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_bytes (struct sfm_writer *, const void *, size_t);
136 static void write_zeros (struct sfm_writer *, size_t);
137 static void write_spaces (struct sfm_writer *, size_t);
138 static void write_value (struct sfm_writer *, const union value *, int width);
140 static void write_case_uncompressed (struct sfm_writer *,
141 const struct ccase *);
142 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
143 static void flush_compressed (struct sfm_writer *);
144 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
145 static void put_cmp_number (struct sfm_writer *, double);
146 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
148 static bool write_error (const struct sfm_writer *);
149 static bool close_writer (struct sfm_writer *);
151 /* Returns default options for writing a system file. */
152 struct sfm_write_options
153 sfm_writer_default_options (void)
155 struct sfm_write_options opts;
156 opts.create_writeable = true;
157 opts.compress = settings_get_scompression ();
162 /* Opens the system file designated by file handle FH for writing
163 cases from dictionary D according to the given OPTS.
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. */
213 if (opts.create_writeable)
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_document_line_cnt (d) > 0)
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. */
274 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
281 /* Returns value of X truncated to two least-significant digits. */
292 /* Calculates the offset of data for TARGET_VAR from the
293 beginning of each case's data for dictionary D. The return
294 value is in "octs" (8-byte units). */
296 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
302 for (i = 0; i < dict_get_var_cnt (d); i++)
304 struct variable *var = dict_get_var (d, i);
305 if (var == target_var)
307 oct_idx += sfm_width_to_octs (var_get_width (var));
312 /* Write the sysfile_header header to system file W. */
314 write_header (struct sfm_writer *w, const struct dictionary *d)
317 char creation_date[10];
318 char creation_time[9];
319 const char *file_label;
320 struct variable *weight;
324 /* Record-type code. */
325 write_string (w, "$FL2", 4);
327 /* Product identification. */
328 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
329 version, host_system);
330 write_string (w, prod_name, 60);
335 /* Number of `union value's per case. */
336 write_int (w, calc_oct_idx (d, NULL));
339 write_int (w, w->compress);
341 /* Weight variable. */
342 weight = dict_get_weight (d);
343 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
345 /* Number of cases. We don't know this in advance, so we write
346 -1 to indicate an unknown number of cases. Later we can
347 come back and overwrite it with the true value. */
350 /* Compression bias. */
351 write_float (w, COMPRESSION_BIAS);
353 /* Creation date and time. */
354 if (time (&t) == (time_t) -1)
356 strcpy (creation_date, "01 Jan 70");
357 strcpy (creation_time, "00:00:00");
361 static const char *const month_name[12] =
363 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
364 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
366 struct tm *tmp = localtime (&t);
367 int day = rerange (tmp->tm_mday);
368 int mon = rerange (tmp->tm_mon + 1);
369 int year = rerange (tmp->tm_year);
370 int hour = rerange (tmp->tm_hour + 1);
371 int min = rerange (tmp->tm_min + 1);
372 int sec = rerange (tmp->tm_sec + 1);
374 snprintf (creation_date, sizeof creation_date,
375 "%02d %s %02d", day, month_name[mon - 1], year);
376 snprintf (creation_time, sizeof creation_time,
377 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
379 write_string (w, creation_date, 9);
380 write_string (w, creation_time, 8);
383 file_label = dict_get_label (d);
384 if (file_label == NULL)
386 write_string (w, file_label, 64);
392 /* Write format spec FMT to W, after adjusting it to be
393 compatible with the given WIDTH. */
395 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
397 assert (fmt_check_output (&fmt));
398 assert (sfm_width_to_segments (width) == 1);
401 fmt_resize (&fmt, width);
402 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
405 /* Write a string continuation variable record for each 8-byte
406 section beyond the initial 8 bytes, for a variable of the
409 write_variable_continuation_records (struct sfm_writer *w, int width)
413 assert (sfm_width_to_segments (width) == 1);
414 for (position = 8; position < width; position += 8)
416 write_int (w, 2); /* Record type. */
417 write_int (w, -1); /* Width. */
418 write_int (w, 0); /* No variable label. */
419 write_int (w, 0); /* No missing values. */
420 write_int (w, 0); /* Print format. */
421 write_int (w, 0); /* Write format. */
422 write_zeros (w, 8); /* Name. */
426 /* Write the variable record(s) for variable V to system file
429 write_variable (struct sfm_writer *w, const struct variable *v)
431 int width = var_get_width (v);
432 int segment_cnt = sfm_width_to_segments (width);
433 int seg0_width = sfm_segment_alloc_width (width, 0);
434 struct missing_values mv;
441 write_int (w, seg0_width);
443 /* Variable has a variable label? */
444 write_int (w, var_has_label (v));
446 /* Number of missing values. If there is a range, then the
447 range counts as 2 missing values and causes the number to be
449 mv_copy (&mv, var_get_missing_values (v));
450 if (mv_get_width (&mv) > 8)
452 if (mv_has_range (&mv))
453 write_int (w, -2 - mv_n_values (&mv));
455 write_int (w, mv_n_values (&mv));
457 /* Print and write formats. */
458 write_format (w, *var_get_print_format (v), seg0_width);
459 write_format (w, *var_get_write_format (v), seg0_width);
462 The full name is in a translation table written
464 write_string (w, var_get_short_name (v, 0), 8);
467 if (var_has_label (v))
469 char *label = recode_string (var_get_encoding (v), UTF8, var_get_label (v), -1);
470 size_t label_len = MIN (strlen (label), 255);
471 size_t padded_len = ROUND_UP (label_len, 4);
472 write_int (w, label_len);
473 write_string (w, label, padded_len);
477 /* Write the missing values, if any, range first. */
478 if (mv_has_range (&mv))
481 mv_get_range (&mv, &x, &y);
485 for (i = 0; i < mv_n_values (&mv); i++)
486 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
488 write_variable_continuation_records (w, seg0_width);
490 /* Write additional segments for very long string variables. */
491 for (i = 1; i < segment_cnt; i++)
493 int seg_width = sfm_segment_alloc_width (width, i);
494 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
496 write_int (w, 2); /* Variable record. */
497 write_int (w, seg_width); /* Width. */
498 write_int (w, 0); /* No variable label. */
499 write_int (w, 0); /* No missing values. */
500 write_format (w, fmt, seg_width); /* Print format. */
501 write_format (w, fmt, seg_width); /* Write format. */
502 write_string (w, var_get_short_name (v, i), 8);
504 write_variable_continuation_records (w, seg_width);
510 /* Writes the value labels for variable V having system file
511 variable index IDX to system file W.
513 Value labels for long string variables are written separately,
514 by write_long_string_value_labels. */
516 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
518 const struct val_labs *val_labs;
519 const struct val_lab **labels;
523 val_labs = var_get_value_labels (v);
524 n_labels = val_labs_count (val_labs);
525 if (n_labels == 0 || var_get_width (v) > 8)
528 /* Value label record. */
529 write_int (w, 3); /* Record type. */
530 write_int (w, val_labs_count (val_labs));
531 labels = val_labs_sorted (val_labs);
532 for (i = 0; i < n_labels; i++)
534 const struct val_lab *vl = labels[i];
535 char *label = recode_string (var_get_encoding (v), UTF8, val_lab_get_label (vl), -1);
536 uint8_t len = MIN (strlen (label), 255);
538 write_value (w, val_lab_get_value (vl), var_get_width (v));
539 write_bytes (w, &len, 1);
540 write_bytes (w, label, len);
541 write_zeros (w, REM_RND_UP (len + 1, 8));
546 /* Value label variable record. */
547 write_int (w, 4); /* Record type. */
548 write_int (w, 1); /* Number of variables. */
549 write_int (w, idx + 1); /* Variable's dictionary index. */
552 /* Writes record type 6, document record. */
554 write_documents (struct sfm_writer *w, const struct dictionary *d)
556 const struct string_array *docs = dict_get_documents (d);
557 const char *enc = dict_get_encoding (d);
560 write_int (w, 6); /* Record type. */
561 write_int (w, docs->n);
562 for (i = 0; i < docs->n; i++)
564 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
565 size_t s_len = strlen (s);
566 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
568 write_bytes (w, s, write_len);
569 write_spaces (w, DOC_LINE_LENGTH - write_len);
575 put_attrset (struct string *string, const struct attrset *attrs)
577 const struct attribute *attr;
578 struct attrset_iterator i;
580 for (attr = attrset_first (attrs, &i); attr != NULL;
581 attr = attrset_next (attrs, &i))
583 size_t n_values = attribute_get_n_values (attr);
586 ds_put_cstr (string, attribute_get_name (attr));
587 ds_put_byte (string, '(');
588 for (j = 0; j < n_values; j++)
589 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
590 ds_put_byte (string, ')');
595 write_attribute_record (struct sfm_writer *w, const struct string *content,
599 write_int (w, subtype);
601 write_int (w, ds_length (content));
602 write_bytes (w, ds_data (content), ds_length (content));
606 write_data_file_attributes (struct sfm_writer *w,
607 const struct dictionary *d)
609 struct string s = DS_EMPTY_INITIALIZER;
610 put_attrset (&s, dict_get_attributes (d));
611 write_attribute_record (w, &s, 17);
616 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
618 struct string s = DS_EMPTY_INITIALIZER;
619 size_t n_vars = dict_get_var_cnt (d);
620 size_t n_attrsets = 0;
623 for (i = 0; i < n_vars; i++)
625 struct variable *v = dict_get_var (d, i);
626 struct attrset *attrs = var_get_attributes (v);
627 if (attrset_count (attrs))
630 ds_put_byte (&s, '/');
631 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
632 put_attrset (&s, attrs);
636 write_attribute_record (w, &s, 18);
640 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
641 SPSS before release 14, otherwise writes sets supported only by later
644 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
647 struct string s = DS_EMPTY_INITIALIZER;
651 n_mrsets = dict_get_n_mrsets (dict);
655 for (i = 0; i < n_mrsets; i++)
657 const struct mrset *mrset = dict_get_mrset (dict, i);
661 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
665 ds_put_format (&s, "%s=", mrset->name);
666 if (mrset->type == MRSET_MD)
670 if (mrset->cat_source == MRSET_COUNTEDVALUES)
671 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
673 ds_put_byte (&s, 'D');
675 if (mrset->width == 0)
676 counted = xasprintf ("%.0f", mrset->counted.f);
678 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
680 ds_put_format (&s, "%zu %s", strlen (counted), counted);
684 ds_put_byte (&s, 'C');
685 ds_put_byte (&s, ' ');
687 label = mrset->label && !mrset->label_from_var_label ? mrset->label : "";
688 ds_put_format (&s, "%zu %s", strlen (label), label);
690 for (j = 0; j < mrset->n_vars; j++)
691 ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0));
692 ds_put_byte (&s, '\n');
694 write_attribute_record (w, &s, pre_v14 ? 7 : 19);
698 /* Write the alignment, width and scale values. */
700 write_variable_display_parameters (struct sfm_writer *w,
701 const struct dictionary *dict)
705 write_int (w, 7); /* Record type. */
706 write_int (w, 11); /* Record subtype. */
707 write_int (w, 4); /* Data item (int32) size. */
708 write_int (w, w->segment_cnt * 3); /* Number of data items. */
710 for (i = 0; i < dict_get_var_cnt (dict); ++i)
712 struct variable *v = dict_get_var (dict, i);
713 int width = var_get_width (v);
714 int segment_cnt = sfm_width_to_segments (width);
715 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
716 : var_get_measure (v) == MEASURE_ORDINAL ? 2
718 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
719 : var_get_alignment (v) == ALIGN_RIGHT ? 1
723 for (i = 0; i < segment_cnt; i++)
725 int width_left = width - sfm_segment_effective_offset (width, i);
726 write_int (w, measure);
727 write_int (w, (i == 0 ? var_get_display_width (v)
728 : var_default_display_width (width_left)));
729 write_int (w, alignment);
734 /* Writes the table of lengths for very long string variables. */
736 write_vls_length_table (struct sfm_writer *w,
737 const struct dictionary *dict)
742 ds_init_empty (&map);
743 for (i = 0; i < dict_get_var_cnt (dict); ++i)
745 const struct variable *v = dict_get_var (dict, i);
746 if (sfm_width_to_segments (var_get_width (v)) > 1)
747 ds_put_format (&map, "%s=%05d%c\t",
748 var_get_short_name (v, 0), var_get_width (v), 0);
750 if (!ds_is_empty (&map))
752 write_int (w, 7); /* Record type. */
753 write_int (w, 14); /* Record subtype. */
754 write_int (w, 1); /* Data item (char) size. */
755 write_int (w, ds_length (&map)); /* Number of data items. */
756 write_bytes (w, ds_data (&map), ds_length (&map));
763 write_long_string_value_labels (struct sfm_writer *w,
764 const struct dictionary *dict)
766 size_t n_vars = dict_get_var_cnt (dict);
770 /* Figure out the size in advance. */
772 for (i = 0; i < n_vars; i++)
774 struct variable *var = dict_get_var (dict, i);
775 const struct val_labs *val_labs = var_get_value_labels (var);
776 int width = var_get_width (var);
777 const struct val_lab *val_lab;
779 if (val_labs_count (val_labs) == 0 || width < 9)
782 size += 12 + strlen (var_get_name (var));
783 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
784 val_lab = val_labs_next (val_labs, val_lab))
785 size += 8 + width + strlen (val_lab_get_label (val_lab));
790 write_int (w, 7); /* Record type. */
791 write_int (w, 21); /* Record subtype */
792 write_int (w, 1); /* Data item (byte) size. */
793 write_int (w, size); /* Number of data items. */
795 start = ftello (w->file);
796 for (i = 0; i < n_vars; i++)
798 struct variable *var = dict_get_var (dict, i);
799 const struct val_labs *val_labs = var_get_value_labels (var);
800 const char *var_name = var_get_name (var);
801 int width = var_get_width (var);
802 const struct val_lab *val_lab;
804 if (val_labs_count (val_labs) == 0 || width < 9)
807 write_int (w, strlen (var_name));
808 write_bytes (w, var_name, strlen (var_name));
809 write_int (w, width);
810 write_int (w, val_labs_count (val_labs));
811 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
812 val_lab = val_labs_next (val_labs, val_lab))
814 const char *label = val_lab_get_label (val_lab);
815 size_t label_length = strlen (label);
817 write_int (w, width);
818 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
820 write_int (w, label_length);
821 write_bytes (w, label, label_length);
824 assert (ftello (w->file) == start + size);
828 write_encoding_record (struct sfm_writer *w,
829 const struct dictionary *d)
831 const char *enc = dict_get_encoding (d);
836 write_int (w, 7); /* Record type. */
837 write_int (w, 20); /* Record subtype. */
838 write_int (w, 1); /* Data item (char) size. */
839 write_int (w, strlen (enc)); /* Number of data items. */
840 write_string (w, enc, strlen (enc));
844 /* Writes the long variable name table. */
846 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
851 ds_init_empty (&map);
852 for (i = 0; i < dict_get_var_cnt (dict); i++)
854 struct variable *v = dict_get_var (dict, i);
855 char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1);
858 ds_put_byte (&map, '\t');
859 ds_put_format (&map, "%s=%s",
860 var_get_short_name (v, 0), longname);
864 write_int (w, 7); /* Record type. */
865 write_int (w, 13); /* Record subtype. */
866 write_int (w, 1); /* Data item (char) size. */
867 write_int (w, ds_length (&map)); /* Number of data items. */
868 write_bytes (w, ds_data (&map), ds_length (&map));
873 /* Write integer information record. */
875 write_integer_info_record (struct sfm_writer *w)
877 int version_component[3];
880 /* Parse the version string. */
881 memset (version_component, 0, sizeof version_component);
882 sscanf (bare_version, "%d.%d.%d",
883 &version_component[0], &version_component[1], &version_component[2]);
885 /* Figure out the floating-point format. */
886 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
887 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
889 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
891 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
897 write_int (w, 7); /* Record type. */
898 write_int (w, 3); /* Record subtype. */
899 write_int (w, 4); /* Data item (int32) size. */
900 write_int (w, 8); /* Number of data items. */
901 write_int (w, version_component[0]);
902 write_int (w, version_component[1]);
903 write_int (w, version_component[2]);
904 write_int (w, -1); /* Machine code. */
905 write_int (w, float_format);
906 write_int (w, 1); /* Compression code. */
907 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
908 write_int (w, 2); /* 7-bit ASCII. */
911 /* Write floating-point information record. */
913 write_float_info_record (struct sfm_writer *w)
915 write_int (w, 7); /* Record type. */
916 write_int (w, 4); /* Record subtype. */
917 write_int (w, 8); /* Data item (flt64) size. */
918 write_int (w, 3); /* Number of data items. */
919 write_float (w, SYSMIS); /* System-missing value. */
920 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
921 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
924 /* Writes case C to system file W. */
926 sys_file_casewriter_write (struct casewriter *writer, void *w_,
929 struct sfm_writer *w = w_;
931 if (ferror (w->file))
933 casewriter_force_error (writer);
941 write_case_uncompressed (w, c);
943 write_case_compressed (w, c);
948 /* Destroys system file writer W. */
950 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
952 struct sfm_writer *w = w_;
953 if (!close_writer (w))
954 casewriter_force_error (writer);
957 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
959 write_error (const struct sfm_writer *writer)
961 return ferror (writer->file);
964 /* Closes a system file after we're done with it.
965 Returns true if successful, false if an I/O error occurred. */
967 close_writer (struct sfm_writer *w)
978 if (w->opcode_cnt > 0)
979 flush_compressed (w);
982 ok = !write_error (w);
984 /* Seek back to the beginning and update the number of cases.
985 This is just a courtesy to later readers, so there's no need
986 to check return values or report errors. */
987 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
989 write_int (w, w->case_cnt);
993 if (fclose (w->file) == EOF)
997 msg (ME, _("An I/O error occurred writing system file `%s'."),
998 fh_get_file_name (w->fh));
1000 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1004 fh_unlock (w->lock);
1013 /* System file writer casewriter class. */
1014 static const struct casewriter_class sys_file_casewriter_class =
1016 sys_file_casewriter_write,
1017 sys_file_casewriter_destroy,
1021 /* Writes case C to system file W, without compressing it. */
1023 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1027 for (i = 0; i < w->sfm_var_cnt; i++)
1029 struct sfm_var *v = &w->sfm_vars[i];
1031 if (v->var_width == 0)
1032 write_float (w, case_num_idx (c, v->case_index));
1035 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1037 write_spaces (w, v->padding);
1042 /* Writes case C to system file W, with compression. */
1044 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1048 for (i = 0; i < w->sfm_var_cnt; i++)
1050 struct sfm_var *v = &w->sfm_vars[i];
1052 if (v->var_width == 0)
1054 double d = case_num_idx (c, v->case_index);
1056 put_cmp_opcode (w, 255);
1057 else if (d >= 1 - COMPRESSION_BIAS
1058 && d <= 251 - COMPRESSION_BIAS
1060 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1063 put_cmp_opcode (w, 253);
1064 put_cmp_number (w, d);
1069 int offset = v->offset;
1072 /* This code properly deals with a width that is not a
1073 multiple of 8, by ensuring that the final partial
1074 oct (8 byte unit) is treated as padded with spaces
1076 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1078 const void *data = case_str_idx (c, v->case_index) + offset;
1079 int chunk_size = MIN (width, 8);
1080 if (!memcmp (data, " ", chunk_size))
1081 put_cmp_opcode (w, 254);
1084 put_cmp_opcode (w, 253);
1085 put_cmp_string (w, data, chunk_size);
1089 /* This code deals properly with padding that is not a
1090 multiple of 8 bytes, by discarding the remainder,
1091 which was already effectively padded with spaces in
1092 the previous loop. (Note that v->width + v->padding
1093 is always a multiple of 8.) */
1094 for (padding = v->padding / 8; padding > 0; padding--)
1095 put_cmp_opcode (w, 254);
1100 /* Flushes buffered compressed opcodes and data to W.
1101 The compression buffer must not be empty. */
1103 flush_compressed (struct sfm_writer *w)
1105 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1107 write_bytes (w, w->opcodes, w->opcode_cnt);
1108 write_zeros (w, 8 - w->opcode_cnt);
1110 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1112 w->opcode_cnt = w->data_cnt = 0;
1115 /* Appends OPCODE to the buffered set of compression opcodes in
1116 W. Flushes the compression buffer beforehand if necessary. */
1118 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1120 if (w->opcode_cnt >= 8)
1121 flush_compressed (w);
1123 w->opcodes[w->opcode_cnt++] = opcode;
1126 /* Appends NUMBER to the buffered compression data in W. The
1127 buffer must not be full; the way to assure that is to call
1128 this function only just after a call to put_cmp_opcode, which
1129 will flush the buffer as necessary. */
1131 put_cmp_number (struct sfm_writer *w, double number)
1133 assert (w->opcode_cnt > 0);
1134 assert (w->data_cnt < 8);
1136 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1139 /* Appends SIZE bytes of DATA to the buffered compression data in
1140 W, followed by enough spaces to pad the output data to exactly
1141 8 bytes (thus, SIZE must be no greater than 8). The buffer
1142 must not be full; the way to assure that is to call this
1143 function only just after a call to put_cmp_opcode, which will
1144 flush the buffer as necessary. */
1146 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1148 assert (w->opcode_cnt > 0);
1149 assert (w->data_cnt < 8);
1152 memset (w->data[w->data_cnt], ' ', 8);
1153 memcpy (w->data[w->data_cnt], data, size);
1157 /* Writes 32-bit integer X to the output file for writer W. */
1159 write_int (struct sfm_writer *w, int32_t x)
1161 write_bytes (w, &x, sizeof x);
1164 /* Converts NATIVE to the 64-bit format used in output files in
1167 convert_double_to_output_format (double native, uint8_t output[8])
1169 /* If "double" is not a 64-bit type, then convert it to a
1170 64-bit type. Otherwise just copy it. */
1171 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1172 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1174 memcpy (output, &native, sizeof native);
1177 /* Writes floating-point number X to the output file for writer
1180 write_float (struct sfm_writer *w, double x)
1183 convert_double_to_output_format (x, output);
1184 write_bytes (w, output, sizeof output);
1187 /* Writes contents of VALUE with the given WIDTH to W, padding
1188 with zeros to a multiple of 8 bytes.
1189 To avoid a branch, and because we don't actually need to
1190 support it, WIDTH must be no bigger than 8. */
1192 write_value (struct sfm_writer *w, const union value *value, int width)
1194 assert (width <= 8);
1196 write_float (w, value->f);
1199 write_bytes (w, value_str (value, width), width);
1200 write_zeros (w, 8 - width);
1204 /* Writes null-terminated STRING in a field of the given WIDTH to
1205 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1206 is narrowed, it is padded on the right with spaces. */
1208 write_string (struct sfm_writer *w, const char *string, size_t width)
1210 size_t data_bytes = MIN (strlen (string), width);
1211 size_t pad_bytes = width - data_bytes;
1212 write_bytes (w, string, data_bytes);
1213 while (pad_bytes-- > 0)
1214 putc (' ', w->file);
1217 /* Writes SIZE bytes of DATA to W's output file. */
1219 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1221 fwrite (data, 1, size, w->file);
1224 /* Writes N zeros to W's output file. */
1226 write_zeros (struct sfm_writer *w, size_t n)
1232 /* Writes N spaces to W's output file. */
1234 write_spaces (struct sfm_writer *w, size_t n)
1237 putc (' ', w->file);