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/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 static bool write_error (const struct sfm_writer *);
148 static 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. */
273 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
280 /* Returns value of X truncated to two least-significant digits. */
291 /* Calculates the offset of data for TARGET_VAR from the
292 beginning of each case's data for dictionary D. The return
293 value is in "octs" (8-byte units). */
295 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
301 for (i = 0; i < dict_get_var_cnt (d); i++)
303 struct variable *var = dict_get_var (d, i);
304 if (var == target_var)
306 oct_idx += sfm_width_to_octs (var_get_width (var));
311 /* Write the sysfile_header header to system file W. */
313 write_header (struct sfm_writer *w, const struct dictionary *d)
316 char creation_date[10];
317 char creation_time[9];
318 const char *file_label;
319 struct variable *weight;
323 /* Record-type code. */
324 write_string (w, "$FL2", 4);
326 /* Product identification. */
327 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
328 version, host_system);
329 write_string (w, prod_name, 60);
334 /* Number of `union value's per case. */
335 write_int (w, calc_oct_idx (d, NULL));
338 write_int (w, w->compress);
340 /* Weight variable. */
341 weight = dict_get_weight (d);
342 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
344 /* Number of cases. We don't know this in advance, so we write
345 -1 to indicate an unknown number of cases. Later we can
346 come back and overwrite it with the true value. */
349 /* Compression bias. */
350 write_float (w, COMPRESSION_BIAS);
352 /* Creation date and time. */
353 if (time (&t) == (time_t) -1)
355 strcpy (creation_date, "01 Jan 70");
356 strcpy (creation_time, "00:00:00");
360 static const char *const month_name[12] =
362 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
363 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
365 struct tm *tmp = localtime (&t);
366 int day = rerange (tmp->tm_mday);
367 int mon = rerange (tmp->tm_mon + 1);
368 int year = rerange (tmp->tm_year);
369 int hour = rerange (tmp->tm_hour + 1);
370 int min = rerange (tmp->tm_min + 1);
371 int sec = rerange (tmp->tm_sec + 1);
373 snprintf (creation_date, sizeof creation_date,
374 "%02d %s %02d", day, month_name[mon - 1], year);
375 snprintf (creation_time, sizeof creation_time,
376 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
378 write_string (w, creation_date, 9);
379 write_string (w, creation_time, 8);
382 file_label = dict_get_label (d);
383 if (file_label == NULL)
385 write_string (w, file_label, 64);
391 /* Write format spec FMT to W, after adjusting it to be
392 compatible with the given WIDTH. */
394 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
396 assert (fmt_check_output (&fmt));
397 assert (sfm_width_to_segments (width) == 1);
400 fmt_resize (&fmt, width);
401 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
404 /* Write a string continuation variable record for each 8-byte
405 section beyond the initial 8 bytes, for a variable of the
408 write_variable_continuation_records (struct sfm_writer *w, int width)
412 assert (sfm_width_to_segments (width) == 1);
413 for (position = 8; position < width; position += 8)
415 write_int (w, 2); /* Record type. */
416 write_int (w, -1); /* Width. */
417 write_int (w, 0); /* No variable label. */
418 write_int (w, 0); /* No missing values. */
419 write_int (w, 0); /* Print format. */
420 write_int (w, 0); /* Write format. */
421 write_zeros (w, 8); /* Name. */
425 /* Write the variable record(s) for variable V to system file
428 write_variable (struct sfm_writer *w, const struct variable *v)
430 int width = var_get_width (v);
431 int segment_cnt = sfm_width_to_segments (width);
432 int seg0_width = sfm_segment_alloc_width (width, 0);
433 struct missing_values mv;
440 write_int (w, seg0_width);
442 /* Variable has a variable label? */
443 write_int (w, var_has_label (v));
445 /* Number of missing values. If there is a range, then the
446 range counts as 2 missing values and causes the number to be
448 mv_copy (&mv, var_get_missing_values (v));
449 if (mv_get_width (&mv) > 8)
451 if (mv_has_range (&mv))
452 write_int (w, -2 - mv_n_values (&mv));
454 write_int (w, mv_n_values (&mv));
456 /* Print and write formats. */
457 write_format (w, *var_get_print_format (v), seg0_width);
458 write_format (w, *var_get_write_format (v), seg0_width);
461 The full name is in a translation table written
463 write_string (w, var_get_short_name (v, 0), 8);
466 if (var_has_label (v))
468 char *label = recode_string (var_get_encoding (v), UTF8, var_get_label (v), -1);
469 size_t label_len = MIN (strlen (label), 255);
470 size_t padded_len = ROUND_UP (label_len, 4);
471 write_int (w, label_len);
472 write_string (w, label, padded_len);
476 /* Write the missing values, if any, range first. */
477 if (mv_has_range (&mv))
480 mv_get_range (&mv, &x, &y);
484 for (i = 0; i < mv_n_values (&mv); i++)
485 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
487 write_variable_continuation_records (w, seg0_width);
489 /* Write additional segments for very long string variables. */
490 for (i = 1; i < segment_cnt; i++)
492 int seg_width = sfm_segment_alloc_width (width, i);
493 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
495 write_int (w, 2); /* Variable record. */
496 write_int (w, seg_width); /* Width. */
497 write_int (w, 0); /* No variable label. */
498 write_int (w, 0); /* No missing values. */
499 write_format (w, fmt, seg_width); /* Print format. */
500 write_format (w, fmt, seg_width); /* Write format. */
501 write_string (w, var_get_short_name (v, i), 8);
503 write_variable_continuation_records (w, seg_width);
509 /* Writes the value labels for variable V having system file
510 variable index IDX to system file W.
512 Value labels for long string variables are written separately,
513 by write_long_string_value_labels. */
515 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
517 const struct val_labs *val_labs;
518 const struct val_lab **labels;
522 val_labs = var_get_value_labels (v);
523 n_labels = val_labs_count (val_labs);
524 if (n_labels == 0 || var_get_width (v) > 8)
527 /* Value label record. */
528 write_int (w, 3); /* Record type. */
529 write_int (w, val_labs_count (val_labs));
530 labels = val_labs_sorted (val_labs);
531 for (i = 0; i < n_labels; i++)
533 const struct val_lab *vl = labels[i];
534 char *label = recode_string (var_get_encoding (v), UTF8, val_lab_get_label (vl), -1);
535 uint8_t len = MIN (strlen (label), 255);
537 write_value (w, val_lab_get_value (vl), var_get_width (v));
538 write_bytes (w, &len, 1);
539 write_bytes (w, label, len);
540 write_zeros (w, REM_RND_UP (len + 1, 8));
545 /* Value label variable record. */
546 write_int (w, 4); /* Record type. */
547 write_int (w, 1); /* Number of variables. */
548 write_int (w, idx + 1); /* Variable's dictionary index. */
551 /* Writes record type 6, document record. */
553 write_documents (struct sfm_writer *w, const struct dictionary *d)
555 size_t line_cnt = dict_get_document_line_cnt (d);
557 write_int (w, 6); /* Record type. */
558 write_int (w, line_cnt);
559 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
563 put_attrset (struct string *string, const struct attrset *attrs)
565 const struct attribute *attr;
566 struct attrset_iterator i;
568 for (attr = attrset_first (attrs, &i); attr != NULL;
569 attr = attrset_next (attrs, &i))
571 size_t n_values = attribute_get_n_values (attr);
574 ds_put_cstr (string, attribute_get_name (attr));
575 ds_put_byte (string, '(');
576 for (j = 0; j < n_values; j++)
577 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
578 ds_put_byte (string, ')');
583 write_attribute_record (struct sfm_writer *w, const struct string *content,
587 write_int (w, subtype);
589 write_int (w, ds_length (content));
590 write_bytes (w, ds_data (content), ds_length (content));
594 write_data_file_attributes (struct sfm_writer *w,
595 const struct dictionary *d)
597 struct string s = DS_EMPTY_INITIALIZER;
598 put_attrset (&s, dict_get_attributes (d));
599 write_attribute_record (w, &s, 17);
604 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
606 struct string s = DS_EMPTY_INITIALIZER;
607 size_t n_vars = dict_get_var_cnt (d);
608 size_t n_attrsets = 0;
611 for (i = 0; i < n_vars; i++)
613 struct variable *v = dict_get_var (d, i);
614 struct attrset *attrs = var_get_attributes (v);
615 if (attrset_count (attrs))
618 ds_put_byte (&s, '/');
619 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
620 put_attrset (&s, attrs);
624 write_attribute_record (w, &s, 18);
628 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
629 SPSS before release 14, otherwise writes sets supported only by later
632 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
635 struct string s = DS_EMPTY_INITIALIZER;
639 n_mrsets = dict_get_n_mrsets (dict);
643 for (i = 0; i < n_mrsets; i++)
645 const struct mrset *mrset = dict_get_mrset (dict, i);
649 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
653 ds_put_format (&s, "%s=", mrset->name);
654 if (mrset->type == MRSET_MD)
658 if (mrset->cat_source == MRSET_COUNTEDVALUES)
659 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
661 ds_put_byte (&s, 'D');
663 if (mrset->width == 0)
664 counted = xasprintf ("%.0f", mrset->counted.f);
666 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
668 ds_put_format (&s, "%zu %s", strlen (counted), counted);
672 ds_put_byte (&s, 'C');
673 ds_put_byte (&s, ' ');
675 label = mrset->label && !mrset->label_from_var_label ? mrset->label : "";
676 ds_put_format (&s, "%zu %s", strlen (label), label);
678 for (j = 0; j < mrset->n_vars; j++)
679 ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0));
680 ds_put_byte (&s, '\n');
682 write_attribute_record (w, &s, pre_v14 ? 7 : 19);
686 /* Write the alignment, width and scale values. */
688 write_variable_display_parameters (struct sfm_writer *w,
689 const struct dictionary *dict)
693 write_int (w, 7); /* Record type. */
694 write_int (w, 11); /* Record subtype. */
695 write_int (w, 4); /* Data item (int32) size. */
696 write_int (w, w->segment_cnt * 3); /* Number of data items. */
698 for (i = 0; i < dict_get_var_cnt (dict); ++i)
700 struct variable *v = dict_get_var (dict, i);
701 int width = var_get_width (v);
702 int segment_cnt = sfm_width_to_segments (width);
703 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
704 : var_get_measure (v) == MEASURE_ORDINAL ? 2
706 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
707 : var_get_alignment (v) == ALIGN_RIGHT ? 1
711 for (i = 0; i < segment_cnt; i++)
713 int width_left = width - sfm_segment_effective_offset (width, i);
714 write_int (w, measure);
715 write_int (w, (i == 0 ? var_get_display_width (v)
716 : var_default_display_width (width_left)));
717 write_int (w, alignment);
722 /* Writes the table of lengths for very long string variables. */
724 write_vls_length_table (struct sfm_writer *w,
725 const struct dictionary *dict)
730 ds_init_empty (&map);
731 for (i = 0; i < dict_get_var_cnt (dict); ++i)
733 const struct variable *v = dict_get_var (dict, i);
734 if (sfm_width_to_segments (var_get_width (v)) > 1)
735 ds_put_format (&map, "%s=%05d%c\t",
736 var_get_short_name (v, 0), var_get_width (v), 0);
738 if (!ds_is_empty (&map))
740 write_int (w, 7); /* Record type. */
741 write_int (w, 14); /* Record subtype. */
742 write_int (w, 1); /* Data item (char) size. */
743 write_int (w, ds_length (&map)); /* Number of data items. */
744 write_bytes (w, ds_data (&map), ds_length (&map));
751 write_long_string_value_labels (struct sfm_writer *w,
752 const struct dictionary *dict)
754 size_t n_vars = dict_get_var_cnt (dict);
758 /* Figure out the size in advance. */
760 for (i = 0; i < n_vars; i++)
762 struct variable *var = dict_get_var (dict, i);
763 const struct val_labs *val_labs = var_get_value_labels (var);
764 int width = var_get_width (var);
765 const struct val_lab *val_lab;
767 if (val_labs_count (val_labs) == 0 || width < 9)
770 size += 12 + strlen (var_get_name (var));
771 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
772 val_lab = val_labs_next (val_labs, val_lab))
773 size += 8 + width + strlen (val_lab_get_label (val_lab));
778 write_int (w, 7); /* Record type. */
779 write_int (w, 21); /* Record subtype */
780 write_int (w, 1); /* Data item (byte) size. */
781 write_int (w, size); /* Number of data items. */
783 start = ftello (w->file);
784 for (i = 0; i < n_vars; i++)
786 struct variable *var = dict_get_var (dict, i);
787 const struct val_labs *val_labs = var_get_value_labels (var);
788 const char *var_name = var_get_name (var);
789 int width = var_get_width (var);
790 const struct val_lab *val_lab;
792 if (val_labs_count (val_labs) == 0 || width < 9)
795 write_int (w, strlen (var_name));
796 write_bytes (w, var_name, strlen (var_name));
797 write_int (w, width);
798 write_int (w, val_labs_count (val_labs));
799 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
800 val_lab = val_labs_next (val_labs, val_lab))
802 const char *label = val_lab_get_label (val_lab);
803 size_t label_length = strlen (label);
805 write_int (w, width);
806 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
808 write_int (w, label_length);
809 write_bytes (w, label, label_length);
812 assert (ftello (w->file) == start + size);
816 write_encoding_record (struct sfm_writer *w,
817 const struct dictionary *d)
819 const char *enc = dict_get_encoding (d);
824 write_int (w, 7); /* Record type. */
825 write_int (w, 20); /* Record subtype. */
826 write_int (w, 1); /* Data item (char) size. */
827 write_int (w, strlen (enc)); /* Number of data items. */
828 write_string (w, enc, strlen (enc));
832 /* Writes the long variable name table. */
834 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
839 ds_init_empty (&map);
840 for (i = 0; i < dict_get_var_cnt (dict); i++)
842 struct variable *v = dict_get_var (dict, i);
843 char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1);
846 ds_put_byte (&map, '\t');
847 ds_put_format (&map, "%s=%s",
848 var_get_short_name (v, 0), longname);
852 write_int (w, 7); /* Record type. */
853 write_int (w, 13); /* Record subtype. */
854 write_int (w, 1); /* Data item (char) size. */
855 write_int (w, ds_length (&map)); /* Number of data items. */
856 write_bytes (w, ds_data (&map), ds_length (&map));
861 /* Write integer information record. */
863 write_integer_info_record (struct sfm_writer *w)
865 int version_component[3];
868 /* Parse the version string. */
869 memset (version_component, 0, sizeof version_component);
870 sscanf (bare_version, "%d.%d.%d",
871 &version_component[0], &version_component[1], &version_component[2]);
873 /* Figure out the floating-point format. */
874 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
875 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
877 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
879 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
885 write_int (w, 7); /* Record type. */
886 write_int (w, 3); /* Record subtype. */
887 write_int (w, 4); /* Data item (int32) size. */
888 write_int (w, 8); /* Number of data items. */
889 write_int (w, version_component[0]);
890 write_int (w, version_component[1]);
891 write_int (w, version_component[2]);
892 write_int (w, -1); /* Machine code. */
893 write_int (w, float_format);
894 write_int (w, 1); /* Compression code. */
895 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
896 write_int (w, 2); /* 7-bit ASCII. */
899 /* Write floating-point information record. */
901 write_float_info_record (struct sfm_writer *w)
903 write_int (w, 7); /* Record type. */
904 write_int (w, 4); /* Record subtype. */
905 write_int (w, 8); /* Data item (flt64) size. */
906 write_int (w, 3); /* Number of data items. */
907 write_float (w, SYSMIS); /* System-missing value. */
908 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
909 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
912 /* Writes case C to system file W. */
914 sys_file_casewriter_write (struct casewriter *writer, void *w_,
917 struct sfm_writer *w = w_;
919 if (ferror (w->file))
921 casewriter_force_error (writer);
929 write_case_uncompressed (w, c);
931 write_case_compressed (w, c);
936 /* Destroys system file writer W. */
938 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
940 struct sfm_writer *w = w_;
941 if (!close_writer (w))
942 casewriter_force_error (writer);
945 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
947 write_error (const struct sfm_writer *writer)
949 return ferror (writer->file);
952 /* Closes a system file after we're done with it.
953 Returns true if successful, false if an I/O error occurred. */
955 close_writer (struct sfm_writer *w)
966 if (w->opcode_cnt > 0)
967 flush_compressed (w);
970 ok = !write_error (w);
972 /* Seek back to the beginning and update the number of cases.
973 This is just a courtesy to later readers, so there's no need
974 to check return values or report errors. */
975 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
977 write_int (w, w->case_cnt);
981 if (fclose (w->file) == EOF)
985 msg (ME, _("An I/O error occurred writing system file `%s'."),
986 fh_get_file_name (w->fh));
988 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1001 /* System file writer casewriter class. */
1002 static const struct casewriter_class sys_file_casewriter_class =
1004 sys_file_casewriter_write,
1005 sys_file_casewriter_destroy,
1009 /* Writes case C to system file W, without compressing it. */
1011 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1015 for (i = 0; i < w->sfm_var_cnt; i++)
1017 struct sfm_var *v = &w->sfm_vars[i];
1019 if (v->var_width == 0)
1020 write_float (w, case_num_idx (c, v->case_index));
1023 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1025 write_spaces (w, v->padding);
1030 /* Writes case C to system file W, with compression. */
1032 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1036 for (i = 0; i < w->sfm_var_cnt; i++)
1038 struct sfm_var *v = &w->sfm_vars[i];
1040 if (v->var_width == 0)
1042 double d = case_num_idx (c, v->case_index);
1044 put_cmp_opcode (w, 255);
1045 else if (d >= 1 - COMPRESSION_BIAS
1046 && d <= 251 - COMPRESSION_BIAS
1048 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1051 put_cmp_opcode (w, 253);
1052 put_cmp_number (w, d);
1057 int offset = v->offset;
1060 /* This code properly deals with a width that is not a
1061 multiple of 8, by ensuring that the final partial
1062 oct (8 byte unit) is treated as padded with spaces
1064 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1066 const void *data = case_str_idx (c, v->case_index) + offset;
1067 int chunk_size = MIN (width, 8);
1068 if (!memcmp (data, " ", chunk_size))
1069 put_cmp_opcode (w, 254);
1072 put_cmp_opcode (w, 253);
1073 put_cmp_string (w, data, chunk_size);
1077 /* This code deals properly with padding that is not a
1078 multiple of 8 bytes, by discarding the remainder,
1079 which was already effectively padded with spaces in
1080 the previous loop. (Note that v->width + v->padding
1081 is always a multiple of 8.) */
1082 for (padding = v->padding / 8; padding > 0; padding--)
1083 put_cmp_opcode (w, 254);
1088 /* Flushes buffered compressed opcodes and data to W.
1089 The compression buffer must not be empty. */
1091 flush_compressed (struct sfm_writer *w)
1093 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1095 write_bytes (w, w->opcodes, w->opcode_cnt);
1096 write_zeros (w, 8 - w->opcode_cnt);
1098 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1100 w->opcode_cnt = w->data_cnt = 0;
1103 /* Appends OPCODE to the buffered set of compression opcodes in
1104 W. Flushes the compression buffer beforehand if necessary. */
1106 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1108 if (w->opcode_cnt >= 8)
1109 flush_compressed (w);
1111 w->opcodes[w->opcode_cnt++] = opcode;
1114 /* Appends NUMBER to the buffered compression data in W. The
1115 buffer must not be full; the way to assure that is to call
1116 this function only just after a call to put_cmp_opcode, which
1117 will flush the buffer as necessary. */
1119 put_cmp_number (struct sfm_writer *w, double number)
1121 assert (w->opcode_cnt > 0);
1122 assert (w->data_cnt < 8);
1124 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1127 /* Appends SIZE bytes of DATA to the buffered compression data in
1128 W, followed by enough spaces to pad the output data to exactly
1129 8 bytes (thus, SIZE must be no greater than 8). The buffer
1130 must not be full; the way to assure that is to call this
1131 function only just after a call to put_cmp_opcode, which will
1132 flush the buffer as necessary. */
1134 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1136 assert (w->opcode_cnt > 0);
1137 assert (w->data_cnt < 8);
1140 memset (w->data[w->data_cnt], ' ', 8);
1141 memcpy (w->data[w->data_cnt], data, size);
1145 /* Writes 32-bit integer X to the output file for writer W. */
1147 write_int (struct sfm_writer *w, int32_t x)
1149 write_bytes (w, &x, sizeof x);
1152 /* Converts NATIVE to the 64-bit format used in output files in
1155 convert_double_to_output_format (double native, uint8_t output[8])
1157 /* If "double" is not a 64-bit type, then convert it to a
1158 64-bit type. Otherwise just copy it. */
1159 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1160 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1162 memcpy (output, &native, sizeof native);
1165 /* Writes floating-point number X to the output file for writer
1168 write_float (struct sfm_writer *w, double x)
1171 convert_double_to_output_format (x, output);
1172 write_bytes (w, output, sizeof output);
1175 /* Writes contents of VALUE with the given WIDTH to W, padding
1176 with zeros to a multiple of 8 bytes.
1177 To avoid a branch, and because we don't actually need to
1178 support it, WIDTH must be no bigger than 8. */
1180 write_value (struct sfm_writer *w, const union value *value, int width)
1182 assert (width <= 8);
1184 write_float (w, value->f);
1187 write_bytes (w, value_str (value, width), width);
1188 write_zeros (w, 8 - width);
1192 /* Writes null-terminated STRING in a field of the given WIDTH to
1193 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1194 is narrowed, it is padded on the right with spaces. */
1196 write_string (struct sfm_writer *w, const char *string, size_t width)
1198 size_t data_bytes = MIN (strlen (string), width);
1199 size_t pad_bytes = width - data_bytes;
1200 write_bytes (w, string, data_bytes);
1201 while (pad_bytes-- > 0)
1202 putc (' ', w->file);
1205 /* Writes SIZE bytes of DATA to W's output file. */
1207 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1209 fwrite (data, 1, size, w->file);
1212 /* Writes N zeros to W's output file. */
1214 write_zeros (struct sfm_writer *w, size_t n)
1220 /* Writes N spaces to W's output file. */
1222 write_spaces (struct sfm_writer *w, size_t n)
1225 putc (' ', w->file);