settings_get_scompression ()
188 opts.create_writeable = true;
193 /* Opens the system file designated by file handle FH for writing
194 cases from dictionary D according to the given OPTS.
196 No reference to D is retained, so it may be modified or
197 destroyed at will after this function returns. D is not
198 modified by this function, except to assign short names. */
200 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
201 struct sfm_write_options opts)
203 struct encoding_info encoding_info;
204 struct sfm_writer *w;
209 if (opts.version != 2 && opts.version != 3)
211 msg (ME, _("Unknown system file version %d. Treating as version %d."),
216 /* Create and initialize writer. */
217 w = xzalloc (sizeof *w);
223 /* Use the requested compression, except that no EBCDIC-based ZLIB compressed
224 files have been observed, so drop back to simple compression for those
226 w->compression = opts.compression;
227 if (w->compression == SFM_COMP_ZLIB
228 && is_encoding_ebcdic_compatible (dict_get_encoding (d)))
229 w->compression = SFM_COMP_SIMPLE;
233 w->n_opcodes = w->n_elements = 0;
234 memset (w->cbuf[0], 0, 8);
236 /* Figure out how to map in-memory case data to on-disk case
237 data. Also count the number of segments. Very long strings
238 occupy multiple segments, otherwise each variable only takes
240 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
243 /* Open file handle as an exclusive writer. */
244 /* TRANSLATORS: this fragment will be interpolated into
245 messages in fh_lock() that identify types of files. */
246 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
250 /* Create the file on disk. */
252 if (opts.create_writeable)
254 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
258 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
259 fh_get_file_name (fh), strerror (errno));
263 get_encoding_info (&encoding_info, dict_get_encoding (d));
264 w->space = encoding_info.space[0];
266 /* Write the file header. */
269 /* Write basic variable info. */
270 short_names_assign (d);
271 for (i = 0; i < dict_get_var_cnt (d); i++)
272 write_variable (w, dict_get_var (d, i));
274 write_value_labels (w, d);
276 if (dict_get_document_line_cnt (d) > 0)
277 write_documents (w, d);
279 write_integer_info_record (w, d);
280 write_float_info_record (w);
282 write_mrsets (w, d, true);
284 write_variable_display_parameters (w, d);
286 if (opts.version >= 3)
287 write_longvar_table (w, d);
289 write_vls_length_table (w, d);
291 write_long_string_value_labels (w, d);
292 write_long_string_missing_values (w, d);
294 if (opts.version >= 3)
296 if (attrset_count (dict_get_attributes (d)))
297 write_data_file_attributes (w, d);
298 write_variable_attributes (w, d);
301 write_mrsets (w, d, false);
303 write_encoding_record (w, d);
305 /* Write end-of-headers record. */
309 if (w->compression == SFM_COMP_ZLIB)
311 w->zstream.zalloc = Z_NULL;
312 w->zstream.zfree = Z_NULL;
313 w->zstream.opaque = Z_NULL;
314 w->zstart = ftello (w->file);
316 write_int64 (w, w->zstart);
326 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
333 /* Returns value of X truncated to two least-significant digits. */
344 /* Calculates the offset of data for TARGET_VAR from the
345 beginning of each case's data for dictionary D. The return
346 value is in "octs" (8-byte units). */
348 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
354 for (i = 0; i < dict_get_var_cnt (d); i++)
356 struct variable *var = dict_get_var (d, i);
357 if (var == target_var)
359 oct_idx += sfm_width_to_octs (var_get_width (var));
364 /* Write the sysfile_header header to system file W. */
366 write_header (struct sfm_writer *w, const struct dictionary *d)
368 const char *dict_encoding = dict_get_encoding (d);
370 char creation_date[10];
371 char creation_time[9];
372 const char *file_label;
373 struct variable *weight;
377 /* Record-type code. */
378 if (is_encoding_ebcdic_compatible (dict_encoding))
379 write_string (w, EBCDIC_MAGIC, 4);
380 else if (w->compression == SFM_COMP_ZLIB)
381 write_string (w, ASCII_ZMAGIC, 4);
383 write_string (w, ASCII_MAGIC, 4);
385 /* Product identification. */
386 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
387 version, host_system);
388 write_utf8_string (w, dict_encoding, prod_name, 60);
393 /* Number of `union value's per case. */
394 write_int (w, calc_oct_idx (d, NULL));
397 write_int (w, (w->compression == SFM_COMP_NONE ? 0
398 : w->compression == SFM_COMP_SIMPLE ? 1
401 /* Weight variable. */
402 weight = dict_get_weight (d);
403 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
405 /* Number of cases. We don't know this in advance, so we write
406 -1 to indicate an unknown number of cases. Later we can
407 come back and overwrite it with the true value. */
410 /* Compression bias. */
411 write_float (w, COMPRESSION_BIAS);
413 /* Creation date and time. */
414 if (time (&t) == (time_t) -1)
416 strcpy (creation_date, "01 Jan 70");
417 strcpy (creation_time, "00:00:00");
421 static const char *const month_name[12] =
423 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
424 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
426 struct tm *tmp = localtime (&t);
427 int day = rerange (tmp->tm_mday);
428 int mon = rerange (tmp->tm_mon + 1);
429 int year = rerange (tmp->tm_year);
430 int hour = rerange (tmp->tm_hour + 1);
431 int min = rerange (tmp->tm_min + 1);
432 int sec = rerange (tmp->tm_sec + 1);
434 snprintf (creation_date, sizeof creation_date,
435 "%02d %s %02d", day, month_name[mon - 1], year);
436 snprintf (creation_time, sizeof creation_time,
437 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
439 write_utf8_string (w, dict_encoding, creation_date, 9);
440 write_utf8_string (w, dict_encoding, creation_time, 8);
443 file_label = dict_get_label (d);
444 if (file_label == NULL)
446 write_utf8_string (w, dict_encoding, file_label, 64);
452 /* Write format spec FMT to W, after adjusting it to be
453 compatible with the given WIDTH. */
455 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
457 assert (fmt_check_output (&fmt));
458 assert (sfm_width_to_segments (width) == 1);
461 fmt_resize (&fmt, width);
462 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
465 /* Write a string continuation variable record for each 8-byte
466 section beyond the initial 8 bytes, for a variable of the
469 write_variable_continuation_records (struct sfm_writer *w, int width)
473 assert (sfm_width_to_segments (width) == 1);
474 for (position = 8; position < width; position += 8)
476 write_int (w, 2); /* Record type. */
477 write_int (w, -1); /* Width. */
478 write_int (w, 0); /* No variable label. */
479 write_int (w, 0); /* No missing values. */
480 write_int (w, 0); /* Print format. */
481 write_int (w, 0); /* Write format. */
482 write_zeros (w, 8); /* Name. */
486 /* Write the variable record(s) for variable V to system file
489 write_variable (struct sfm_writer *w, const struct variable *v)
491 int width = var_get_width (v);
492 int segment_cnt = sfm_width_to_segments (width);
493 int seg0_width = sfm_segment_alloc_width (width, 0);
494 const char *encoding = var_get_encoding (v);
501 write_int (w, seg0_width);
503 /* Variable has a variable label? */
504 write_int (w, var_has_label (v));
506 /* Number of missing values. If there is a range, then the
507 range counts as 2 missing values and causes the number to be
510 Missing values for long string variables are written in a separate
512 if (width <= MAX_SHORT_STRING)
514 const struct missing_values *mv = var_get_missing_values (v);
515 if (mv_has_range (mv))
516 write_int (w, -2 - mv_n_values (mv));
518 write_int (w, mv_n_values (mv));
523 /* Print and write formats. */
524 write_format (w, *var_get_print_format (v), seg0_width);
525 write_format (w, *var_get_write_format (v), seg0_width);
528 The full name is in a translation table written
530 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
533 if (var_has_label (v))
535 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
536 size_t label_len = MIN (strlen (label), 255);
537 size_t padded_len = ROUND_UP (label_len, 4);
538 write_int (w, label_len);
539 write_string (w, label, padded_len);
543 /* Write the missing values, if any, range first. */
544 if (width <= MAX_SHORT_STRING)
546 const struct missing_values *mv = var_get_missing_values (v);
547 if (mv_has_range (mv))
550 mv_get_range (mv, &x, &y);
554 for (i = 0; i < mv_n_values (mv); i++)
555 write_value (w, mv_get_value (mv, i), width);
558 write_variable_continuation_records (w, seg0_width);
560 /* Write additional segments for very long string variables. */
561 for (i = 1; i < segment_cnt; i++)
563 int seg_width = sfm_segment_alloc_width (width, i);
564 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
566 write_int (w, 2); /* Variable record. */
567 write_int (w, seg_width); /* Width. */
568 write_int (w, 0); /* No variable label. */
569 write_int (w, 0); /* No missing values. */
570 write_format (w, fmt, seg_width); /* Print format. */
571 write_format (w, fmt, seg_width); /* Write format. */
572 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
574 write_variable_continuation_records (w, seg_width);
578 /* Writes the value labels to system file W.
580 Value labels for long string variables are written separately,
581 by write_long_string_value_labels. */
583 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
587 struct hmap_node hmap_node;
588 const struct val_labs *val_labs;
590 size_t n_indexes, allocated_indexes;
593 size_t n_sets, allocated_sets;
594 struct label_set **sets;
595 struct hmap same_sets;
599 n_sets = allocated_sets = 0;
601 hmap_init (&same_sets);
604 for (i = 0; i < dict_get_var_cnt (d); i++)
606 struct variable *v = dict_get_var (d, i);
608 if (var_has_value_labels (v) && var_get_width (v) <= 8)
610 const struct val_labs *val_labs = var_get_value_labels (v);
611 unsigned int hash = val_labs_hash (val_labs, 0);
612 struct label_set *set;
614 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
617 if (val_labs_equal (set->val_labs, val_labs))
619 if (set->n_indexes >= set->allocated_indexes)
620 set->indexes = x2nrealloc (set->indexes,
621 &set->allocated_indexes,
622 sizeof *set->indexes);
623 set->indexes[set->n_indexes++] = idx;
628 set = xmalloc (sizeof *set);
629 set->val_labs = val_labs;
630 set->indexes = xmalloc (sizeof *set->indexes);
631 set->indexes[0] = idx;
633 set->allocated_indexes = 1;
634 hmap_insert (&same_sets, &set->hmap_node, hash);
636 if (n_sets >= allocated_sets)
637 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
638 sets[n_sets++] = set;
642 idx += sfm_width_to_octs (var_get_width (v));
645 for (i = 0; i < n_sets; i++)
647 const struct label_set *set = sets[i];
648 const struct val_labs *val_labs = set->val_labs;
649 size_t n_labels = val_labs_count (val_labs);
650 int width = val_labs_get_width (val_labs);
651 const struct val_lab **labels;
654 /* Value label record. */
655 write_int (w, 3); /* Record type. */
656 write_int (w, n_labels);
657 labels = val_labs_sorted (val_labs);
658 for (j = 0; j < n_labels; j++)
660 const struct val_lab *vl = labels[j];
661 char *label = recode_string (dict_get_encoding (d), UTF8,
662 val_lab_get_escaped_label (vl), -1);
663 uint8_t len = MIN (strlen (label), 255);
665 write_value (w, val_lab_get_value (vl), width);
666 write_bytes (w, &len, 1);
667 write_bytes (w, label, len);
668 write_zeros (w, REM_RND_UP (len + 1, 8));
673 /* Value label variable record. */
674 write_int (w, 4); /* Record type. */
675 write_int (w, set->n_indexes);
676 for (j = 0; j < set->n_indexes; j++)
677 write_int (w, set->indexes[j] + 1);
680 for (i = 0; i < n_sets; i++)
682 struct label_set *set = sets[i];
688 hmap_destroy (&same_sets);
691 /* Writes record type 6, document record. */
693 write_documents (struct sfm_writer *w, const struct dictionary *d)
695 const struct string_array *docs = dict_get_documents (d);
696 const char *enc = dict_get_encoding (d);
699 write_int (w, 6); /* Record type. */
700 write_int (w, docs->n);
701 for (i = 0; i < docs->n; i++)
703 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
704 size_t s_len = strlen (s);
705 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
707 write_bytes (w, s, write_len);
708 write_spaces (w, DOC_LINE_LENGTH - write_len);
714 put_attrset (struct string *string, const struct attrset *attrs)
716 const struct attribute *attr;
717 struct attrset_iterator i;
719 for (attr = attrset_first (attrs, &i); attr != NULL;
720 attr = attrset_next (attrs, &i))
722 size_t n_values = attribute_get_n_values (attr);
725 ds_put_cstr (string, attribute_get_name (attr));
726 ds_put_byte (string, '(');
727 for (j = 0; j < n_values; j++)
728 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
729 ds_put_byte (string, ')');
734 write_data_file_attributes (struct sfm_writer *w,
735 const struct dictionary *d)
737 struct string s = DS_EMPTY_INITIALIZER;
738 put_attrset (&s, dict_get_attributes (d));
739 write_utf8_record (w, dict_get_encoding (d), &s, 17);
744 add_role_attribute (enum var_role role, struct attrset *attrs)
746 struct attribute *attr;
776 attrset_delete (attrs, "$@Role");
778 attr = attribute_create ("$@Role");
779 attribute_add_value (attr, s);
780 attrset_add (attrs, attr);
784 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
786 struct string s = DS_EMPTY_INITIALIZER;
787 size_t n_vars = dict_get_var_cnt (d);
788 size_t n_attrsets = 0;
791 for (i = 0; i < n_vars; i++)
793 struct variable *v = dict_get_var (d, i);
794 struct attrset attrs;
796 attrset_clone (&attrs, var_get_attributes (v));
798 add_role_attribute (var_get_role (v), &attrs);
800 ds_put_byte (&s, '/');
801 ds_put_format (&s, "%s:", var_get_name (v));
802 put_attrset (&s, &attrs);
803 attrset_destroy (&attrs);
806 write_utf8_record (w, dict_get_encoding (d), &s, 18);
810 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
811 SPSS before release 14, otherwise writes sets supported only by later
814 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
817 const char *encoding = dict_get_encoding (dict);
818 struct string s = DS_EMPTY_INITIALIZER;
822 if (is_encoding_ebcdic_compatible (encoding))
828 n_mrsets = dict_get_n_mrsets (dict);
832 for (i = 0; i < n_mrsets; i++)
834 const struct mrset *mrset = dict_get_mrset (dict, i);
838 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
842 name = recode_string (encoding, "UTF-8", mrset->name, -1);
843 ds_put_format (&s, "%s=", name);
846 if (mrset->type == MRSET_MD)
850 if (mrset->cat_source == MRSET_COUNTEDVALUES)
851 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
853 ds_put_byte (&s, 'D');
855 if (mrset->width == 0)
856 counted = xasprintf ("%.0f", mrset->counted.f);
858 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
860 ds_put_format (&s, "%zu %s", strlen (counted), counted);
864 ds_put_byte (&s, 'C');
865 ds_put_byte (&s, ' ');
867 if (mrset->label && !mrset->label_from_var_label)
869 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
870 ds_put_format (&s, "%zu %s", strlen (label), label);
874 ds_put_cstr (&s, "0 ");
876 for (j = 0; j < mrset->n_vars; j++)
878 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
879 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
880 char *short_name = recode_string (encoding, "UTF-8",
881 lower_name_utf8, -1);
882 ds_put_format (&s, " %s", short_name);
884 free (lower_name_utf8);
886 ds_put_byte (&s, '\n');
889 if (!ds_is_empty (&s))
890 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
894 /* Write the alignment, width and scale values. */
896 write_variable_display_parameters (struct sfm_writer *w,
897 const struct dictionary *dict)
901 write_int (w, 7); /* Record type. */
902 write_int (w, 11); /* Record subtype. */
903 write_int (w, 4); /* Data item (int32) size. */
904 write_int (w, w->segment_cnt * 3); /* Number of data items. */
906 for (i = 0; i < dict_get_var_cnt (dict); ++i)
908 struct variable *v = dict_get_var (dict, i);
909 int width = var_get_width (v);
910 int segment_cnt = sfm_width_to_segments (width);
911 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
912 : var_get_measure (v) == MEASURE_ORDINAL ? 2
914 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
915 : var_get_alignment (v) == ALIGN_RIGHT ? 1
919 for (i = 0; i < segment_cnt; i++)
921 int width_left = width - sfm_segment_effective_offset (width, i);
922 write_int (w, measure);
923 write_int (w, (i == 0 ? var_get_display_width (v)
924 : var_default_display_width (width_left)));
925 write_int (w, alignment);
930 /* Writes the table of lengths for very long string variables. */
932 write_vls_length_table (struct sfm_writer *w,
933 const struct dictionary *dict)
938 ds_init_empty (&map);
939 for (i = 0; i < dict_get_var_cnt (dict); ++i)
941 const struct variable *v = dict_get_var (dict, i);
942 if (sfm_width_to_segments (var_get_width (v)) > 1)
943 ds_put_format (&map, "%s=%05d%c\t",
944 var_get_short_name (v, 0), var_get_width (v), 0);
946 if (!ds_is_empty (&map))
947 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
952 write_long_string_value_labels (struct sfm_writer *w,
953 const struct dictionary *dict)
955 const char *encoding = dict_get_encoding (dict);
956 size_t n_vars = dict_get_var_cnt (dict);
960 /* Figure out the size in advance. */
962 for (i = 0; i < n_vars; i++)
964 struct variable *var = dict_get_var (dict, i);
965 const struct val_labs *val_labs = var_get_value_labels (var);
966 int width = var_get_width (var);
967 const struct val_lab *val_lab;
969 if (val_labs_count (val_labs) == 0 || width < 9)
973 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
974 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
975 val_lab = val_labs_next (val_labs, val_lab))
978 size += recode_string_len (encoding, "UTF-8",
979 val_lab_get_escaped_label (val_lab), -1);
985 write_int (w, 7); /* Record type. */
986 write_int (w, 21); /* Record subtype */
987 write_int (w, 1); /* Data item (byte) size. */
988 write_int (w, size); /* Number of data items. */
990 start = ftello (w->file);
991 for (i = 0; i < n_vars; i++)
993 struct variable *var = dict_get_var (dict, i);
994 const struct val_labs *val_labs = var_get_value_labels (var);
995 int width = var_get_width (var);
996 const struct val_lab *val_lab;
999 if (val_labs_count (val_labs) == 0 || width < 9)
1002 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1003 write_int (w, strlen (var_name));
1004 write_bytes (w, var_name, strlen (var_name));
1007 write_int (w, width);
1008 write_int (w, val_labs_count (val_labs));
1009 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
1010 val_lab = val_labs_next (val_labs, val_lab))
1015 write_int (w, width);
1016 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
1019 label = recode_string (var_get_encoding (var), "UTF-8",
1020 val_lab_get_escaped_label (val_lab), -1);
1021 len = strlen (label);
1023 write_bytes (w, label, len);
1027 assert (ftello (w->file) == start + size);
1031 write_long_string_missing_values (struct sfm_writer *w,
1032 const struct dictionary *dict)
1034 const char *encoding = dict_get_encoding (dict);
1035 size_t n_vars = dict_get_var_cnt (dict);
1039 /* Figure out the size in advance. */
1041 for (i = 0; i < n_vars; i++)
1043 struct variable *var = dict_get_var (dict, i);
1044 const struct missing_values *mv = var_get_missing_values (var);
1045 int width = var_get_width (var);
1047 if (mv_is_empty (mv) || width < 9)
1051 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1053 size += mv_n_values (mv) * (4 + 8);
1058 write_int (w, 7); /* Record type. */
1059 write_int (w, 22); /* Record subtype */
1060 write_int (w, 1); /* Data item (byte) size. */
1061 write_int (w, size); /* Number of data items. */
1063 start = ftello (w->file);
1064 for (i = 0; i < n_vars; i++)
1066 struct variable *var = dict_get_var (dict, i);
1067 const struct missing_values *mv = var_get_missing_values (var);
1068 int width = var_get_width (var);
1069 uint8_t n_missing_values;
1073 if (mv_is_empty (mv) || width < 9)
1076 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1077 write_int (w, strlen (var_name));
1078 write_bytes (w, var_name, strlen (var_name));
1081 n_missing_values = mv_n_values (mv);
1082 write_bytes (w, &n_missing_values, 1);
1084 for (j = 0; j < n_missing_values; j++)
1086 const union value *value = mv_get_value (mv, j);
1089 write_bytes (w, value_str (value, width), 8);
1092 assert (ftello (w->file) == start + size);
1096 write_encoding_record (struct sfm_writer *w,
1097 const struct dictionary *d)
1099 /* IANA says "...character set names may be up to 40 characters taken
1100 from the printable characters of US-ASCII," so character set names
1101 don't need to be recoded to be in UTF-8.
1103 We convert encoding names to uppercase because SPSS writes encoding
1104 names in uppercase. */
1105 char *encoding = xstrdup (dict_get_encoding (d));
1106 str_uppercase (encoding);
1107 write_string_record (w, ss_cstr (encoding), 20);
1111 /* Writes the long variable name table. */
1113 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1118 ds_init_empty (&map);
1119 for (i = 0; i < dict_get_var_cnt (dict); i++)
1121 struct variable *v = dict_get_var (dict, i);
1123 ds_put_byte (&map, '\t');
1124 ds_put_format (&map, "%s=%s",
1125 var_get_short_name (v, 0), var_get_name (v));
1127 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1131 /* Write integer information record. */
1133 write_integer_info_record (struct sfm_writer *w,
1134 const struct dictionary *d)
1136 const char *dict_encoding = dict_get_encoding (d);
1137 int version_component[3];
1141 /* Parse the version string. */
1142 memset (version_component, 0, sizeof version_component);
1143 sscanf (bare_version, "%d.%d.%d",
1144 &version_component[0], &version_component[1], &version_component[2]);
1146 /* Figure out the floating-point format. */
1147 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1148 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1150 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1152 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1157 /* Choose codepage. */
1158 codepage = sys_get_codepage_from_encoding (dict_encoding);
1161 /* The codepage is unknown. Choose a default.
1163 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1165 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1166 many files use this codepage number regardless of their actual
1169 if (is_encoding_ascii_compatible (dict_encoding))
1171 else if (is_encoding_ebcdic_compatible (dict_encoding))
1176 write_int (w, 7); /* Record type. */
1177 write_int (w, 3); /* Record subtype. */
1178 write_int (w, 4); /* Data item (int32) size. */
1179 write_int (w, 8); /* Number of data items. */
1180 write_int (w, version_component[0]);
1181 write_int (w, version_component[1]);
1182 write_int (w, version_component[2]);
1183 write_int (w, -1); /* Machine code. */
1184 write_int (w, float_format);
1185 write_int (w, 1); /* Compression code. */
1186 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1187 write_int (w, codepage);
1190 /* Write floating-point information record. */
1192 write_float_info_record (struct sfm_writer *w)
1194 write_int (w, 7); /* Record type. */
1195 write_int (w, 4); /* Record subtype. */
1196 write_int (w, 8); /* Data item (flt64) size. */
1197 write_int (w, 3); /* Number of data items. */
1198 write_float (w, SYSMIS); /* System-missing value. */
1199 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1200 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1203 /* Writes case C to system file W. */
1205 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1208 struct sfm_writer *w = w_;
1210 if (ferror (w->file))
1212 casewriter_force_error (writer);
1219 if (w->compression == SFM_COMP_NONE)
1220 write_case_uncompressed (w, c);
1222 write_case_compressed (w, c);
1227 /* Destroys system file writer W. */
1229 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1231 struct sfm_writer *w = w_;
1232 if (!close_writer (w))
1233 casewriter_force_error (writer);
1236 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1238 write_error (const struct sfm_writer *writer)
1240 return ferror (writer->file);
1243 /* Closes a system file after we're done with it.
1244 Returns true if successful, false if an I/O error occurred. */
1246 close_writer (struct sfm_writer *w)
1254 if (w->file != NULL)
1257 flush_compressed (w);
1258 if (w->compression == SFM_COMP_ZLIB)
1265 ok = !write_error (w);
1267 /* Seek back to the beginning and update the number of cases.
1268 This is just a courtesy to later readers, so there's no need
1269 to check return values or report errors. */
1270 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1272 write_int (w, w->case_cnt);
1276 if (fclose (w->file) == EOF)
1280 msg (ME, _("An I/O error occurred writing system file `%s'."),
1281 fh_get_file_name (w->fh));
1283 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1289 fh_unlock (w->lock);
1298 /* System file writer casewriter class. */
1299 static const struct casewriter_class sys_file_casewriter_class =
1301 sys_file_casewriter_write,
1302 sys_file_casewriter_destroy,
1306 /* Writes case C to system file W, without compressing it. */
1308 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1312 for (i = 0; i < w->sfm_var_cnt; i++)
1314 struct sfm_var *v = &w->sfm_vars[i];
1316 if (v->var_width == 0)
1317 write_float (w, case_num_idx (c, v->case_index));
1320 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1322 write_spaces (w, v->padding);
1327 /* Writes case C to system file W, with compression. */
1329 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1333 for (i = 0; i < w->sfm_var_cnt; i++)
1335 struct sfm_var *v = &w->sfm_vars[i];
1337 if (v->var_width == 0)
1339 double d = case_num_idx (c, v->case_index);
1341 put_cmp_opcode (w, 255);
1342 else if (d >= 1 - COMPRESSION_BIAS
1343 && d <= 251 - COMPRESSION_BIAS
1345 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1347 put_cmp_number (w, d);
1351 int offset = v->offset;
1354 /* This code properly deals with a width that is not a
1355 multiple of 8, by ensuring that the final partial
1356 oct (8 byte unit) is treated as padded with spaces
1358 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1360 const void *data = case_str_idx (c, v->case_index) + offset;
1361 int chunk_size = MIN (width, 8);
1362 if (!memcmp (data, " ", chunk_size))
1363 put_cmp_opcode (w, 254);
1365 put_cmp_string (w, data, chunk_size);
1368 /* This code deals properly with padding that is not a
1369 multiple of 8 bytes, by discarding the remainder,
1370 which was already effectively padded with spaces in
1371 the previous loop. (Note that v->width + v->padding
1372 is always a multiple of 8.) */
1373 for (padding = v->padding / 8; padding > 0; padding--)
1374 put_cmp_opcode (w, 254);
1380 start_zstream (struct sfm_writer *w)
1384 error = deflateInit (&w->zstream, 1);
1387 msg (ME, _("Failed to initialize ZLIB for compression (%s)."),
1395 finish_zstream (struct sfm_writer *w)
1397 struct zblock *block;
1400 assert (w->zstream.total_in <= ZBLOCK_SIZE);
1402 w->zstream.next_in = NULL;
1403 w->zstream.avail_in = 0;
1408 w->zstream.next_out = buf;
1409 w->zstream.avail_out = sizeof buf;
1410 error = deflate (&w->zstream, Z_FINISH);
1411 write_bytes (w, buf, w->zstream.next_out - buf);
1413 while (error == Z_OK);
1415 if (error != Z_STREAM_END)
1416 msg (ME, _("Failed to complete ZLIB stream compression (%s)."),
1419 if (w->n_blocks >= w->allocated_blocks)
1420 w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks,
1422 block = &w->blocks[w->n_blocks++];
1423 block->uncompressed_size = w->zstream.total_in;
1424 block->compressed_size = w->zstream.total_out;
1428 write_zlib (struct sfm_writer *w, const void *data_, unsigned int n)
1430 const uint8_t *data = data_;
1436 if (w->zstream.total_in >= ZBLOCK_SIZE)
1442 chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in);
1444 w->zstream.next_in = CONST_CAST (uint8_t *, data);
1445 w->zstream.avail_in = chunk;
1451 w->zstream.next_out = buf;
1452 w->zstream.avail_out = sizeof buf;
1453 error = deflate (&w->zstream, Z_NO_FLUSH);
1454 write_bytes (w, buf, w->zstream.next_out - buf);
1457 msg (ME, _("ZLIB stream compression failed (%s)."),
1462 while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0);
1469 write_ztrailer (struct sfm_writer *w)
1471 long long int uncompressed_ofs;
1472 long long int compressed_ofs;
1473 const struct zblock *block;
1475 write_int64 (w, -COMPRESSION_BIAS);
1477 write_int (w, ZBLOCK_SIZE);
1478 write_int (w, w->n_blocks);
1480 uncompressed_ofs = w->zstart;
1481 compressed_ofs = w->zstart + 24;
1482 for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++)
1484 write_int64 (w, uncompressed_ofs);
1485 write_int64 (w, compressed_ofs);
1486 write_int (w, block->uncompressed_size);
1487 write_int (w, block->compressed_size);
1489 uncompressed_ofs += block->uncompressed_size;
1490 compressed_ofs += block->compressed_size;
1493 if (!fseeko (w->file, w->zstart + 8, SEEK_SET))
1495 write_int64 (w, compressed_ofs);
1496 write_int64 (w, 24 + (w->n_blocks * 24));
1499 msg (ME, _("%s: Seek failed (%s)."),
1500 fh_get_file_name (w->fh), strerror (errno));
1503 /* Flushes buffered compressed opcodes and data to W. */
1505 flush_compressed (struct sfm_writer *w)
1509 unsigned int n = 8 * (1 + w->n_elements);
1510 if (w->compression == SFM_COMP_SIMPLE)
1511 write_bytes (w, w->cbuf, n);
1513 write_zlib (w, w->cbuf, n);
1515 w->n_opcodes = w->n_elements = 0;
1516 memset (w->cbuf[0], 0, 8);
1520 /* Appends OPCODE to the buffered set of compression opcodes in
1521 W. Flushes the compression buffer beforehand if necessary. */
1523 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1525 if (w->n_opcodes >= 8)
1526 flush_compressed (w);
1528 w->cbuf[0][w->n_opcodes++] = opcode;
1531 /* Appends NUMBER to the buffered compression data in W. */
1533 put_cmp_number (struct sfm_writer *w, double number)
1535 put_cmp_opcode (w, 253);
1536 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1539 /* Appends SIZE bytes of DATA to the buffered compression data in
1540 W, followed by enough spaces to pad the output data to exactly
1541 8 bytes (thus, SIZE must be no greater than 8). */
1543 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1547 put_cmp_opcode (w, 253);
1549 memset (w->cbuf[w->n_elements], w->space, 8);
1550 memcpy (w->cbuf[w->n_elements], data, size);
1553 /* Writes 32-bit integer X to the output file for writer W. */
1555 write_int (struct sfm_writer *w, int32_t x)
1557 write_bytes (w, &x, sizeof x);
1560 /* Writes 64-bit integer X to the output file for writer W. */
1562 write_int64 (struct sfm_writer *w, int64_t x)
1564 write_bytes (w, &x, sizeof x);
1567 /* Converts NATIVE to the 64-bit format used in output files in
1570 convert_double_to_output_format (double native, uint8_t output[8])
1572 /* If "double" is not a 64-bit type, then convert it to a
1573 64-bit type. Otherwise just copy it. */
1574 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1575 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1577 memcpy (output, &native, sizeof native);
1580 /* Writes floating-point number X to the output file for writer
1583 write_float (struct sfm_writer *w, double x)
1586 convert_double_to_output_format (x, output);
1587 write_bytes (w, output, sizeof output);
1590 /* Writes contents of VALUE with the given WIDTH to W, padding
1591 with zeros to a multiple of 8 bytes.
1592 To avoid a branch, and because we don't actually need to
1593 support it, WIDTH must be no bigger than 8. */
1595 write_value (struct sfm_writer *w, const union value *value, int width)
1597 assert (width <= 8);
1599 write_float (w, value->f);
1602 write_bytes (w, value_str (value, width), width);
1603 write_zeros (w, 8 - width);
1607 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1608 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1609 is padded on the right with spaces. */
1611 write_string (struct sfm_writer *w, const char *string, size_t width)
1613 size_t data_bytes = MIN (strlen (string), width);
1614 size_t pad_bytes = width - data_bytes;
1615 write_bytes (w, string, data_bytes);
1616 while (pad_bytes-- > 0)
1617 putc (w->space, w->file);
1620 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1621 recoded version in a field of the given WIDTH to W. The string is truncated
1622 or padded on the right with spaces to exactly WIDTH bytes. */
1624 write_utf8_string (struct sfm_writer *w, const char *encoding,
1625 const char *string, size_t width)
1627 char *s = recode_string (encoding, "UTF-8", string, -1);
1628 write_string (w, s, width);
1632 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1633 from UTF-8 encoded into ENCODING. */
1635 write_utf8_record (struct sfm_writer *w, const char *encoding,
1636 const struct string *content, int subtype)
1640 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1641 write_string_record (w, s, subtype);
1645 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1648 write_string_record (struct sfm_writer *w,
1649 const struct substring content, int subtype)
1652 write_int (w, subtype);
1654 write_int (w, ss_length (content));
1655 write_bytes (w, ss_data (content), ss_length (content));
1658 /* Writes SIZE bytes of DATA to W's output file. */
1660 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1662 fwrite (data, 1, size, w->file);
1665 /* Writes N zeros to W's output file. */
1667 write_zeros (struct sfm_writer *w, size_t n)
1673 /* Writes N spaces to W's output file. */
1675 write_spaces (struct sfm_writer *w, size_t n)
1678 putc (w->space, w->file);