1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2007, 2009, 2010, 2011 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include "data/sys-file-writer.h"
20 #include "data/sys-file-private.h"
29 #include "data/attributes.h"
30 #include "data/case.h"
31 #include "data/casewriter-provider.h"
32 #include "data/casewriter.h"
33 #include "data/dictionary.h"
34 #include "data/file-handle-def.h"
35 #include "data/file-name.h"
36 #include "data/format.h"
37 #include "data/make-file.h"
38 #include "data/missing-values.h"
39 #include "data/mrset.h"
40 #include "data/settings.h"
41 #include "data/short-names.h"
42 #include "data/value-labels.h"
43 #include "data/variable.h"
44 #include "libpspp/float-format.h"
45 #include "libpspp/i18n.h"
46 #include "libpspp/integer-format.h"
47 #include "libpspp/message.h"
48 #include "libpspp/misc.h"
49 #include "libpspp/str.h"
50 #include "libpspp/string-array.h"
51 #include "libpspp/version.h"
53 #include "gl/xmemdup0.h"
54 #include "gl/minmax.h"
55 #include "gl/unlocked-io.h"
56 #include "gl/xalloc.h"
59 #define _(msgid) gettext (msgid)
60 #define N_(msgid) (msgid)
62 /* Compression bias used by PSPP. Values between (1 -
63 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
65 #define COMPRESSION_BIAS 100
67 /* System file writer. */
70 struct file_handle *fh; /* File handle. */
71 struct fh_lock *lock; /* Mutual exclusion for file. */
72 FILE *file; /* File stream. */
73 struct replace_file *rf; /* Ticket for replacing output file. */
75 bool compress; /* 1=compressed, 0=not compressed. */
76 casenumber case_cnt; /* Number of cases written so far. */
78 /* Compression buffering.
80 Compressed data is output as groups of 8 1-byte opcodes
81 followed by up to 8 (depending on the opcodes) 8-byte data
82 items. Data items and opcodes arrive at the same time but
83 must be reordered for writing to disk, thus a small amount
85 uint8_t opcodes[8]; /* Buffered opcodes. */
86 int opcode_cnt; /* Number of buffered opcodes. */
87 uint8_t data[8][8]; /* Buffered data. */
88 int data_cnt; /* Number of buffered data items. */
91 struct sfm_var *sfm_vars; /* Variables. */
92 size_t sfm_var_cnt; /* Number of variables. */
93 size_t segment_cnt; /* Number of variables including extra segments
94 for long string variables. */
97 static const struct casewriter_class sys_file_casewriter_class;
99 static void write_header (struct sfm_writer *, const struct dictionary *);
100 static void write_variable (struct sfm_writer *, const struct variable *);
101 static void write_value_labels (struct sfm_writer *, struct variable *,
103 static void write_integer_info_record (struct sfm_writer *);
104 static void write_float_info_record (struct sfm_writer *);
106 static void write_longvar_table (struct sfm_writer *w,
107 const struct dictionary *dict);
109 static void write_encoding_record (struct sfm_writer *w,
110 const struct dictionary *);
112 static void write_vls_length_table (struct sfm_writer *w,
113 const struct dictionary *dict);
115 static void write_long_string_value_labels (struct sfm_writer *,
116 const struct dictionary *);
118 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
121 static void write_variable_display_parameters (struct sfm_writer *w,
122 const struct dictionary *dict);
124 static void write_documents (struct sfm_writer *, const struct dictionary *);
126 static void write_data_file_attributes (struct sfm_writer *,
127 const struct dictionary *);
128 static void write_variable_attributes (struct sfm_writer *,
129 const struct dictionary *);
131 static void write_int (struct sfm_writer *, int32_t);
132 static inline void convert_double_to_output_format (double, uint8_t[8]);
133 static void write_float (struct sfm_writer *, double);
134 static void write_string (struct sfm_writer *, const char *, size_t);
135 static void write_utf8_string (struct sfm_writer *, const char *encoding,
136 const char *string, size_t width);
137 static void write_utf8_record (struct sfm_writer *, const char *encoding,
138 const struct string *content, int subtype);
139 static void write_bytes (struct sfm_writer *, const void *, size_t);
140 static void write_zeros (struct sfm_writer *, size_t);
141 static void write_spaces (struct sfm_writer *, size_t);
142 static void write_value (struct sfm_writer *, const union value *, int width);
144 static void write_case_uncompressed (struct sfm_writer *,
145 const struct ccase *);
146 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
147 static void flush_compressed (struct sfm_writer *);
148 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
149 static void put_cmp_number (struct sfm_writer *, double);
150 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
152 static bool write_error (const struct sfm_writer *);
153 static bool close_writer (struct sfm_writer *);
155 /* Returns default options for writing a system file. */
156 struct sfm_write_options
157 sfm_writer_default_options (void)
159 struct sfm_write_options opts;
160 opts.create_writeable = true;
161 opts.compress = settings_get_scompression ();
166 /* Opens the system file designated by file handle FH for writing
167 cases from dictionary D according to the given OPTS.
169 No reference to D is retained, so it may be modified or
170 destroyed at will after this function returns. D is not
171 modified by this function, except to assign short names. */
173 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
174 struct sfm_write_options opts)
176 struct sfm_writer *w;
182 if (opts.version != 2 && opts.version != 3)
184 msg (ME, _("Unknown system file version %d. Treating as version %d."),
189 /* Create and initialize writer. */
190 w = xmalloc (sizeof *w);
196 w->compress = opts.compress;
199 w->opcode_cnt = w->data_cnt = 0;
201 /* Figure out how to map in-memory case data to on-disk case
202 data. Also count the number of segments. Very long strings
203 occupy multiple segments, otherwise each variable only takes
205 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
208 /* Open file handle as an exclusive writer. */
209 /* TRANSLATORS: this fragment will be interpolated into
210 messages in fh_lock() that identify types of files. */
211 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
215 /* Create the file on disk. */
217 if (opts.create_writeable)
219 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
223 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
224 fh_get_file_name (fh), strerror (errno));
228 /* Write the file header. */
231 /* Write basic variable info. */
232 short_names_assign (d);
233 for (i = 0; i < dict_get_var_cnt (d); i++)
234 write_variable (w, dict_get_var (d, i));
236 /* Write out value labels. */
238 for (i = 0; i < dict_get_var_cnt (d); i++)
240 struct variable *v = dict_get_var (d, i);
242 write_value_labels (w, v, idx);
243 idx += sfm_width_to_octs (var_get_width (v));
246 if (dict_get_document_line_cnt (d) > 0)
247 write_documents (w, d);
249 write_integer_info_record (w);
250 write_float_info_record (w);
252 write_mrsets (w, d, true);
254 write_variable_display_parameters (w, d);
256 if (opts.version >= 3)
257 write_longvar_table (w, d);
259 write_vls_length_table (w, d);
261 write_long_string_value_labels (w, d);
263 if (attrset_count (dict_get_attributes (d)))
264 write_data_file_attributes (w, d);
265 write_variable_attributes (w, d);
267 write_mrsets (w, d, false);
269 write_encoding_record (w, d);
271 /* Write end-of-headers record. */
278 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
285 /* Returns value of X truncated to two least-significant digits. */
296 /* Calculates the offset of data for TARGET_VAR from the
297 beginning of each case's data for dictionary D. The return
298 value is in "octs" (8-byte units). */
300 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
306 for (i = 0; i < dict_get_var_cnt (d); i++)
308 struct variable *var = dict_get_var (d, i);
309 if (var == target_var)
311 oct_idx += sfm_width_to_octs (var_get_width (var));
316 /* Write the sysfile_header header to system file W. */
318 write_header (struct sfm_writer *w, const struct dictionary *d)
321 char creation_date[10];
322 char creation_time[9];
323 const char *file_label;
324 struct variable *weight;
328 /* Record-type code. */
329 write_string (w, "$FL2", 4);
331 /* Product identification. */
332 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
333 version, host_system);
334 write_string (w, prod_name, 60);
339 /* Number of `union value's per case. */
340 write_int (w, calc_oct_idx (d, NULL));
343 write_int (w, w->compress);
345 /* Weight variable. */
346 weight = dict_get_weight (d);
347 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
349 /* Number of cases. We don't know this in advance, so we write
350 -1 to indicate an unknown number of cases. Later we can
351 come back and overwrite it with the true value. */
354 /* Compression bias. */
355 write_float (w, COMPRESSION_BIAS);
357 /* Creation date and time. */
358 if (time (&t) == (time_t) -1)
360 strcpy (creation_date, "01 Jan 70");
361 strcpy (creation_time, "00:00:00");
365 static const char *const month_name[12] =
367 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
368 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
370 struct tm *tmp = localtime (&t);
371 int day = rerange (tmp->tm_mday);
372 int mon = rerange (tmp->tm_mon + 1);
373 int year = rerange (tmp->tm_year);
374 int hour = rerange (tmp->tm_hour + 1);
375 int min = rerange (tmp->tm_min + 1);
376 int sec = rerange (tmp->tm_sec + 1);
378 snprintf (creation_date, sizeof creation_date,
379 "%02d %s %02d", day, month_name[mon - 1], year);
380 snprintf (creation_time, sizeof creation_time,
381 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
383 write_string (w, creation_date, 9);
384 write_string (w, creation_time, 8);
387 file_label = dict_get_label (d);
388 if (file_label == NULL)
390 write_utf8_string (w, dict_get_encoding (d), file_label, 64);
396 /* Write format spec FMT to W, after adjusting it to be
397 compatible with the given WIDTH. */
399 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
401 assert (fmt_check_output (&fmt));
402 assert (sfm_width_to_segments (width) == 1);
405 fmt_resize (&fmt, width);
406 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
409 /* Write a string continuation variable record for each 8-byte
410 section beyond the initial 8 bytes, for a variable of the
413 write_variable_continuation_records (struct sfm_writer *w, int width)
417 assert (sfm_width_to_segments (width) == 1);
418 for (position = 8; position < width; position += 8)
420 write_int (w, 2); /* Record type. */
421 write_int (w, -1); /* Width. */
422 write_int (w, 0); /* No variable label. */
423 write_int (w, 0); /* No missing values. */
424 write_int (w, 0); /* Print format. */
425 write_int (w, 0); /* Write format. */
426 write_zeros (w, 8); /* Name. */
430 /* Write the variable record(s) for variable V to system file
433 write_variable (struct sfm_writer *w, const struct variable *v)
435 int width = var_get_width (v);
436 int segment_cnt = sfm_width_to_segments (width);
437 int seg0_width = sfm_segment_alloc_width (width, 0);
438 const char *encoding = var_get_encoding (v);
439 struct missing_values mv;
446 write_int (w, seg0_width);
448 /* Variable has a variable label? */
449 write_int (w, var_has_label (v));
451 /* Number of missing values. If there is a range, then the
452 range counts as 2 missing values and causes the number to be
454 mv_copy (&mv, var_get_missing_values (v));
455 if (mv_get_width (&mv) > 8)
457 if (mv_has_range (&mv))
458 write_int (w, -2 - mv_n_values (&mv));
460 write_int (w, mv_n_values (&mv));
462 /* Print and write formats. */
463 write_format (w, *var_get_print_format (v), seg0_width);
464 write_format (w, *var_get_write_format (v), seg0_width);
467 The full name is in a translation table written
469 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
472 if (var_has_label (v))
474 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
475 size_t label_len = MIN (strlen (label), 255);
476 size_t padded_len = ROUND_UP (label_len, 4);
477 write_int (w, label_len);
478 write_string (w, label, padded_len);
482 /* Write the missing values, if any, range first. */
483 if (mv_has_range (&mv))
486 mv_get_range (&mv, &x, &y);
490 for (i = 0; i < mv_n_values (&mv); i++)
491 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
493 write_variable_continuation_records (w, seg0_width);
495 /* Write additional segments for very long string variables. */
496 for (i = 1; i < segment_cnt; i++)
498 int seg_width = sfm_segment_alloc_width (width, i);
499 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
501 write_int (w, 2); /* Variable record. */
502 write_int (w, seg_width); /* Width. */
503 write_int (w, 0); /* No variable label. */
504 write_int (w, 0); /* No missing values. */
505 write_format (w, fmt, seg_width); /* Print format. */
506 write_format (w, fmt, seg_width); /* Write format. */
507 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
509 write_variable_continuation_records (w, seg_width);
515 /* Writes the value labels for variable V having system file
516 variable index IDX to system file W.
518 Value labels for long string variables are written separately,
519 by write_long_string_value_labels. */
521 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
523 const struct val_labs *val_labs;
524 const struct val_lab **labels;
528 val_labs = var_get_value_labels (v);
529 n_labels = val_labs_count (val_labs);
530 if (n_labels == 0 || var_get_width (v) > 8)
533 /* Value label record. */
534 write_int (w, 3); /* Record type. */
535 write_int (w, val_labs_count (val_labs));
536 labels = val_labs_sorted (val_labs);
537 for (i = 0; i < n_labels; i++)
539 const struct val_lab *vl = labels[i];
540 char *label = recode_string (var_get_encoding (v), UTF8,
541 val_lab_get_escaped_label (vl), -1);
542 uint8_t len = MIN (strlen (label), 255);
544 write_value (w, val_lab_get_value (vl), var_get_width (v));
545 write_bytes (w, &len, 1);
546 write_bytes (w, label, len);
547 write_zeros (w, REM_RND_UP (len + 1, 8));
552 /* Value label variable record. */
553 write_int (w, 4); /* Record type. */
554 write_int (w, 1); /* Number of variables. */
555 write_int (w, idx + 1); /* Variable's dictionary index. */
558 /* Writes record type 6, document record. */
560 write_documents (struct sfm_writer *w, const struct dictionary *d)
562 const struct string_array *docs = dict_get_documents (d);
563 const char *enc = dict_get_encoding (d);
566 write_int (w, 6); /* Record type. */
567 write_int (w, docs->n);
568 for (i = 0; i < docs->n; i++)
570 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
571 size_t s_len = strlen (s);
572 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
574 write_bytes (w, s, write_len);
575 write_spaces (w, DOC_LINE_LENGTH - write_len);
581 put_attrset (struct string *string, const struct attrset *attrs)
583 const struct attribute *attr;
584 struct attrset_iterator i;
586 for (attr = attrset_first (attrs, &i); attr != NULL;
587 attr = attrset_next (attrs, &i))
589 size_t n_values = attribute_get_n_values (attr);
592 ds_put_cstr (string, attribute_get_name (attr));
593 ds_put_byte (string, '(');
594 for (j = 0; j < n_values; j++)
595 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
596 ds_put_byte (string, ')');
601 write_data_file_attributes (struct sfm_writer *w,
602 const struct dictionary *d)
604 struct string s = DS_EMPTY_INITIALIZER;
605 put_attrset (&s, dict_get_attributes (d));
606 write_utf8_record (w, dict_get_encoding (d), &s, 17);
611 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
613 struct string s = DS_EMPTY_INITIALIZER;
614 size_t n_vars = dict_get_var_cnt (d);
615 size_t n_attrsets = 0;
618 for (i = 0; i < n_vars; i++)
620 struct variable *v = dict_get_var (d, i);
621 struct attrset *attrs = var_get_attributes (v);
622 if (attrset_count (attrs))
625 ds_put_byte (&s, '/');
626 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
627 put_attrset (&s, attrs);
631 write_utf8_record (w, dict_get_encoding (d), &s, 18);
635 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
636 SPSS before release 14, otherwise writes sets supported only by later
639 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
642 struct string s = DS_EMPTY_INITIALIZER;
646 n_mrsets = dict_get_n_mrsets (dict);
650 for (i = 0; i < n_mrsets; i++)
652 const struct mrset *mrset = dict_get_mrset (dict, i);
656 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
660 ds_put_format (&s, "%s=", mrset->name);
661 if (mrset->type == MRSET_MD)
665 if (mrset->cat_source == MRSET_COUNTEDVALUES)
666 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
668 ds_put_byte (&s, 'D');
670 if (mrset->width == 0)
671 counted = xasprintf ("%.0f", mrset->counted.f);
673 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
675 ds_put_format (&s, "%zu %s", strlen (counted), counted);
679 ds_put_byte (&s, 'C');
680 ds_put_byte (&s, ' ');
682 label = mrset->label && !mrset->label_from_var_label ? mrset->label : "";
683 ds_put_format (&s, "%zu %s", strlen (label), label);
685 for (j = 0; j < mrset->n_vars; j++)
686 ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0));
687 ds_put_byte (&s, '\n');
689 write_utf8_record (w, dict_get_encoding (dict), &s, pre_v14 ? 7 : 19);
693 /* Write the alignment, width and scale values. */
695 write_variable_display_parameters (struct sfm_writer *w,
696 const struct dictionary *dict)
700 write_int (w, 7); /* Record type. */
701 write_int (w, 11); /* Record subtype. */
702 write_int (w, 4); /* Data item (int32) size. */
703 write_int (w, w->segment_cnt * 3); /* Number of data items. */
705 for (i = 0; i < dict_get_var_cnt (dict); ++i)
707 struct variable *v = dict_get_var (dict, i);
708 int width = var_get_width (v);
709 int segment_cnt = sfm_width_to_segments (width);
710 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
711 : var_get_measure (v) == MEASURE_ORDINAL ? 2
713 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
714 : var_get_alignment (v) == ALIGN_RIGHT ? 1
718 for (i = 0; i < segment_cnt; i++)
720 int width_left = width - sfm_segment_effective_offset (width, i);
721 write_int (w, measure);
722 write_int (w, (i == 0 ? var_get_display_width (v)
723 : var_default_display_width (width_left)));
724 write_int (w, alignment);
729 /* Writes the table of lengths for very long string variables. */
731 write_vls_length_table (struct sfm_writer *w,
732 const struct dictionary *dict)
737 ds_init_empty (&map);
738 for (i = 0; i < dict_get_var_cnt (dict); ++i)
740 const struct variable *v = dict_get_var (dict, i);
741 if (sfm_width_to_segments (var_get_width (v)) > 1)
742 ds_put_format (&map, "%s=%05d%c\t",
743 var_get_short_name (v, 0), var_get_width (v), 0);
745 if (!ds_is_empty (&map))
746 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
752 write_long_string_value_labels (struct sfm_writer *w,
753 const struct dictionary *dict)
755 size_t n_vars = dict_get_var_cnt (dict);
759 /* Figure out the size in advance. */
761 for (i = 0; i < n_vars; i++)
763 struct variable *var = dict_get_var (dict, i);
764 const struct val_labs *val_labs = var_get_value_labels (var);
765 const char *encoding = var_get_encoding (var);
766 int width = var_get_width (var);
767 const struct val_lab *val_lab;
769 if (val_labs_count (val_labs) == 0 || width < 9)
773 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
774 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
775 val_lab = val_labs_next (val_labs, val_lab))
778 size += recode_string_len (encoding, "UTF-8",
779 val_lab_get_escaped_label (val_lab), -1);
785 write_int (w, 7); /* Record type. */
786 write_int (w, 21); /* Record subtype */
787 write_int (w, 1); /* Data item (byte) size. */
788 write_int (w, size); /* Number of data items. */
790 start = ftello (w->file);
791 for (i = 0; i < n_vars; i++)
793 struct variable *var = dict_get_var (dict, i);
794 const struct val_labs *val_labs = var_get_value_labels (var);
795 const char *encoding = var_get_encoding (var);
796 int width = var_get_width (var);
797 const struct val_lab *val_lab;
800 if (val_labs_count (val_labs) == 0 || width < 9)
803 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
804 write_int (w, strlen (var_name));
805 write_bytes (w, var_name, strlen (var_name));
808 write_int (w, width);
809 write_int (w, val_labs_count (val_labs));
810 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
811 val_lab = val_labs_next (val_labs, val_lab))
816 write_int (w, width);
817 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
820 label = recode_string (var_get_encoding (var), "UTF-8",
821 val_lab_get_escaped_label (val_lab), -1);
822 len = strlen (label);
824 write_bytes (w, label, len);
828 assert (ftello (w->file) == start + size);
832 write_encoding_record (struct sfm_writer *w,
833 const struct dictionary *d)
835 const char *enc = dict_get_encoding (d);
840 write_int (w, 7); /* Record type. */
841 write_int (w, 20); /* Record subtype. */
842 write_int (w, 1); /* Data item (char) size. */
844 /* IANA says "...character set names may be up to 40 characters taken from
845 the printable characters of US-ASCII," so character set names don't need
847 write_int (w, strlen (enc)); /* Number of data items. */
848 write_string (w, enc, strlen (enc));
852 /* Writes the long variable name table. */
854 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
859 ds_init_empty (&map);
860 for (i = 0; i < dict_get_var_cnt (dict); i++)
862 struct variable *v = dict_get_var (dict, i);
864 ds_put_byte (&map, '\t');
865 ds_put_format (&map, "%s=%s",
866 var_get_short_name (v, 0), var_get_name (v));
868 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
872 /* Write integer information record. */
874 write_integer_info_record (struct sfm_writer *w)
876 int version_component[3];
879 /* Parse the version string. */
880 memset (version_component, 0, sizeof version_component);
881 sscanf (bare_version, "%d.%d.%d",
882 &version_component[0], &version_component[1], &version_component[2]);
884 /* Figure out the floating-point format. */
885 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
886 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
888 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
890 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
896 write_int (w, 7); /* Record type. */
897 write_int (w, 3); /* Record subtype. */
898 write_int (w, 4); /* Data item (int32) size. */
899 write_int (w, 8); /* Number of data items. */
900 write_int (w, version_component[0]);
901 write_int (w, version_component[1]);
902 write_int (w, version_component[2]);
903 write_int (w, -1); /* Machine code. */
904 write_int (w, float_format);
905 write_int (w, 1); /* Compression code. */
906 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
907 write_int (w, 2); /* 7-bit ASCII. */
910 /* Write floating-point information record. */
912 write_float_info_record (struct sfm_writer *w)
914 write_int (w, 7); /* Record type. */
915 write_int (w, 4); /* Record subtype. */
916 write_int (w, 8); /* Data item (flt64) size. */
917 write_int (w, 3); /* Number of data items. */
918 write_float (w, SYSMIS); /* System-missing value. */
919 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
920 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
923 /* Writes case C to system file W. */
925 sys_file_casewriter_write (struct casewriter *writer, void *w_,
928 struct sfm_writer *w = w_;
930 if (ferror (w->file))
932 casewriter_force_error (writer);
940 write_case_uncompressed (w, c);
942 write_case_compressed (w, c);
947 /* Destroys system file writer W. */
949 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
951 struct sfm_writer *w = w_;
952 if (!close_writer (w))
953 casewriter_force_error (writer);
956 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
958 write_error (const struct sfm_writer *writer)
960 return ferror (writer->file);
963 /* Closes a system file after we're done with it.
964 Returns true if successful, false if an I/O error occurred. */
966 close_writer (struct sfm_writer *w)
977 if (w->opcode_cnt > 0)
978 flush_compressed (w);
981 ok = !write_error (w);
983 /* Seek back to the beginning and update the number of cases.
984 This is just a courtesy to later readers, so there's no need
985 to check return values or report errors. */
986 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
988 write_int (w, w->case_cnt);
992 if (fclose (w->file) == EOF)
996 msg (ME, _("An I/O error occurred writing system file `%s'."),
997 fh_get_file_name (w->fh));
999 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1003 fh_unlock (w->lock);
1012 /* System file writer casewriter class. */
1013 static const struct casewriter_class sys_file_casewriter_class =
1015 sys_file_casewriter_write,
1016 sys_file_casewriter_destroy,
1020 /* Writes case C to system file W, without compressing it. */
1022 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1026 for (i = 0; i < w->sfm_var_cnt; i++)
1028 struct sfm_var *v = &w->sfm_vars[i];
1030 if (v->var_width == 0)
1031 write_float (w, case_num_idx (c, v->case_index));
1034 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1036 write_spaces (w, v->padding);
1041 /* Writes case C to system file W, with compression. */
1043 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1047 for (i = 0; i < w->sfm_var_cnt; i++)
1049 struct sfm_var *v = &w->sfm_vars[i];
1051 if (v->var_width == 0)
1053 double d = case_num_idx (c, v->case_index);
1055 put_cmp_opcode (w, 255);
1056 else if (d >= 1 - COMPRESSION_BIAS
1057 && d <= 251 - COMPRESSION_BIAS
1059 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1062 put_cmp_opcode (w, 253);
1063 put_cmp_number (w, d);
1068 int offset = v->offset;
1071 /* This code properly deals with a width that is not a
1072 multiple of 8, by ensuring that the final partial
1073 oct (8 byte unit) is treated as padded with spaces
1075 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1077 const void *data = case_str_idx (c, v->case_index) + offset;
1078 int chunk_size = MIN (width, 8);
1079 if (!memcmp (data, " ", chunk_size))
1080 put_cmp_opcode (w, 254);
1083 put_cmp_opcode (w, 253);
1084 put_cmp_string (w, data, chunk_size);
1088 /* This code deals properly with padding that is not a
1089 multiple of 8 bytes, by discarding the remainder,
1090 which was already effectively padded with spaces in
1091 the previous loop. (Note that v->width + v->padding
1092 is always a multiple of 8.) */
1093 for (padding = v->padding / 8; padding > 0; padding--)
1094 put_cmp_opcode (w, 254);
1099 /* Flushes buffered compressed opcodes and data to W.
1100 The compression buffer must not be empty. */
1102 flush_compressed (struct sfm_writer *w)
1104 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1106 write_bytes (w, w->opcodes, w->opcode_cnt);
1107 write_zeros (w, 8 - w->opcode_cnt);
1109 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1111 w->opcode_cnt = w->data_cnt = 0;
1114 /* Appends OPCODE to the buffered set of compression opcodes in
1115 W. Flushes the compression buffer beforehand if necessary. */
1117 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1119 if (w->opcode_cnt >= 8)
1120 flush_compressed (w);
1122 w->opcodes[w->opcode_cnt++] = opcode;
1125 /* Appends NUMBER to the buffered compression data in W. The
1126 buffer must not be full; the way to assure that is to call
1127 this function only just after a call to put_cmp_opcode, which
1128 will flush the buffer as necessary. */
1130 put_cmp_number (struct sfm_writer *w, double number)
1132 assert (w->opcode_cnt > 0);
1133 assert (w->data_cnt < 8);
1135 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1138 /* Appends SIZE bytes of DATA to the buffered compression data in
1139 W, followed by enough spaces to pad the output data to exactly
1140 8 bytes (thus, SIZE must be no greater than 8). The buffer
1141 must not be full; the way to assure that is to call this
1142 function only just after a call to put_cmp_opcode, which will
1143 flush the buffer as necessary. */
1145 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1147 assert (w->opcode_cnt > 0);
1148 assert (w->data_cnt < 8);
1151 memset (w->data[w->data_cnt], ' ', 8);
1152 memcpy (w->data[w->data_cnt], data, size);
1156 /* Writes 32-bit integer X to the output file for writer W. */
1158 write_int (struct sfm_writer *w, int32_t x)
1160 write_bytes (w, &x, sizeof x);
1163 /* Converts NATIVE to the 64-bit format used in output files in
1166 convert_double_to_output_format (double native, uint8_t output[8])
1168 /* If "double" is not a 64-bit type, then convert it to a
1169 64-bit type. Otherwise just copy it. */
1170 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1171 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1173 memcpy (output, &native, sizeof native);
1176 /* Writes floating-point number X to the output file for writer
1179 write_float (struct sfm_writer *w, double x)
1182 convert_double_to_output_format (x, output);
1183 write_bytes (w, output, sizeof output);
1186 /* Writes contents of VALUE with the given WIDTH to W, padding
1187 with zeros to a multiple of 8 bytes.
1188 To avoid a branch, and because we don't actually need to
1189 support it, WIDTH must be no bigger than 8. */
1191 write_value (struct sfm_writer *w, const union value *value, int width)
1193 assert (width <= 8);
1195 write_float (w, value->f);
1198 write_bytes (w, value_str (value, width), width);
1199 write_zeros (w, 8 - width);
1203 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1204 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1205 is padded on the right with spaces. */
1207 write_string (struct sfm_writer *w, const char *string, size_t width)
1209 size_t data_bytes = MIN (strlen (string), width);
1210 size_t pad_bytes = width - data_bytes;
1211 write_bytes (w, string, data_bytes);
1212 while (pad_bytes-- > 0)
1213 putc (' ', w->file);
1216 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1217 recoded version in a field of the given WIDTH to W. The string is truncated
1218 or padded on the right with spaces to exactly WIDTH bytes. */
1220 write_utf8_string (struct sfm_writer *w, const char *encoding,
1221 const char *string, size_t width)
1223 char *s = recode_string (encoding, "UTF-8", string, -1);
1224 write_string (w, s, width);
1228 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1229 from UTF-8 encoded into ENCODING. */
1231 write_utf8_record (struct sfm_writer *w, const char *encoding,
1232 const struct string *content, int subtype)
1236 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1238 write_int (w, subtype);
1240 write_int (w, ss_length (s));
1241 write_bytes (w, ss_data (s), ss_length (s));
1245 /* Writes SIZE bytes of DATA to W's output file. */
1247 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1249 fwrite (data, 1, size, w->file);
1252 /* Writes N zeros to W's output file. */
1254 write_zeros (struct sfm_writer *w, size_t n)
1260 /* Writes N spaces to W's output file. */
1262 write_spaces (struct sfm_writer *w, size_t n)
1265 putc (' ', w->file);