1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2007, 2009, 2010 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/version.h"
52 #include "gl/xmemdup0.h"
53 #include "gl/minmax.h"
54 #include "gl/unlocked-io.h"
55 #include "gl/xalloc.h"
58 #define _(msgid) gettext (msgid)
59 #define N_(msgid) (msgid)
61 /* Compression bias used by PSPP. Values between (1 -
62 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
64 #define COMPRESSION_BIAS 100
66 /* System file writer. */
69 struct file_handle *fh; /* File handle. */
70 struct fh_lock *lock; /* Mutual exclusion for file. */
71 FILE *file; /* File stream. */
72 struct replace_file *rf; /* Ticket for replacing output file. */
74 bool compress; /* 1=compressed, 0=not compressed. */
75 casenumber case_cnt; /* Number of cases written so far. */
77 /* Compression buffering.
79 Compressed data is output as groups of 8 1-byte opcodes
80 followed by up to 8 (depending on the opcodes) 8-byte data
81 items. Data items and opcodes arrive at the same time but
82 must be reordered for writing to disk, thus a small amount
84 uint8_t opcodes[8]; /* Buffered opcodes. */
85 int opcode_cnt; /* Number of buffered opcodes. */
86 uint8_t data[8][8]; /* Buffered data. */
87 int data_cnt; /* Number of buffered data items. */
90 struct sfm_var *sfm_vars; /* Variables. */
91 size_t sfm_var_cnt; /* Number of variables. */
92 size_t segment_cnt; /* Number of variables including extra segments
93 for long string variables. */
96 static const struct casewriter_class sys_file_casewriter_class;
98 static void write_header (struct sfm_writer *, const struct dictionary *);
99 static void write_variable (struct sfm_writer *, const struct variable *);
100 static void write_value_labels (struct sfm_writer *, struct variable *,
102 static void write_integer_info_record (struct sfm_writer *);
103 static void write_float_info_record (struct sfm_writer *);
105 static void write_longvar_table (struct sfm_writer *w,
106 const struct dictionary *dict);
108 static void write_encoding_record (struct sfm_writer *w,
109 const struct dictionary *);
111 static void write_vls_length_table (struct sfm_writer *w,
112 const struct dictionary *dict);
114 static void write_long_string_value_labels (struct sfm_writer *,
115 const struct dictionary *);
117 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
120 static void write_variable_display_parameters (struct sfm_writer *w,
121 const struct dictionary *dict);
123 static void write_documents (struct sfm_writer *, const struct dictionary *);
125 static void write_data_file_attributes (struct sfm_writer *,
126 const struct dictionary *);
127 static void write_variable_attributes (struct sfm_writer *,
128 const struct dictionary *);
130 static void write_int (struct sfm_writer *, int32_t);
131 static inline void convert_double_to_output_format (double, uint8_t[8]);
132 static void write_float (struct sfm_writer *, double);
133 static void write_string (struct sfm_writer *, const char *, size_t);
134 static void write_bytes (struct sfm_writer *, const void *, size_t);
135 static void write_zeros (struct sfm_writer *, size_t);
136 static void write_spaces (struct sfm_writer *, size_t);
137 static void write_value (struct sfm_writer *, const union value *, int width);
139 static void write_case_uncompressed (struct sfm_writer *,
140 const struct ccase *);
141 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
142 static void flush_compressed (struct sfm_writer *);
143 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
144 static void put_cmp_number (struct sfm_writer *, double);
145 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
147 bool write_error (const struct sfm_writer *);
148 bool close_writer (struct sfm_writer *);
150 /* Returns default options for writing a system file. */
151 struct sfm_write_options
152 sfm_writer_default_options (void)
154 struct sfm_write_options opts;
155 opts.create_writeable = true;
156 opts.compress = settings_get_scompression ();
161 /* Opens the system file designated by file handle FH for writing
162 cases from dictionary D according to the given OPTS. If
163 COMPRESS is nonzero, the system file will be compressed.
165 No reference to D is retained, so it may be modified or
166 destroyed at will after this function returns. D is not
167 modified by this function, except to assign short names. */
169 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
170 struct sfm_write_options opts)
172 struct sfm_writer *w;
178 if (opts.version != 2 && opts.version != 3)
180 msg (ME, _("Unknown system file version %d. Treating as version %d."),
185 /* Create and initialize writer. */
186 w = xmalloc (sizeof *w);
192 w->compress = opts.compress;
195 w->opcode_cnt = w->data_cnt = 0;
197 /* Figure out how to map in-memory case data to on-disk case
198 data. Also count the number of segments. Very long strings
199 occupy multiple segments, otherwise each variable only takes
201 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
204 /* Open file handle as an exclusive writer. */
205 /* TRANSLATORS: this fragment will be interpolated into
206 messages in fh_lock() that identify types of files. */
207 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
211 /* Create the file on disk. */
212 mode = S_IRUSR | S_IRGRP | S_IROTH;
213 if (opts.create_writeable)
214 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
215 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
219 msg (ME, _("Error opening \"%s\" for writing as a system file: %s."),
220 fh_get_file_name (fh), strerror (errno));
224 /* Write the file header. */
227 /* Write basic variable info. */
228 short_names_assign (d);
229 for (i = 0; i < dict_get_var_cnt (d); i++)
230 write_variable (w, dict_get_var (d, i));
232 /* Write out value labels. */
234 for (i = 0; i < dict_get_var_cnt (d); i++)
236 struct variable *v = dict_get_var (d, i);
238 write_value_labels (w, v, idx);
239 idx += sfm_width_to_octs (var_get_width (v));
242 if (dict_get_documents (d) != NULL)
243 write_documents (w, d);
245 write_integer_info_record (w);
246 write_float_info_record (w);
248 write_mrsets (w, d, true);
250 write_variable_display_parameters (w, d);
252 if (opts.version >= 3)
253 write_longvar_table (w, d);
255 write_vls_length_table (w, d);
257 write_long_string_value_labels (w, d);
259 if (attrset_count (dict_get_attributes (d)))
260 write_data_file_attributes (w, d);
261 write_variable_attributes (w, d);
263 write_mrsets (w, d, false);
265 write_encoding_record (w, d);
267 /* Write end-of-headers record. */
277 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
284 /* Returns value of X truncated to two least-significant digits. */
295 /* Calculates the offset of data for TARGET_VAR from the
296 beginning of each case's data for dictionary D. The return
297 value is in "octs" (8-byte units). */
299 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
305 for (i = 0; i < dict_get_var_cnt (d); i++)
307 struct variable *var = dict_get_var (d, i);
308 if (var == target_var)
310 oct_idx += sfm_width_to_octs (var_get_width (var));
315 /* Write the sysfile_header header to system file W. */
317 write_header (struct sfm_writer *w, const struct dictionary *d)
320 char creation_date[10];
321 char creation_time[9];
322 const char *file_label;
323 struct variable *weight;
327 /* Record-type code. */
328 write_string (w, "$FL2", 4);
330 /* Product identification. */
331 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
332 version, host_system);
333 write_string (w, prod_name, 60);
338 /* Number of `union value's per case. */
339 write_int (w, calc_oct_idx (d, NULL));
342 write_int (w, w->compress);
344 /* Weight variable. */
345 weight = dict_get_weight (d);
346 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
348 /* Number of cases. We don't know this in advance, so we write
349 -1 to indicate an unknown number of cases. Later we can
350 come back and overwrite it with the true value. */
353 /* Compression bias. */
354 write_float (w, COMPRESSION_BIAS);
356 /* Creation date and time. */
357 if (time (&t) == (time_t) -1)
359 strcpy (creation_date, "01 Jan 70");
360 strcpy (creation_time, "00:00:00");
364 static const char *const month_name[12] =
366 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
367 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
369 struct tm *tmp = localtime (&t);
370 int day = rerange (tmp->tm_mday);
371 int mon = rerange (tmp->tm_mon + 1);
372 int year = rerange (tmp->tm_year);
373 int hour = rerange (tmp->tm_hour + 1);
374 int min = rerange (tmp->tm_min + 1);
375 int sec = rerange (tmp->tm_sec + 1);
377 snprintf (creation_date, sizeof creation_date,
378 "%02d %s %02d", day, month_name[mon - 1], year);
379 snprintf (creation_time, sizeof creation_time,
380 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
382 write_string (w, creation_date, 9);
383 write_string (w, creation_time, 8);
386 file_label = dict_get_label (d);
387 if (file_label == NULL)
389 write_string (w, file_label, 64);
395 /* Write format spec FMT to W, after adjusting it to be
396 compatible with the given WIDTH. */
398 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
400 assert (fmt_check_output (&fmt));
401 assert (sfm_width_to_segments (width) == 1);
404 fmt_resize (&fmt, width);
405 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
408 /* Write a string continuation variable record for each 8-byte
409 section beyond the initial 8 bytes, for a variable of the
412 write_variable_continuation_records (struct sfm_writer *w, int width)
416 assert (sfm_width_to_segments (width) == 1);
417 for (position = 8; position < width; position += 8)
419 write_int (w, 2); /* Record type. */
420 write_int (w, -1); /* Width. */
421 write_int (w, 0); /* No variable label. */
422 write_int (w, 0); /* No missing values. */
423 write_int (w, 0); /* Print format. */
424 write_int (w, 0); /* Write format. */
425 write_zeros (w, 8); /* Name. */
429 /* Write the variable record(s) for variable V to system file
432 write_variable (struct sfm_writer *w, const struct variable *v)
434 int width = var_get_width (v);
435 int segment_cnt = sfm_width_to_segments (width);
436 int seg0_width = sfm_segment_alloc_width (width, 0);
437 struct missing_values mv;
444 write_int (w, seg0_width);
446 /* Variable has a variable label? */
447 write_int (w, var_has_label (v));
449 /* Number of missing values. If there is a range, then the
450 range counts as 2 missing values and causes the number to be
452 mv_copy (&mv, var_get_missing_values (v));
453 if (mv_get_width (&mv) > 8)
455 if (mv_has_range (&mv))
456 write_int (w, -2 - mv_n_values (&mv));
458 write_int (w, mv_n_values (&mv));
460 /* Print and write formats. */
461 write_format (w, *var_get_print_format (v), seg0_width);
462 write_format (w, *var_get_write_format (v), seg0_width);
465 The full name is in a translation table written
467 write_string (w, var_get_short_name (v, 0), 8);
470 if (var_has_label (v))
472 char *label = recode_string (var_get_encoding (v), UTF8, var_get_label (v), -1);
473 size_t label_len = MIN (strlen (label), 255);
474 size_t padded_len = ROUND_UP (label_len, 4);
475 write_int (w, label_len);
476 write_string (w, label, padded_len);
480 /* Write the missing values, if any, range first. */
481 if (mv_has_range (&mv))
484 mv_get_range (&mv, &x, &y);
488 for (i = 0; i < mv_n_values (&mv); i++)
489 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
491 write_variable_continuation_records (w, seg0_width);
493 /* Write additional segments for very long string variables. */
494 for (i = 1; i < segment_cnt; i++)
496 int seg_width = sfm_segment_alloc_width (width, i);
497 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
499 write_int (w, 2); /* Variable record. */
500 write_int (w, seg_width); /* Width. */
501 write_int (w, 0); /* No variable label. */
502 write_int (w, 0); /* No missing values. */
503 write_format (w, fmt, seg_width); /* Print format. */
504 write_format (w, fmt, seg_width); /* Write format. */
505 write_string (w, var_get_short_name (v, i), 8);
507 write_variable_continuation_records (w, seg_width);
513 /* Writes the value labels for variable V having system file
514 variable index IDX to system file W.
516 Value labels for long string variables are written separately,
517 by write_long_string_value_labels. */
519 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
521 const struct val_labs *val_labs;
522 const struct val_lab **labels;
526 val_labs = var_get_value_labels (v);
527 n_labels = val_labs_count (val_labs);
528 if (n_labels == 0 || var_get_width (v) > 8)
531 /* Value label record. */
532 write_int (w, 3); /* Record type. */
533 write_int (w, val_labs_count (val_labs));
534 labels = val_labs_sorted (val_labs);
535 for (i = 0; i < n_labels; i++)
537 const struct val_lab *vl = labels[i];
538 char *label = recode_string (var_get_encoding (v), UTF8, val_lab_get_label (vl), -1);
539 uint8_t len = MIN (strlen (label), 255);
541 write_value (w, val_lab_get_value (vl), var_get_width (v));
542 write_bytes (w, &len, 1);
543 write_bytes (w, label, len);
544 write_zeros (w, REM_RND_UP (len + 1, 8));
549 /* Value label variable record. */
550 write_int (w, 4); /* Record type. */
551 write_int (w, 1); /* Number of variables. */
552 write_int (w, idx + 1); /* Variable's dictionary index. */
555 /* Writes record type 6, document record. */
557 write_documents (struct sfm_writer *w, const struct dictionary *d)
559 size_t line_cnt = dict_get_document_line_cnt (d);
561 write_int (w, 6); /* Record type. */
562 write_int (w, line_cnt);
563 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
567 put_attrset (struct string *string, const struct attrset *attrs)
569 const struct attribute *attr;
570 struct attrset_iterator i;
572 for (attr = attrset_first (attrs, &i); attr != NULL;
573 attr = attrset_next (attrs, &i))
575 size_t n_values = attribute_get_n_values (attr);
578 ds_put_cstr (string, attribute_get_name (attr));
579 ds_put_char (string, '(');
580 for (j = 0; j < n_values; j++)
581 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
582 ds_put_char (string, ')');
587 write_attribute_record (struct sfm_writer *w, const struct string *content,
591 write_int (w, subtype);
593 write_int (w, ds_length (content));
594 write_bytes (w, ds_data (content), ds_length (content));
598 write_data_file_attributes (struct sfm_writer *w,
599 const struct dictionary *d)
601 struct string s = DS_EMPTY_INITIALIZER;
602 put_attrset (&s, dict_get_attributes (d));
603 write_attribute_record (w, &s, 17);
608 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
610 struct string s = DS_EMPTY_INITIALIZER;
611 size_t n_vars = dict_get_var_cnt (d);
612 size_t n_attrsets = 0;
615 for (i = 0; i < n_vars; i++)
617 struct variable *v = dict_get_var (d, i);
618 struct attrset *attrs = var_get_attributes (v);
619 if (attrset_count (attrs))
622 ds_put_char (&s, '/');
623 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
624 put_attrset (&s, attrs);
628 write_attribute_record (w, &s, 18);
632 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
633 SPSS before release 14, otherwise writes sets supported only by later
636 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
639 struct string s = DS_EMPTY_INITIALIZER;
643 n_mrsets = dict_get_n_mrsets (dict);
647 for (i = 0; i < n_mrsets; i++)
649 const struct mrset *mrset = dict_get_mrset (dict, i);
653 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
657 ds_put_format (&s, "%s=", mrset->name);
658 if (mrset->type == MRSET_MD)
662 if (mrset->cat_source == MRSET_COUNTEDVALUES)
663 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
665 ds_put_char (&s, 'D');
667 if (mrset->width == 0)
668 counted = xasprintf ("%.0f", mrset->counted.f);
670 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
672 ds_put_format (&s, "%zu %s", strlen (counted), counted);
676 ds_put_char (&s, 'C');
677 ds_put_char (&s, ' ');
679 label = mrset->label && !mrset->label_from_var_label ? mrset->label : "";
680 ds_put_format (&s, "%zu %s", strlen (label), label);
682 for (j = 0; j < mrset->n_vars; j++)
683 ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0));
684 ds_put_char (&s, '\n');
686 write_attribute_record (w, &s, 7);
690 /* Write the alignment, width and scale values. */
692 write_variable_display_parameters (struct sfm_writer *w,
693 const struct dictionary *dict)
697 write_int (w, 7); /* Record type. */
698 write_int (w, 11); /* Record subtype. */
699 write_int (w, 4); /* Data item (int32) size. */
700 write_int (w, w->segment_cnt * 3); /* Number of data items. */
702 for (i = 0; i < dict_get_var_cnt (dict); ++i)
704 struct variable *v = dict_get_var (dict, i);
705 int width = var_get_width (v);
706 int segment_cnt = sfm_width_to_segments (width);
707 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
708 : var_get_measure (v) == MEASURE_ORDINAL ? 2
710 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
711 : var_get_alignment (v) == ALIGN_RIGHT ? 1
715 for (i = 0; i < segment_cnt; i++)
717 int width_left = width - sfm_segment_effective_offset (width, i);
718 write_int (w, measure);
719 write_int (w, (i == 0 ? var_get_display_width (v)
720 : var_default_display_width (width_left)));
721 write_int (w, alignment);
726 /* Writes the table of lengths for very long string variables. */
728 write_vls_length_table (struct sfm_writer *w,
729 const struct dictionary *dict)
734 ds_init_empty (&map);
735 for (i = 0; i < dict_get_var_cnt (dict); ++i)
737 const struct variable *v = dict_get_var (dict, i);
738 if (sfm_width_to_segments (var_get_width (v)) > 1)
739 ds_put_format (&map, "%s=%05d%c\t",
740 var_get_short_name (v, 0), var_get_width (v), 0);
742 if (!ds_is_empty (&map))
744 write_int (w, 7); /* Record type. */
745 write_int (w, 14); /* Record subtype. */
746 write_int (w, 1); /* Data item (char) size. */
747 write_int (w, ds_length (&map)); /* Number of data items. */
748 write_bytes (w, ds_data (&map), ds_length (&map));
755 write_long_string_value_labels (struct sfm_writer *w,
756 const struct dictionary *dict)
758 size_t n_vars = dict_get_var_cnt (dict);
762 /* Figure out the size in advance. */
764 for (i = 0; i < n_vars; i++)
766 struct variable *var = dict_get_var (dict, i);
767 const struct val_labs *val_labs = var_get_value_labels (var);
768 int width = var_get_width (var);
769 const struct val_lab *val_lab;
771 if (val_labs_count (val_labs) == 0 || width < 9)
774 size += 12 + strlen (var_get_name (var));
775 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
776 val_lab = val_labs_next (val_labs, val_lab))
777 size += 8 + width + strlen (val_lab_get_label (val_lab));
782 write_int (w, 7); /* Record type. */
783 write_int (w, 21); /* Record subtype */
784 write_int (w, 1); /* Data item (byte) size. */
785 write_int (w, size); /* Number of data items. */
787 start = ftello (w->file);
788 for (i = 0; i < n_vars; i++)
790 struct variable *var = dict_get_var (dict, i);
791 const struct val_labs *val_labs = var_get_value_labels (var);
792 const char *var_name = var_get_name (var);
793 int width = var_get_width (var);
794 const struct val_lab *val_lab;
796 if (val_labs_count (val_labs) == 0 || width < 9)
799 write_int (w, strlen (var_name));
800 write_bytes (w, var_name, strlen (var_name));
801 write_int (w, width);
802 write_int (w, val_labs_count (val_labs));
803 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
804 val_lab = val_labs_next (val_labs, val_lab))
806 const char *label = val_lab_get_label (val_lab);
807 size_t label_length = strlen (label);
809 write_int (w, width);
810 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
812 write_int (w, label_length);
813 write_bytes (w, label, label_length);
816 assert (ftello (w->file) == start + size);
820 write_encoding_record (struct sfm_writer *w,
821 const struct dictionary *d)
823 const char *enc = dict_get_encoding (d);
828 write_int (w, 7); /* Record type. */
829 write_int (w, 20); /* Record subtype. */
830 write_int (w, 1); /* Data item (char) size. */
831 write_int (w, strlen (enc)); /* Number of data items. */
832 write_string (w, enc, strlen (enc));
836 /* Writes the long variable name table. */
838 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
843 ds_init_empty (&map);
844 for (i = 0; i < dict_get_var_cnt (dict); i++)
846 struct variable *v = dict_get_var (dict, i);
847 char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1);
850 ds_put_char (&map, '\t');
851 ds_put_format (&map, "%s=%s",
852 var_get_short_name (v, 0), longname);
856 write_int (w, 7); /* Record type. */
857 write_int (w, 13); /* Record subtype. */
858 write_int (w, 1); /* Data item (char) size. */
859 write_int (w, ds_length (&map)); /* Number of data items. */
860 write_bytes (w, ds_data (&map), ds_length (&map));
865 /* Write integer information record. */
867 write_integer_info_record (struct sfm_writer *w)
869 int version_component[3];
872 /* Parse the version string. */
873 memset (version_component, 0, sizeof version_component);
874 sscanf (bare_version, "%d.%d.%d",
875 &version_component[0], &version_component[1], &version_component[2]);
877 /* Figure out the floating-point format. */
878 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
879 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
881 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
883 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
889 write_int (w, 7); /* Record type. */
890 write_int (w, 3); /* Record subtype. */
891 write_int (w, 4); /* Data item (int32) size. */
892 write_int (w, 8); /* Number of data items. */
893 write_int (w, version_component[0]);
894 write_int (w, version_component[1]);
895 write_int (w, version_component[2]);
896 write_int (w, -1); /* Machine code. */
897 write_int (w, float_format);
898 write_int (w, 1); /* Compression code. */
899 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
900 write_int (w, 2); /* 7-bit ASCII. */
903 /* Write floating-point information record. */
905 write_float_info_record (struct sfm_writer *w)
907 write_int (w, 7); /* Record type. */
908 write_int (w, 4); /* Record subtype. */
909 write_int (w, 8); /* Data item (flt64) size. */
910 write_int (w, 3); /* Number of data items. */
911 write_float (w, SYSMIS); /* System-missing value. */
912 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
913 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
916 /* Writes case C to system file W. */
918 sys_file_casewriter_write (struct casewriter *writer, void *w_,
921 struct sfm_writer *w = w_;
923 if (ferror (w->file))
925 casewriter_force_error (writer);
933 write_case_uncompressed (w, c);
935 write_case_compressed (w, c);
940 /* Destroys system file writer W. */
942 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
944 struct sfm_writer *w = w_;
945 if (!close_writer (w))
946 casewriter_force_error (writer);
949 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
951 write_error (const struct sfm_writer *writer)
953 return ferror (writer->file);
956 /* Closes a system file after we're done with it.
957 Returns true if successful, false if an I/O error occurred. */
959 close_writer (struct sfm_writer *w)
970 if (w->opcode_cnt > 0)
971 flush_compressed (w);
974 ok = !write_error (w);
976 /* Seek back to the beginning and update the number of cases.
977 This is just a courtesy to later readers, so there's no need
978 to check return values or report errors. */
979 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
981 write_int (w, w->case_cnt);
985 if (fclose (w->file) == EOF)
989 msg (ME, _("An I/O error occurred writing system file \"%s\"."),
990 fh_get_file_name (w->fh));
992 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1005 /* System file writer casewriter class. */
1006 static const struct casewriter_class sys_file_casewriter_class =
1008 sys_file_casewriter_write,
1009 sys_file_casewriter_destroy,
1013 /* Writes case C to system file W, without compressing it. */
1015 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1019 for (i = 0; i < w->sfm_var_cnt; i++)
1021 struct sfm_var *v = &w->sfm_vars[i];
1023 if (v->var_width == 0)
1024 write_float (w, case_num_idx (c, v->case_index));
1027 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1029 write_spaces (w, v->padding);
1034 /* Writes case C to system file W, with compression. */
1036 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1040 for (i = 0; i < w->sfm_var_cnt; i++)
1042 struct sfm_var *v = &w->sfm_vars[i];
1044 if (v->var_width == 0)
1046 double d = case_num_idx (c, v->case_index);
1048 put_cmp_opcode (w, 255);
1049 else if (d >= 1 - COMPRESSION_BIAS
1050 && d <= 251 - COMPRESSION_BIAS
1052 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1055 put_cmp_opcode (w, 253);
1056 put_cmp_number (w, d);
1061 int offset = v->offset;
1064 /* This code properly deals with a width that is not a
1065 multiple of 8, by ensuring that the final partial
1066 oct (8 byte unit) is treated as padded with spaces
1068 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1070 const void *data = case_str_idx (c, v->case_index) + offset;
1071 int chunk_size = MIN (width, 8);
1072 if (!memcmp (data, " ", chunk_size))
1073 put_cmp_opcode (w, 254);
1076 put_cmp_opcode (w, 253);
1077 put_cmp_string (w, data, chunk_size);
1081 /* This code deals properly with padding that is not a
1082 multiple of 8 bytes, by discarding the remainder,
1083 which was already effectively padded with spaces in
1084 the previous loop. (Note that v->width + v->padding
1085 is always a multiple of 8.) */
1086 for (padding = v->padding / 8; padding > 0; padding--)
1087 put_cmp_opcode (w, 254);
1092 /* Flushes buffered compressed opcodes and data to W.
1093 The compression buffer must not be empty. */
1095 flush_compressed (struct sfm_writer *w)
1097 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1099 write_bytes (w, w->opcodes, w->opcode_cnt);
1100 write_zeros (w, 8 - w->opcode_cnt);
1102 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1104 w->opcode_cnt = w->data_cnt = 0;
1107 /* Appends OPCODE to the buffered set of compression opcodes in
1108 W. Flushes the compression buffer beforehand if necessary. */
1110 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1112 if (w->opcode_cnt >= 8)
1113 flush_compressed (w);
1115 w->opcodes[w->opcode_cnt++] = opcode;
1118 /* Appends NUMBER to the buffered compression data in W. The
1119 buffer must not be full; the way to assure that is to call
1120 this function only just after a call to put_cmp_opcode, which
1121 will flush the buffer as necessary. */
1123 put_cmp_number (struct sfm_writer *w, double number)
1125 assert (w->opcode_cnt > 0);
1126 assert (w->data_cnt < 8);
1128 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1131 /* Appends SIZE bytes of DATA to the buffered compression data in
1132 W, followed by enough spaces to pad the output data to exactly
1133 8 bytes (thus, SIZE must be no greater than 8). The buffer
1134 must not be full; the way to assure that is to call this
1135 function only just after a call to put_cmp_opcode, which will
1136 flush the buffer as necessary. */
1138 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1140 assert (w->opcode_cnt > 0);
1141 assert (w->data_cnt < 8);
1144 memset (w->data[w->data_cnt], ' ', 8);
1145 memcpy (w->data[w->data_cnt], data, size);
1149 /* Writes 32-bit integer X to the output file for writer W. */
1151 write_int (struct sfm_writer *w, int32_t x)
1153 write_bytes (w, &x, sizeof x);
1156 /* Converts NATIVE to the 64-bit format used in output files in
1159 convert_double_to_output_format (double native, uint8_t output[8])
1161 /* If "double" is not a 64-bit type, then convert it to a
1162 64-bit type. Otherwise just copy it. */
1163 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1164 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1166 memcpy (output, &native, sizeof native);
1169 /* Writes floating-point number X to the output file for writer
1172 write_float (struct sfm_writer *w, double x)
1175 convert_double_to_output_format (x, output);
1176 write_bytes (w, output, sizeof output);
1179 /* Writes contents of VALUE with the given WIDTH to W, padding
1180 with zeros to a multiple of 8 bytes.
1181 To avoid a branch, and because we don't actually need to
1182 support it, WIDTH must be no bigger than 8. */
1184 write_value (struct sfm_writer *w, const union value *value, int width)
1186 assert (width <= 8);
1188 write_float (w, value->f);
1191 write_bytes (w, value_str (value, width), width);
1192 write_zeros (w, 8 - width);
1196 /* Writes null-terminated STRING in a field of the given WIDTH to
1197 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1198 is narrowed, it is padded on the right with spaces. */
1200 write_string (struct sfm_writer *w, const char *string, size_t width)
1202 size_t data_bytes = MIN (strlen (string), width);
1203 size_t pad_bytes = width - data_bytes;
1204 write_bytes (w, string, data_bytes);
1205 while (pad_bytes-- > 0)
1206 putc (' ', w->file);
1209 /* Writes SIZE bytes of DATA to W's output file. */
1211 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1213 fwrite (data, 1, size, w->file);
1216 /* Writes N zeros to W's output file. */
1218 write_zeros (struct sfm_writer *w, size_t n)
1224 /* Writes N spaces to W's output file. */
1226 write_spaces (struct sfm_writer *w, size_t n)
1229 putc (' ', w->file);