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.
164 No reference to D is retained, so it may be modified or
165 destroyed at will after this function returns. D is not
166 modified by this function, except to assign short names. */
168 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
169 struct sfm_write_options opts)
171 struct sfm_writer *w;
177 if (opts.version != 2 && opts.version != 3)
179 msg (ME, _("Unknown system file version %d. Treating as version %d."),
184 /* Create and initialize writer. */
185 w = xmalloc (sizeof *w);
191 w->compress = opts.compress;
194 w->opcode_cnt = w->data_cnt = 0;
196 /* Figure out how to map in-memory case data to on-disk case
197 data. Also count the number of segments. Very long strings
198 occupy multiple segments, otherwise each variable only takes
200 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
203 /* Open file handle as an exclusive writer. */
204 /* TRANSLATORS: this fragment will be interpolated into
205 messages in fh_lock() that identify types of files. */
206 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
210 /* Create the file on disk. */
212 if (opts.create_writeable)
214 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
218 msg (ME, _("Error opening \"%s\" for writing as a system file: %s."),
219 fh_get_file_name (fh), strerror (errno));
223 /* Write the file header. */
226 /* Write basic variable info. */
227 short_names_assign (d);
228 for (i = 0; i < dict_get_var_cnt (d); i++)
229 write_variable (w, dict_get_var (d, i));
231 /* Write out value labels. */
233 for (i = 0; i < dict_get_var_cnt (d); i++)
235 struct variable *v = dict_get_var (d, i);
237 write_value_labels (w, v, idx);
238 idx += sfm_width_to_octs (var_get_width (v));
241 if (dict_get_documents (d) != NULL)
242 write_documents (w, d);
244 write_integer_info_record (w);
245 write_float_info_record (w);
247 write_mrsets (w, d, true);
249 write_variable_display_parameters (w, d);
251 if (opts.version >= 3)
252 write_longvar_table (w, d);
254 write_vls_length_table (w, d);
256 write_long_string_value_labels (w, d);
258 if (attrset_count (dict_get_attributes (d)))
259 write_data_file_attributes (w, d);
260 write_variable_attributes (w, d);
262 write_mrsets (w, d, false);
264 write_encoding_record (w, d);
266 /* Write end-of-headers record. */
276 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
283 /* Returns value of X truncated to two least-significant digits. */
294 /* Calculates the offset of data for TARGET_VAR from the
295 beginning of each case's data for dictionary D. The return
296 value is in "octs" (8-byte units). */
298 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
304 for (i = 0; i < dict_get_var_cnt (d); i++)
306 struct variable *var = dict_get_var (d, i);
307 if (var == target_var)
309 oct_idx += sfm_width_to_octs (var_get_width (var));
314 /* Write the sysfile_header header to system file W. */
316 write_header (struct sfm_writer *w, const struct dictionary *d)
319 char creation_date[10];
320 char creation_time[9];
321 const char *file_label;
322 struct variable *weight;
326 /* Record-type code. */
327 write_string (w, "$FL2", 4);
329 /* Product identification. */
330 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
331 version, host_system);
332 write_string (w, prod_name, 60);
337 /* Number of `union value's per case. */
338 write_int (w, calc_oct_idx (d, NULL));
341 write_int (w, w->compress);
343 /* Weight variable. */
344 weight = dict_get_weight (d);
345 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
347 /* Number of cases. We don't know this in advance, so we write
348 -1 to indicate an unknown number of cases. Later we can
349 come back and overwrite it with the true value. */
352 /* Compression bias. */
353 write_float (w, COMPRESSION_BIAS);
355 /* Creation date and time. */
356 if (time (&t) == (time_t) -1)
358 strcpy (creation_date, "01 Jan 70");
359 strcpy (creation_time, "00:00:00");
363 static const char *const month_name[12] =
365 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
366 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
368 struct tm *tmp = localtime (&t);
369 int day = rerange (tmp->tm_mday);
370 int mon = rerange (tmp->tm_mon + 1);
371 int year = rerange (tmp->tm_year);
372 int hour = rerange (tmp->tm_hour + 1);
373 int min = rerange (tmp->tm_min + 1);
374 int sec = rerange (tmp->tm_sec + 1);
376 snprintf (creation_date, sizeof creation_date,
377 "%02d %s %02d", day, month_name[mon - 1], year);
378 snprintf (creation_time, sizeof creation_time,
379 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
381 write_string (w, creation_date, 9);
382 write_string (w, creation_time, 8);
385 file_label = dict_get_label (d);
386 if (file_label == NULL)
388 write_string (w, file_label, 64);
394 /* Write format spec FMT to W, after adjusting it to be
395 compatible with the given WIDTH. */
397 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
399 assert (fmt_check_output (&fmt));
400 assert (sfm_width_to_segments (width) == 1);
403 fmt_resize (&fmt, width);
404 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
407 /* Write a string continuation variable record for each 8-byte
408 section beyond the initial 8 bytes, for a variable of the
411 write_variable_continuation_records (struct sfm_writer *w, int width)
415 assert (sfm_width_to_segments (width) == 1);
416 for (position = 8; position < width; position += 8)
418 write_int (w, 2); /* Record type. */
419 write_int (w, -1); /* Width. */
420 write_int (w, 0); /* No variable label. */
421 write_int (w, 0); /* No missing values. */
422 write_int (w, 0); /* Print format. */
423 write_int (w, 0); /* Write format. */
424 write_zeros (w, 8); /* Name. */
428 /* Write the variable record(s) for variable V to system file
431 write_variable (struct sfm_writer *w, const struct variable *v)
433 int width = var_get_width (v);
434 int segment_cnt = sfm_width_to_segments (width);
435 int seg0_width = sfm_segment_alloc_width (width, 0);
436 struct missing_values mv;
443 write_int (w, seg0_width);
445 /* Variable has a variable label? */
446 write_int (w, var_has_label (v));
448 /* Number of missing values. If there is a range, then the
449 range counts as 2 missing values and causes the number to be
451 mv_copy (&mv, var_get_missing_values (v));
452 if (mv_get_width (&mv) > 8)
454 if (mv_has_range (&mv))
455 write_int (w, -2 - mv_n_values (&mv));
457 write_int (w, mv_n_values (&mv));
459 /* Print and write formats. */
460 write_format (w, *var_get_print_format (v), seg0_width);
461 write_format (w, *var_get_write_format (v), seg0_width);
464 The full name is in a translation table written
466 write_string (w, var_get_short_name (v, 0), 8);
469 if (var_has_label (v))
471 char *label = recode_string (var_get_encoding (v), UTF8, var_get_label (v), -1);
472 size_t label_len = MIN (strlen (label), 255);
473 size_t padded_len = ROUND_UP (label_len, 4);
474 write_int (w, label_len);
475 write_string (w, label, padded_len);
479 /* Write the missing values, if any, range first. */
480 if (mv_has_range (&mv))
483 mv_get_range (&mv, &x, &y);
487 for (i = 0; i < mv_n_values (&mv); i++)
488 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
490 write_variable_continuation_records (w, seg0_width);
492 /* Write additional segments for very long string variables. */
493 for (i = 1; i < segment_cnt; i++)
495 int seg_width = sfm_segment_alloc_width (width, i);
496 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
498 write_int (w, 2); /* Variable record. */
499 write_int (w, seg_width); /* Width. */
500 write_int (w, 0); /* No variable label. */
501 write_int (w, 0); /* No missing values. */
502 write_format (w, fmt, seg_width); /* Print format. */
503 write_format (w, fmt, seg_width); /* Write format. */
504 write_string (w, var_get_short_name (v, i), 8);
506 write_variable_continuation_records (w, seg_width);
512 /* Writes the value labels for variable V having system file
513 variable index IDX to system file W.
515 Value labels for long string variables are written separately,
516 by write_long_string_value_labels. */
518 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
520 const struct val_labs *val_labs;
521 const struct val_lab **labels;
525 val_labs = var_get_value_labels (v);
526 n_labels = val_labs_count (val_labs);
527 if (n_labels == 0 || var_get_width (v) > 8)
530 /* Value label record. */
531 write_int (w, 3); /* Record type. */
532 write_int (w, val_labs_count (val_labs));
533 labels = val_labs_sorted (val_labs);
534 for (i = 0; i < n_labels; i++)
536 const struct val_lab *vl = labels[i];
537 char *label = recode_string (var_get_encoding (v), UTF8, val_lab_get_label (vl), -1);
538 uint8_t len = MIN (strlen (label), 255);
540 write_value (w, val_lab_get_value (vl), var_get_width (v));
541 write_bytes (w, &len, 1);
542 write_bytes (w, label, len);
543 write_zeros (w, REM_RND_UP (len + 1, 8));
548 /* Value label variable record. */
549 write_int (w, 4); /* Record type. */
550 write_int (w, 1); /* Number of variables. */
551 write_int (w, idx + 1); /* Variable's dictionary index. */
554 /* Writes record type 6, document record. */
556 write_documents (struct sfm_writer *w, const struct dictionary *d)
558 size_t line_cnt = dict_get_document_line_cnt (d);
560 write_int (w, 6); /* Record type. */
561 write_int (w, line_cnt);
562 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
566 put_attrset (struct string *string, const struct attrset *attrs)
568 const struct attribute *attr;
569 struct attrset_iterator i;
571 for (attr = attrset_first (attrs, &i); attr != NULL;
572 attr = attrset_next (attrs, &i))
574 size_t n_values = attribute_get_n_values (attr);
577 ds_put_cstr (string, attribute_get_name (attr));
578 ds_put_char (string, '(');
579 for (j = 0; j < n_values; j++)
580 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
581 ds_put_char (string, ')');
586 write_attribute_record (struct sfm_writer *w, const struct string *content,
590 write_int (w, subtype);
592 write_int (w, ds_length (content));
593 write_bytes (w, ds_data (content), ds_length (content));
597 write_data_file_attributes (struct sfm_writer *w,
598 const struct dictionary *d)
600 struct string s = DS_EMPTY_INITIALIZER;
601 put_attrset (&s, dict_get_attributes (d));
602 write_attribute_record (w, &s, 17);
607 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
609 struct string s = DS_EMPTY_INITIALIZER;
610 size_t n_vars = dict_get_var_cnt (d);
611 size_t n_attrsets = 0;
614 for (i = 0; i < n_vars; i++)
616 struct variable *v = dict_get_var (d, i);
617 struct attrset *attrs = var_get_attributes (v);
618 if (attrset_count (attrs))
621 ds_put_char (&s, '/');
622 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
623 put_attrset (&s, attrs);
627 write_attribute_record (w, &s, 18);
631 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
632 SPSS before release 14, otherwise writes sets supported only by later
635 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
638 struct string s = DS_EMPTY_INITIALIZER;
642 n_mrsets = dict_get_n_mrsets (dict);
646 for (i = 0; i < n_mrsets; i++)
648 const struct mrset *mrset = dict_get_mrset (dict, i);
652 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
656 ds_put_format (&s, "%s=", mrset->name);
657 if (mrset->type == MRSET_MD)
661 if (mrset->cat_source == MRSET_COUNTEDVALUES)
662 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
664 ds_put_char (&s, 'D');
666 if (mrset->width == 0)
667 counted = xasprintf ("%.0f", mrset->counted.f);
669 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
671 ds_put_format (&s, "%zu %s", strlen (counted), counted);
675 ds_put_char (&s, 'C');
676 ds_put_char (&s, ' ');
678 label = mrset->label && !mrset->label_from_var_label ? mrset->label : "";
679 ds_put_format (&s, "%zu %s", strlen (label), label);
681 for (j = 0; j < mrset->n_vars; j++)
682 ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0));
683 ds_put_char (&s, '\n');
685 write_attribute_record (w, &s, 7);
689 /* Write the alignment, width and scale values. */
691 write_variable_display_parameters (struct sfm_writer *w,
692 const struct dictionary *dict)
696 write_int (w, 7); /* Record type. */
697 write_int (w, 11); /* Record subtype. */
698 write_int (w, 4); /* Data item (int32) size. */
699 write_int (w, w->segment_cnt * 3); /* Number of data items. */
701 for (i = 0; i < dict_get_var_cnt (dict); ++i)
703 struct variable *v = dict_get_var (dict, i);
704 int width = var_get_width (v);
705 int segment_cnt = sfm_width_to_segments (width);
706 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
707 : var_get_measure (v) == MEASURE_ORDINAL ? 2
709 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
710 : var_get_alignment (v) == ALIGN_RIGHT ? 1
714 for (i = 0; i < segment_cnt; i++)
716 int width_left = width - sfm_segment_effective_offset (width, i);
717 write_int (w, measure);
718 write_int (w, (i == 0 ? var_get_display_width (v)
719 : var_default_display_width (width_left)));
720 write_int (w, alignment);
725 /* Writes the table of lengths for very long string variables. */
727 write_vls_length_table (struct sfm_writer *w,
728 const struct dictionary *dict)
733 ds_init_empty (&map);
734 for (i = 0; i < dict_get_var_cnt (dict); ++i)
736 const struct variable *v = dict_get_var (dict, i);
737 if (sfm_width_to_segments (var_get_width (v)) > 1)
738 ds_put_format (&map, "%s=%05d%c\t",
739 var_get_short_name (v, 0), var_get_width (v), 0);
741 if (!ds_is_empty (&map))
743 write_int (w, 7); /* Record type. */
744 write_int (w, 14); /* Record subtype. */
745 write_int (w, 1); /* Data item (char) size. */
746 write_int (w, ds_length (&map)); /* Number of data items. */
747 write_bytes (w, ds_data (&map), ds_length (&map));
754 write_long_string_value_labels (struct sfm_writer *w,
755 const struct dictionary *dict)
757 size_t n_vars = dict_get_var_cnt (dict);
761 /* Figure out the size in advance. */
763 for (i = 0; i < n_vars; i++)
765 struct variable *var = dict_get_var (dict, i);
766 const struct val_labs *val_labs = var_get_value_labels (var);
767 int width = var_get_width (var);
768 const struct val_lab *val_lab;
770 if (val_labs_count (val_labs) == 0 || width < 9)
773 size += 12 + strlen (var_get_name (var));
774 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
775 val_lab = val_labs_next (val_labs, val_lab))
776 size += 8 + width + strlen (val_lab_get_label (val_lab));
781 write_int (w, 7); /* Record type. */
782 write_int (w, 21); /* Record subtype */
783 write_int (w, 1); /* Data item (byte) size. */
784 write_int (w, size); /* Number of data items. */
786 start = ftello (w->file);
787 for (i = 0; i < n_vars; i++)
789 struct variable *var = dict_get_var (dict, i);
790 const struct val_labs *val_labs = var_get_value_labels (var);
791 const char *var_name = var_get_name (var);
792 int width = var_get_width (var);
793 const struct val_lab *val_lab;
795 if (val_labs_count (val_labs) == 0 || width < 9)
798 write_int (w, strlen (var_name));
799 write_bytes (w, var_name, strlen (var_name));
800 write_int (w, width);
801 write_int (w, val_labs_count (val_labs));
802 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
803 val_lab = val_labs_next (val_labs, val_lab))
805 const char *label = val_lab_get_label (val_lab);
806 size_t label_length = strlen (label);
808 write_int (w, width);
809 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
811 write_int (w, label_length);
812 write_bytes (w, label, label_length);
815 assert (ftello (w->file) == start + size);
819 write_encoding_record (struct sfm_writer *w,
820 const struct dictionary *d)
822 const char *enc = dict_get_encoding (d);
827 write_int (w, 7); /* Record type. */
828 write_int (w, 20); /* Record subtype. */
829 write_int (w, 1); /* Data item (char) size. */
830 write_int (w, strlen (enc)); /* Number of data items. */
831 write_string (w, enc, strlen (enc));
835 /* Writes the long variable name table. */
837 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
842 ds_init_empty (&map);
843 for (i = 0; i < dict_get_var_cnt (dict); i++)
845 struct variable *v = dict_get_var (dict, i);
846 char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1);
849 ds_put_char (&map, '\t');
850 ds_put_format (&map, "%s=%s",
851 var_get_short_name (v, 0), longname);
855 write_int (w, 7); /* Record type. */
856 write_int (w, 13); /* Record subtype. */
857 write_int (w, 1); /* Data item (char) size. */
858 write_int (w, ds_length (&map)); /* Number of data items. */
859 write_bytes (w, ds_data (&map), ds_length (&map));
864 /* Write integer information record. */
866 write_integer_info_record (struct sfm_writer *w)
868 int version_component[3];
871 /* Parse the version string. */
872 memset (version_component, 0, sizeof version_component);
873 sscanf (bare_version, "%d.%d.%d",
874 &version_component[0], &version_component[1], &version_component[2]);
876 /* Figure out the floating-point format. */
877 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
878 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
880 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
882 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
888 write_int (w, 7); /* Record type. */
889 write_int (w, 3); /* Record subtype. */
890 write_int (w, 4); /* Data item (int32) size. */
891 write_int (w, 8); /* Number of data items. */
892 write_int (w, version_component[0]);
893 write_int (w, version_component[1]);
894 write_int (w, version_component[2]);
895 write_int (w, -1); /* Machine code. */
896 write_int (w, float_format);
897 write_int (w, 1); /* Compression code. */
898 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
899 write_int (w, 2); /* 7-bit ASCII. */
902 /* Write floating-point information record. */
904 write_float_info_record (struct sfm_writer *w)
906 write_int (w, 7); /* Record type. */
907 write_int (w, 4); /* Record subtype. */
908 write_int (w, 8); /* Data item (flt64) size. */
909 write_int (w, 3); /* Number of data items. */
910 write_float (w, SYSMIS); /* System-missing value. */
911 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
912 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
915 /* Writes case C to system file W. */
917 sys_file_casewriter_write (struct casewriter *writer, void *w_,
920 struct sfm_writer *w = w_;
922 if (ferror (w->file))
924 casewriter_force_error (writer);
932 write_case_uncompressed (w, c);
934 write_case_compressed (w, c);
939 /* Destroys system file writer W. */
941 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
943 struct sfm_writer *w = w_;
944 if (!close_writer (w))
945 casewriter_force_error (writer);
948 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
950 write_error (const struct sfm_writer *writer)
952 return ferror (writer->file);
955 /* Closes a system file after we're done with it.
956 Returns true if successful, false if an I/O error occurred. */
958 close_writer (struct sfm_writer *w)
969 if (w->opcode_cnt > 0)
970 flush_compressed (w);
973 ok = !write_error (w);
975 /* Seek back to the beginning and update the number of cases.
976 This is just a courtesy to later readers, so there's no need
977 to check return values or report errors. */
978 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
980 write_int (w, w->case_cnt);
984 if (fclose (w->file) == EOF)
988 msg (ME, _("An I/O error occurred writing system file \"%s\"."),
989 fh_get_file_name (w->fh));
991 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1004 /* System file writer casewriter class. */
1005 static const struct casewriter_class sys_file_casewriter_class =
1007 sys_file_casewriter_write,
1008 sys_file_casewriter_destroy,
1012 /* Writes case C to system file W, without compressing it. */
1014 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1018 for (i = 0; i < w->sfm_var_cnt; i++)
1020 struct sfm_var *v = &w->sfm_vars[i];
1022 if (v->var_width == 0)
1023 write_float (w, case_num_idx (c, v->case_index));
1026 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1028 write_spaces (w, v->padding);
1033 /* Writes case C to system file W, with compression. */
1035 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1039 for (i = 0; i < w->sfm_var_cnt; i++)
1041 struct sfm_var *v = &w->sfm_vars[i];
1043 if (v->var_width == 0)
1045 double d = case_num_idx (c, v->case_index);
1047 put_cmp_opcode (w, 255);
1048 else if (d >= 1 - COMPRESSION_BIAS
1049 && d <= 251 - COMPRESSION_BIAS
1051 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1054 put_cmp_opcode (w, 253);
1055 put_cmp_number (w, d);
1060 int offset = v->offset;
1063 /* This code properly deals with a width that is not a
1064 multiple of 8, by ensuring that the final partial
1065 oct (8 byte unit) is treated as padded with spaces
1067 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1069 const void *data = case_str_idx (c, v->case_index) + offset;
1070 int chunk_size = MIN (width, 8);
1071 if (!memcmp (data, " ", chunk_size))
1072 put_cmp_opcode (w, 254);
1075 put_cmp_opcode (w, 253);
1076 put_cmp_string (w, data, chunk_size);
1080 /* This code deals properly with padding that is not a
1081 multiple of 8 bytes, by discarding the remainder,
1082 which was already effectively padded with spaces in
1083 the previous loop. (Note that v->width + v->padding
1084 is always a multiple of 8.) */
1085 for (padding = v->padding / 8; padding > 0; padding--)
1086 put_cmp_opcode (w, 254);
1091 /* Flushes buffered compressed opcodes and data to W.
1092 The compression buffer must not be empty. */
1094 flush_compressed (struct sfm_writer *w)
1096 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1098 write_bytes (w, w->opcodes, w->opcode_cnt);
1099 write_zeros (w, 8 - w->opcode_cnt);
1101 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1103 w->opcode_cnt = w->data_cnt = 0;
1106 /* Appends OPCODE to the buffered set of compression opcodes in
1107 W. Flushes the compression buffer beforehand if necessary. */
1109 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1111 if (w->opcode_cnt >= 8)
1112 flush_compressed (w);
1114 w->opcodes[w->opcode_cnt++] = opcode;
1117 /* Appends NUMBER to the buffered compression data in W. The
1118 buffer must not be full; the way to assure that is to call
1119 this function only just after a call to put_cmp_opcode, which
1120 will flush the buffer as necessary. */
1122 put_cmp_number (struct sfm_writer *w, double number)
1124 assert (w->opcode_cnt > 0);
1125 assert (w->data_cnt < 8);
1127 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1130 /* Appends SIZE bytes of DATA to the buffered compression data in
1131 W, followed by enough spaces to pad the output data to exactly
1132 8 bytes (thus, SIZE must be no greater than 8). The buffer
1133 must not be full; the way to assure that is to call this
1134 function only just after a call to put_cmp_opcode, which will
1135 flush the buffer as necessary. */
1137 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1139 assert (w->opcode_cnt > 0);
1140 assert (w->data_cnt < 8);
1143 memset (w->data[w->data_cnt], ' ', 8);
1144 memcpy (w->data[w->data_cnt], data, size);
1148 /* Writes 32-bit integer X to the output file for writer W. */
1150 write_int (struct sfm_writer *w, int32_t x)
1152 write_bytes (w, &x, sizeof x);
1155 /* Converts NATIVE to the 64-bit format used in output files in
1158 convert_double_to_output_format (double native, uint8_t output[8])
1160 /* If "double" is not a 64-bit type, then convert it to a
1161 64-bit type. Otherwise just copy it. */
1162 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1163 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1165 memcpy (output, &native, sizeof native);
1168 /* Writes floating-point number X to the output file for writer
1171 write_float (struct sfm_writer *w, double x)
1174 convert_double_to_output_format (x, output);
1175 write_bytes (w, output, sizeof output);
1178 /* Writes contents of VALUE with the given WIDTH to W, padding
1179 with zeros to a multiple of 8 bytes.
1180 To avoid a branch, and because we don't actually need to
1181 support it, WIDTH must be no bigger than 8. */
1183 write_value (struct sfm_writer *w, const union value *value, int width)
1185 assert (width <= 8);
1187 write_float (w, value->f);
1190 write_bytes (w, value_str (value, width), width);
1191 write_zeros (w, 8 - width);
1195 /* Writes null-terminated STRING in a field of the given WIDTH to
1196 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1197 is narrowed, it is padded on the right with spaces. */
1199 write_string (struct sfm_writer *w, const char *string, size_t width)
1201 size_t data_bytes = MIN (strlen (string), width);
1202 size_t pad_bytes = width - data_bytes;
1203 write_bytes (w, string, data_bytes);
1204 while (pad_bytes-- > 0)
1205 putc (' ', w->file);
1208 /* Writes SIZE bytes of DATA to W's output file. */
1210 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1212 fwrite (data, 1, size, w->file);
1215 /* Writes N zeros to W's output file. */
1217 write_zeros (struct sfm_writer *w, size_t n)
1223 /* Writes N spaces to W's output file. */
1225 write_spaces (struct sfm_writer *w, size_t n)
1228 putc (' ', w->file);