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 a series of 8-byte elements, with 1 to 9
82 such elements clustered together. The first element in a cluster is 8
83 1-byte opcodes. Some opcodes call for an additional element in the
84 cluster (hence, if there are eight such opcodes, then the cluster
85 contains a full 9 elements).
87 cbuf[] holds a cluster at a time. */
89 int n_opcodes; /* Number of opcodes in cbuf[0] so far. */
90 int n_elements; /* Number of elements in cbuf[] so far. */
93 struct sfm_var *sfm_vars; /* Variables. */
94 size_t sfm_var_cnt; /* Number of variables. */
95 size_t segment_cnt; /* Number of variables including extra segments
96 for long string variables. */
99 static const struct casewriter_class sys_file_casewriter_class;
101 static void write_header (struct sfm_writer *, const struct dictionary *);
102 static void write_variable (struct sfm_writer *, const struct variable *);
103 static void write_value_labels (struct sfm_writer *,
104 const struct dictionary *);
105 static void write_integer_info_record (struct sfm_writer *,
106 const struct dictionary *);
107 static void write_float_info_record (struct sfm_writer *);
109 static void write_longvar_table (struct sfm_writer *w,
110 const struct dictionary *dict);
112 static void write_encoding_record (struct sfm_writer *w,
113 const struct dictionary *);
115 static void write_vls_length_table (struct sfm_writer *w,
116 const struct dictionary *dict);
118 static void write_long_string_value_labels (struct sfm_writer *,
119 const struct dictionary *);
120 static void write_long_string_missing_values (struct sfm_writer *,
121 const struct dictionary *);
123 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
126 static void write_variable_display_parameters (struct sfm_writer *w,
127 const struct dictionary *dict);
129 static void write_documents (struct sfm_writer *, const struct dictionary *);
131 static void write_data_file_attributes (struct sfm_writer *,
132 const struct dictionary *);
133 static void write_variable_attributes (struct sfm_writer *,
134 const struct dictionary *);
136 static void write_int (struct sfm_writer *, int32_t);
137 static inline void convert_double_to_output_format (double, uint8_t[8]);
138 static void write_float (struct sfm_writer *, double);
139 static void write_string (struct sfm_writer *, const char *, size_t);
140 static void write_utf8_string (struct sfm_writer *, const char *encoding,
141 const char *string, size_t width);
142 static void write_utf8_record (struct sfm_writer *, const char *encoding,
143 const struct string *content, int subtype);
144 static void write_string_record (struct sfm_writer *,
145 const struct substring content, int subtype);
146 static void write_bytes (struct sfm_writer *, const void *, size_t);
147 static void write_zeros (struct sfm_writer *, size_t);
148 static void write_spaces (struct sfm_writer *, size_t);
149 static void write_value (struct sfm_writer *, const union value *, int width);
151 static void write_case_uncompressed (struct sfm_writer *,
152 const struct ccase *);
153 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
154 static void flush_compressed (struct sfm_writer *);
155 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
156 static void put_cmp_number (struct sfm_writer *, double);
157 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
159 static bool write_error (const struct sfm_writer *);
160 static bool close_writer (struct sfm_writer *);
162 /* Returns default options for writing a system file. */
163 struct sfm_write_options
164 sfm_writer_default_options (void)
166 struct sfm_write_options opts;
167 opts.create_writeable = true;
168 opts.compress = settings_get_scompression ();
173 /* Opens the system file designated by file handle FH for writing
174 cases from dictionary D according to the given OPTS.
176 No reference to D is retained, so it may be modified or
177 destroyed at will after this function returns. D is not
178 modified by this function, except to assign short names. */
180 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
181 struct sfm_write_options opts)
183 struct encoding_info encoding_info;
184 struct sfm_writer *w;
189 if (opts.version != 2 && opts.version != 3)
191 msg (ME, _("Unknown system file version %d. Treating as version %d."),
196 /* Create and initialize writer. */
197 w = xmalloc (sizeof *w);
203 w->compress = opts.compress;
206 w->n_opcodes = w->n_elements = 0;
207 memset (w->cbuf[0], 0, 8);
209 /* Figure out how to map in-memory case data to on-disk case
210 data. Also count the number of segments. Very long strings
211 occupy multiple segments, otherwise each variable only takes
213 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
216 /* Open file handle as an exclusive writer. */
217 /* TRANSLATORS: this fragment will be interpolated into
218 messages in fh_lock() that identify types of files. */
219 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
223 /* Create the file on disk. */
225 if (opts.create_writeable)
227 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
231 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
232 fh_get_file_name (fh), strerror (errno));
236 get_encoding_info (&encoding_info, dict_get_encoding (d));
237 w->space = encoding_info.space[0];
239 /* Write the file header. */
242 /* Write basic variable info. */
243 short_names_assign (d);
244 for (i = 0; i < dict_get_var_cnt (d); i++)
245 write_variable (w, dict_get_var (d, i));
247 write_value_labels (w, d);
249 if (dict_get_document_line_cnt (d) > 0)
250 write_documents (w, d);
252 write_integer_info_record (w, d);
253 write_float_info_record (w);
255 write_mrsets (w, d, true);
257 write_variable_display_parameters (w, d);
259 if (opts.version >= 3)
260 write_longvar_table (w, d);
262 write_vls_length_table (w, d);
264 write_long_string_value_labels (w, d);
265 write_long_string_missing_values (w, d);
267 if (opts.version >= 3)
269 if (attrset_count (dict_get_attributes (d)))
270 write_data_file_attributes (w, d);
271 write_variable_attributes (w, d);
274 write_mrsets (w, d, false);
276 write_encoding_record (w, d);
278 /* Write end-of-headers record. */
285 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
292 /* Returns value of X truncated to two least-significant digits. */
303 /* Calculates the offset of data for TARGET_VAR from the
304 beginning of each case's data for dictionary D. The return
305 value is in "octs" (8-byte units). */
307 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
313 for (i = 0; i < dict_get_var_cnt (d); i++)
315 struct variable *var = dict_get_var (d, i);
316 if (var == target_var)
318 oct_idx += sfm_width_to_octs (var_get_width (var));
323 /* Write the sysfile_header header to system file W. */
325 write_header (struct sfm_writer *w, const struct dictionary *d)
327 const char *dict_encoding = dict_get_encoding (d);
329 char creation_date[10];
330 char creation_time[9];
331 const char *file_label;
332 struct variable *weight;
336 /* Record-type code. */
337 if (is_encoding_ebcdic_compatible (dict_encoding))
338 write_string (w, EBCDIC_MAGIC, 4);
340 write_string (w, ASCII_MAGIC, 4);
342 /* Product identification. */
343 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
344 version, host_system);
345 write_utf8_string (w, dict_encoding, prod_name, 60);
350 /* Number of `union value's per case. */
351 write_int (w, calc_oct_idx (d, NULL));
354 write_int (w, w->compress);
356 /* Weight variable. */
357 weight = dict_get_weight (d);
358 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
360 /* Number of cases. We don't know this in advance, so we write
361 -1 to indicate an unknown number of cases. Later we can
362 come back and overwrite it with the true value. */
365 /* Compression bias. */
366 write_float (w, COMPRESSION_BIAS);
368 /* Creation date and time. */
369 if (time (&t) == (time_t) -1)
371 strcpy (creation_date, "01 Jan 70");
372 strcpy (creation_time, "00:00:00");
376 static const char *const month_name[12] =
378 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
379 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
381 struct tm *tmp = localtime (&t);
382 int day = rerange (tmp->tm_mday);
383 int mon = rerange (tmp->tm_mon + 1);
384 int year = rerange (tmp->tm_year);
385 int hour = rerange (tmp->tm_hour + 1);
386 int min = rerange (tmp->tm_min + 1);
387 int sec = rerange (tmp->tm_sec + 1);
389 snprintf (creation_date, sizeof creation_date,
390 "%02d %s %02d", day, month_name[mon - 1], year);
391 snprintf (creation_time, sizeof creation_time,
392 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
394 write_utf8_string (w, dict_encoding, creation_date, 9);
395 write_utf8_string (w, dict_encoding, creation_time, 8);
398 file_label = dict_get_label (d);
399 if (file_label == NULL)
401 write_utf8_string (w, dict_encoding, file_label, 64);
407 /* Write format spec FMT to W, after adjusting it to be
408 compatible with the given WIDTH. */
410 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
412 assert (fmt_check_output (&fmt));
413 assert (sfm_width_to_segments (width) == 1);
416 fmt_resize (&fmt, width);
417 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
420 /* Write a string continuation variable record for each 8-byte
421 section beyond the initial 8 bytes, for a variable of the
424 write_variable_continuation_records (struct sfm_writer *w, int width)
428 assert (sfm_width_to_segments (width) == 1);
429 for (position = 8; position < width; position += 8)
431 write_int (w, 2); /* Record type. */
432 write_int (w, -1); /* Width. */
433 write_int (w, 0); /* No variable label. */
434 write_int (w, 0); /* No missing values. */
435 write_int (w, 0); /* Print format. */
436 write_int (w, 0); /* Write format. */
437 write_zeros (w, 8); /* Name. */
441 /* Write the variable record(s) for variable V to system file
444 write_variable (struct sfm_writer *w, const struct variable *v)
446 int width = var_get_width (v);
447 int segment_cnt = sfm_width_to_segments (width);
448 int seg0_width = sfm_segment_alloc_width (width, 0);
449 const char *encoding = var_get_encoding (v);
456 write_int (w, seg0_width);
458 /* Variable has a variable label? */
459 write_int (w, var_has_label (v));
461 /* Number of missing values. If there is a range, then the
462 range counts as 2 missing values and causes the number to be
465 Missing values for long string variables are written in a separate
467 if (width <= MAX_SHORT_STRING)
469 const struct missing_values *mv = var_get_missing_values (v);
470 if (mv_has_range (mv))
471 write_int (w, -2 - mv_n_values (mv));
473 write_int (w, mv_n_values (mv));
478 /* Print and write formats. */
479 write_format (w, *var_get_print_format (v), seg0_width);
480 write_format (w, *var_get_write_format (v), seg0_width);
483 The full name is in a translation table written
485 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
488 if (var_has_label (v))
490 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
491 size_t label_len = MIN (strlen (label), 255);
492 size_t padded_len = ROUND_UP (label_len, 4);
493 write_int (w, label_len);
494 write_string (w, label, padded_len);
498 /* Write the missing values, if any, range first. */
499 if (width <= MAX_SHORT_STRING)
501 const struct missing_values *mv = var_get_missing_values (v);
502 if (mv_has_range (mv))
505 mv_get_range (mv, &x, &y);
509 for (i = 0; i < mv_n_values (mv); i++)
510 write_value (w, mv_get_value (mv, i), width);
513 write_variable_continuation_records (w, seg0_width);
515 /* Write additional segments for very long string variables. */
516 for (i = 1; i < segment_cnt; i++)
518 int seg_width = sfm_segment_alloc_width (width, i);
519 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
521 write_int (w, 2); /* Variable record. */
522 write_int (w, seg_width); /* Width. */
523 write_int (w, 0); /* No variable label. */
524 write_int (w, 0); /* No missing values. */
525 write_format (w, fmt, seg_width); /* Print format. */
526 write_format (w, fmt, seg_width); /* Write format. */
527 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
529 write_variable_continuation_records (w, seg_width);
533 /* Writes the value labels to system file W.
535 Value labels for long string variables are written separately,
536 by write_long_string_value_labels. */
538 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
542 struct hmap_node hmap_node;
543 const struct val_labs *val_labs;
545 size_t n_indexes, allocated_indexes;
548 size_t n_sets, allocated_sets;
549 struct label_set **sets;
550 struct hmap same_sets;
554 n_sets = allocated_sets = 0;
556 hmap_init (&same_sets);
559 for (i = 0; i < dict_get_var_cnt (d); i++)
561 struct variable *v = dict_get_var (d, i);
563 if (var_has_value_labels (v) && var_get_width (v) <= 8)
565 const struct val_labs *val_labs = var_get_value_labels (v);
566 unsigned int hash = val_labs_hash (val_labs, 0);
567 struct label_set *set;
569 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
572 if (val_labs_equal (set->val_labs, val_labs))
574 if (set->n_indexes >= set->allocated_indexes)
575 set->indexes = x2nrealloc (set->indexes,
576 &set->allocated_indexes,
577 sizeof *set->indexes);
578 set->indexes[set->n_indexes++] = idx;
583 set = xmalloc (sizeof *set);
584 set->val_labs = val_labs;
585 set->indexes = xmalloc (sizeof *set->indexes);
586 set->indexes[0] = idx;
588 set->allocated_indexes = 1;
589 hmap_insert (&same_sets, &set->hmap_node, hash);
591 if (n_sets >= allocated_sets)
592 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
593 sets[n_sets++] = set;
597 idx += sfm_width_to_octs (var_get_width (v));
600 for (i = 0; i < n_sets; i++)
602 const struct label_set *set = sets[i];
603 const struct val_labs *val_labs = set->val_labs;
604 size_t n_labels = val_labs_count (val_labs);
605 int width = val_labs_get_width (val_labs);
606 const struct val_lab **labels;
609 /* Value label record. */
610 write_int (w, 3); /* Record type. */
611 write_int (w, n_labels);
612 labels = val_labs_sorted (val_labs);
613 for (j = 0; j < n_labels; j++)
615 const struct val_lab *vl = labels[j];
616 char *label = recode_string (dict_get_encoding (d), UTF8,
617 val_lab_get_escaped_label (vl), -1);
618 uint8_t len = MIN (strlen (label), 255);
620 write_value (w, val_lab_get_value (vl), width);
621 write_bytes (w, &len, 1);
622 write_bytes (w, label, len);
623 write_zeros (w, REM_RND_UP (len + 1, 8));
628 /* Value label variable record. */
629 write_int (w, 4); /* Record type. */
630 write_int (w, set->n_indexes);
631 for (j = 0; j < set->n_indexes; j++)
632 write_int (w, set->indexes[j] + 1);
635 for (i = 0; i < n_sets; i++)
637 struct label_set *set = sets[i];
643 hmap_destroy (&same_sets);
646 /* Writes record type 6, document record. */
648 write_documents (struct sfm_writer *w, const struct dictionary *d)
650 const struct string_array *docs = dict_get_documents (d);
651 const char *enc = dict_get_encoding (d);
654 write_int (w, 6); /* Record type. */
655 write_int (w, docs->n);
656 for (i = 0; i < docs->n; i++)
658 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
659 size_t s_len = strlen (s);
660 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
662 write_bytes (w, s, write_len);
663 write_spaces (w, DOC_LINE_LENGTH - write_len);
669 put_attrset (struct string *string, const struct attrset *attrs)
671 const struct attribute *attr;
672 struct attrset_iterator i;
674 for (attr = attrset_first (attrs, &i); attr != NULL;
675 attr = attrset_next (attrs, &i))
677 size_t n_values = attribute_get_n_values (attr);
680 ds_put_cstr (string, attribute_get_name (attr));
681 ds_put_byte (string, '(');
682 for (j = 0; j < n_values; j++)
683 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
684 ds_put_byte (string, ')');
689 write_data_file_attributes (struct sfm_writer *w,
690 const struct dictionary *d)
692 struct string s = DS_EMPTY_INITIALIZER;
693 put_attrset (&s, dict_get_attributes (d));
694 write_utf8_record (w, dict_get_encoding (d), &s, 17);
699 add_role_attribute (enum var_role role, struct attrset *attrs)
701 struct attribute *attr;
731 attrset_delete (attrs, "$@Role");
733 attr = attribute_create ("$@Role");
734 attribute_add_value (attr, s);
735 attrset_add (attrs, attr);
739 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
741 struct string s = DS_EMPTY_INITIALIZER;
742 size_t n_vars = dict_get_var_cnt (d);
743 size_t n_attrsets = 0;
746 for (i = 0; i < n_vars; i++)
748 struct variable *v = dict_get_var (d, i);
749 struct attrset attrs;
751 attrset_clone (&attrs, var_get_attributes (v));
753 add_role_attribute (var_get_role (v), &attrs);
755 ds_put_byte (&s, '/');
756 ds_put_format (&s, "%s:", var_get_name (v));
757 put_attrset (&s, &attrs);
758 attrset_destroy (&attrs);
761 write_utf8_record (w, dict_get_encoding (d), &s, 18);
765 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
766 SPSS before release 14, otherwise writes sets supported only by later
769 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
772 const char *encoding = dict_get_encoding (dict);
773 struct string s = DS_EMPTY_INITIALIZER;
777 if (is_encoding_ebcdic_compatible (encoding))
783 n_mrsets = dict_get_n_mrsets (dict);
787 for (i = 0; i < n_mrsets; i++)
789 const struct mrset *mrset = dict_get_mrset (dict, i);
793 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
797 name = recode_string (encoding, "UTF-8", mrset->name, -1);
798 ds_put_format (&s, "%s=", name);
801 if (mrset->type == MRSET_MD)
805 if (mrset->cat_source == MRSET_COUNTEDVALUES)
806 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
808 ds_put_byte (&s, 'D');
810 if (mrset->width == 0)
811 counted = xasprintf ("%.0f", mrset->counted.f);
813 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
815 ds_put_format (&s, "%zu %s", strlen (counted), counted);
819 ds_put_byte (&s, 'C');
820 ds_put_byte (&s, ' ');
822 if (mrset->label && !mrset->label_from_var_label)
824 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
825 ds_put_format (&s, "%zu %s", strlen (label), label);
829 ds_put_cstr (&s, "0 ");
831 for (j = 0; j < mrset->n_vars; j++)
833 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
834 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
835 char *short_name = recode_string (encoding, "UTF-8",
836 lower_name_utf8, -1);
837 ds_put_format (&s, " %s", short_name);
839 free (lower_name_utf8);
841 ds_put_byte (&s, '\n');
844 if (!ds_is_empty (&s))
845 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
849 /* Write the alignment, width and scale values. */
851 write_variable_display_parameters (struct sfm_writer *w,
852 const struct dictionary *dict)
856 write_int (w, 7); /* Record type. */
857 write_int (w, 11); /* Record subtype. */
858 write_int (w, 4); /* Data item (int32) size. */
859 write_int (w, w->segment_cnt * 3); /* Number of data items. */
861 for (i = 0; i < dict_get_var_cnt (dict); ++i)
863 struct variable *v = dict_get_var (dict, i);
864 int width = var_get_width (v);
865 int segment_cnt = sfm_width_to_segments (width);
866 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
867 : var_get_measure (v) == MEASURE_ORDINAL ? 2
869 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
870 : var_get_alignment (v) == ALIGN_RIGHT ? 1
874 for (i = 0; i < segment_cnt; i++)
876 int width_left = width - sfm_segment_effective_offset (width, i);
877 write_int (w, measure);
878 write_int (w, (i == 0 ? var_get_display_width (v)
879 : var_default_display_width (width_left)));
880 write_int (w, alignment);
885 /* Writes the table of lengths for very long string variables. */
887 write_vls_length_table (struct sfm_writer *w,
888 const struct dictionary *dict)
893 ds_init_empty (&map);
894 for (i = 0; i < dict_get_var_cnt (dict); ++i)
896 const struct variable *v = dict_get_var (dict, i);
897 if (sfm_width_to_segments (var_get_width (v)) > 1)
898 ds_put_format (&map, "%s=%05d%c\t",
899 var_get_short_name (v, 0), var_get_width (v), 0);
901 if (!ds_is_empty (&map))
902 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
907 write_long_string_value_labels (struct sfm_writer *w,
908 const struct dictionary *dict)
910 const char *encoding = dict_get_encoding (dict);
911 size_t n_vars = dict_get_var_cnt (dict);
915 /* Figure out the size in advance. */
917 for (i = 0; i < n_vars; i++)
919 struct variable *var = dict_get_var (dict, i);
920 const struct val_labs *val_labs = var_get_value_labels (var);
921 int width = var_get_width (var);
922 const struct val_lab *val_lab;
924 if (val_labs_count (val_labs) == 0 || width < 9)
928 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
929 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
930 val_lab = val_labs_next (val_labs, val_lab))
933 size += recode_string_len (encoding, "UTF-8",
934 val_lab_get_escaped_label (val_lab), -1);
940 write_int (w, 7); /* Record type. */
941 write_int (w, 21); /* Record subtype */
942 write_int (w, 1); /* Data item (byte) size. */
943 write_int (w, size); /* Number of data items. */
945 start = ftello (w->file);
946 for (i = 0; i < n_vars; i++)
948 struct variable *var = dict_get_var (dict, i);
949 const struct val_labs *val_labs = var_get_value_labels (var);
950 int width = var_get_width (var);
951 const struct val_lab *val_lab;
954 if (val_labs_count (val_labs) == 0 || width < 9)
957 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
958 write_int (w, strlen (var_name));
959 write_bytes (w, var_name, strlen (var_name));
962 write_int (w, width);
963 write_int (w, val_labs_count (val_labs));
964 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
965 val_lab = val_labs_next (val_labs, val_lab))
970 write_int (w, width);
971 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
974 label = recode_string (var_get_encoding (var), "UTF-8",
975 val_lab_get_escaped_label (val_lab), -1);
976 len = strlen (label);
978 write_bytes (w, label, len);
982 assert (ftello (w->file) == start + size);
986 write_long_string_missing_values (struct sfm_writer *w,
987 const struct dictionary *dict)
989 const char *encoding = dict_get_encoding (dict);
990 size_t n_vars = dict_get_var_cnt (dict);
994 /* Figure out the size in advance. */
996 for (i = 0; i < n_vars; i++)
998 struct variable *var = dict_get_var (dict, i);
999 const struct missing_values *mv = var_get_missing_values (var);
1000 int width = var_get_width (var);
1002 if (mv_is_empty (mv) || width < 9)
1006 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1008 size += mv_n_values (mv) * (4 + 8);
1013 write_int (w, 7); /* Record type. */
1014 write_int (w, 22); /* Record subtype */
1015 write_int (w, 1); /* Data item (byte) size. */
1016 write_int (w, size); /* Number of data items. */
1018 start = ftello (w->file);
1019 for (i = 0; i < n_vars; i++)
1021 struct variable *var = dict_get_var (dict, i);
1022 const struct missing_values *mv = var_get_missing_values (var);
1023 int width = var_get_width (var);
1024 uint8_t n_missing_values;
1028 if (mv_is_empty (mv) || width < 9)
1031 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1032 write_int (w, strlen (var_name));
1033 write_bytes (w, var_name, strlen (var_name));
1036 n_missing_values = mv_n_values (mv);
1037 write_bytes (w, &n_missing_values, 1);
1039 for (j = 0; j < n_missing_values; j++)
1041 const union value *value = mv_get_value (mv, j);
1044 write_bytes (w, value_str (value, width), 8);
1047 assert (ftello (w->file) == start + size);
1051 write_encoding_record (struct sfm_writer *w,
1052 const struct dictionary *d)
1054 /* IANA says "...character set names may be up to 40 characters taken
1055 from the printable characters of US-ASCII," so character set names
1056 don't need to be recoded to be in UTF-8.
1058 We convert encoding names to uppercase because SPSS writes encoding
1059 names in uppercase. */
1060 char *encoding = xstrdup (dict_get_encoding (d));
1061 str_uppercase (encoding);
1062 write_string_record (w, ss_cstr (encoding), 20);
1066 /* Writes the long variable name table. */
1068 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1073 ds_init_empty (&map);
1074 for (i = 0; i < dict_get_var_cnt (dict); i++)
1076 struct variable *v = dict_get_var (dict, i);
1078 ds_put_byte (&map, '\t');
1079 ds_put_format (&map, "%s=%s",
1080 var_get_short_name (v, 0), var_get_name (v));
1082 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1086 /* Write integer information record. */
1088 write_integer_info_record (struct sfm_writer *w,
1089 const struct dictionary *d)
1091 const char *dict_encoding = dict_get_encoding (d);
1092 int version_component[3];
1096 /* Parse the version string. */
1097 memset (version_component, 0, sizeof version_component);
1098 sscanf (bare_version, "%d.%d.%d",
1099 &version_component[0], &version_component[1], &version_component[2]);
1101 /* Figure out the floating-point format. */
1102 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1103 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1105 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1107 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1112 /* Choose codepage. */
1113 codepage = sys_get_codepage_from_encoding (dict_encoding);
1116 /* The codepage is unknown. Choose a default.
1118 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1120 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1121 many files use this codepage number regardless of their actual
1124 if (is_encoding_ascii_compatible (dict_encoding))
1126 else if (is_encoding_ebcdic_compatible (dict_encoding))
1131 write_int (w, 7); /* Record type. */
1132 write_int (w, 3); /* Record subtype. */
1133 write_int (w, 4); /* Data item (int32) size. */
1134 write_int (w, 8); /* Number of data items. */
1135 write_int (w, version_component[0]);
1136 write_int (w, version_component[1]);
1137 write_int (w, version_component[2]);
1138 write_int (w, -1); /* Machine code. */
1139 write_int (w, float_format);
1140 write_int (w, 1); /* Compression code. */
1141 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1142 write_int (w, codepage);
1145 /* Write floating-point information record. */
1147 write_float_info_record (struct sfm_writer *w)
1149 write_int (w, 7); /* Record type. */
1150 write_int (w, 4); /* Record subtype. */
1151 write_int (w, 8); /* Data item (flt64) size. */
1152 write_int (w, 3); /* Number of data items. */
1153 write_float (w, SYSMIS); /* System-missing value. */
1154 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1155 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1158 /* Writes case C to system file W. */
1160 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1163 struct sfm_writer *w = w_;
1165 if (ferror (w->file))
1167 casewriter_force_error (writer);
1175 write_case_uncompressed (w, c);
1177 write_case_compressed (w, c);
1182 /* Destroys system file writer W. */
1184 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1186 struct sfm_writer *w = w_;
1187 if (!close_writer (w))
1188 casewriter_force_error (writer);
1191 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1193 write_error (const struct sfm_writer *writer)
1195 return ferror (writer->file);
1198 /* Closes a system file after we're done with it.
1199 Returns true if successful, false if an I/O error occurred. */
1201 close_writer (struct sfm_writer *w)
1209 if (w->file != NULL)
1212 flush_compressed (w);
1215 ok = !write_error (w);
1217 /* Seek back to the beginning and update the number of cases.
1218 This is just a courtesy to later readers, so there's no need
1219 to check return values or report errors. */
1220 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1222 write_int (w, w->case_cnt);
1226 if (fclose (w->file) == EOF)
1230 msg (ME, _("An I/O error occurred writing system file `%s'."),
1231 fh_get_file_name (w->fh));
1233 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1237 fh_unlock (w->lock);
1246 /* System file writer casewriter class. */
1247 static const struct casewriter_class sys_file_casewriter_class =
1249 sys_file_casewriter_write,
1250 sys_file_casewriter_destroy,
1254 /* Writes case C to system file W, without compressing it. */
1256 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1260 for (i = 0; i < w->sfm_var_cnt; i++)
1262 struct sfm_var *v = &w->sfm_vars[i];
1264 if (v->var_width == 0)
1265 write_float (w, case_num_idx (c, v->case_index));
1268 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1270 write_spaces (w, v->padding);
1275 /* Writes case C to system file W, with compression. */
1277 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1281 for (i = 0; i < w->sfm_var_cnt; i++)
1283 struct sfm_var *v = &w->sfm_vars[i];
1285 if (v->var_width == 0)
1287 double d = case_num_idx (c, v->case_index);
1289 put_cmp_opcode (w, 255);
1290 else if (d >= 1 - COMPRESSION_BIAS
1291 && d <= 251 - COMPRESSION_BIAS
1293 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1296 put_cmp_opcode (w, 253);
1297 put_cmp_number (w, d);
1302 int offset = v->offset;
1305 /* This code properly deals with a width that is not a
1306 multiple of 8, by ensuring that the final partial
1307 oct (8 byte unit) is treated as padded with spaces
1309 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1311 const void *data = case_str_idx (c, v->case_index) + offset;
1312 int chunk_size = MIN (width, 8);
1313 if (!memcmp (data, " ", chunk_size))
1314 put_cmp_opcode (w, 254);
1317 put_cmp_opcode (w, 253);
1318 put_cmp_string (w, data, chunk_size);
1322 /* This code deals properly with padding that is not a
1323 multiple of 8 bytes, by discarding the remainder,
1324 which was already effectively padded with spaces in
1325 the previous loop. (Note that v->width + v->padding
1326 is always a multiple of 8.) */
1327 for (padding = v->padding / 8; padding > 0; padding--)
1328 put_cmp_opcode (w, 254);
1333 /* Flushes buffered compressed opcodes and data to W. */
1335 flush_compressed (struct sfm_writer *w)
1339 write_bytes (w, w->cbuf, 8 * (1 + w->n_elements));
1340 w->n_opcodes = w->n_elements = 0;
1341 memset (w->cbuf[0], 0, 8);
1345 /* Appends OPCODE to the buffered set of compression opcodes in
1346 W. Flushes the compression buffer beforehand if necessary. */
1348 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1350 if (w->n_opcodes >= 8)
1351 flush_compressed (w);
1353 w->cbuf[0][w->n_opcodes++] = opcode;
1356 /* Appends NUMBER to the buffered compression data in W. The
1357 buffer must not be full; the way to assure that is to call
1358 this function only just after a call to put_cmp_opcode, which
1359 will flush the buffer as necessary. */
1361 put_cmp_number (struct sfm_writer *w, double number)
1363 assert (w->opcode_cnt > 0);
1364 assert (w->data_cnt < 8);
1366 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1369 /* Appends SIZE bytes of DATA to the buffered compression data in
1370 W, followed by enough spaces to pad the output data to exactly
1371 8 bytes (thus, SIZE must be no greater than 8). The buffer
1372 must not be full; the way to assure that is to call this
1373 function only just after a call to put_cmp_opcode, which will
1374 flush the buffer as necessary. */
1376 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1378 assert (w->opcode_cnt > 0);
1379 assert (w->data_cnt < 8);
1383 memset (w->cbuf[w->n_elements], w->space, 8);
1384 memcpy (w->cbuf[w->n_elements], data, size);
1387 /* Writes 32-bit integer X to the output file for writer W. */
1389 write_int (struct sfm_writer *w, int32_t x)
1391 write_bytes (w, &x, sizeof x);
1394 /* Converts NATIVE to the 64-bit format used in output files in
1397 convert_double_to_output_format (double native, uint8_t output[8])
1399 /* If "double" is not a 64-bit type, then convert it to a
1400 64-bit type. Otherwise just copy it. */
1401 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1402 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1404 memcpy (output, &native, sizeof native);
1407 /* Writes floating-point number X to the output file for writer
1410 write_float (struct sfm_writer *w, double x)
1413 convert_double_to_output_format (x, output);
1414 write_bytes (w, output, sizeof output);
1417 /* Writes contents of VALUE with the given WIDTH to W, padding
1418 with zeros to a multiple of 8 bytes.
1419 To avoid a branch, and because we don't actually need to
1420 support it, WIDTH must be no bigger than 8. */
1422 write_value (struct sfm_writer *w, const union value *value, int width)
1424 assert (width <= 8);
1426 write_float (w, value->f);
1429 write_bytes (w, value_str (value, width), width);
1430 write_zeros (w, 8 - width);
1434 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1435 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1436 is padded on the right with spaces. */
1438 write_string (struct sfm_writer *w, const char *string, size_t width)
1440 size_t data_bytes = MIN (strlen (string), width);
1441 size_t pad_bytes = width - data_bytes;
1442 write_bytes (w, string, data_bytes);
1443 while (pad_bytes-- > 0)
1444 putc (w->space, w->file);
1447 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1448 recoded version in a field of the given WIDTH to W. The string is truncated
1449 or padded on the right with spaces to exactly WIDTH bytes. */
1451 write_utf8_string (struct sfm_writer *w, const char *encoding,
1452 const char *string, size_t width)
1454 char *s = recode_string (encoding, "UTF-8", string, -1);
1455 write_string (w, s, width);
1459 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1460 from UTF-8 encoded into ENCODING. */
1462 write_utf8_record (struct sfm_writer *w, const char *encoding,
1463 const struct string *content, int subtype)
1467 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1468 write_string_record (w, s, subtype);
1472 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1475 write_string_record (struct sfm_writer *w,
1476 const struct substring content, int subtype)
1479 write_int (w, subtype);
1481 write_int (w, ss_length (content));
1482 write_bytes (w, ss_data (content), ss_length (content));
1485 /* Writes SIZE bytes of DATA to W's output file. */
1487 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1489 fwrite (data, 1, size, w->file);
1492 /* Writes N zeros to W's output file. */
1494 write_zeros (struct sfm_writer *w, size_t n)
1500 /* Writes N spaces to W's output file. */
1502 write_spaces (struct sfm_writer *w, size_t n)
1505 putc (w->space, w->file);