1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2007, 2009 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 "sys-file-writer.h"
20 #include "sys-file-private.h"
29 #include <libpspp/float-format.h>
30 #include <libpspp/integer-format.h>
31 #include <libpspp/message.h>
32 #include <libpspp/misc.h>
33 #include <libpspp/str.h>
34 #include <libpspp/version.h>
36 #include <data/attributes.h>
37 #include <data/case.h>
38 #include <data/casewriter-provider.h>
39 #include <data/casewriter.h>
40 #include <data/dictionary.h>
41 #include <data/file-handle-def.h>
42 #include <data/file-name.h>
43 #include <data/format.h>
44 #include <data/make-file.h>
45 #include <data/missing-values.h>
46 #include <data/settings.h>
47 #include <data/short-names.h>
48 #include <data/value-labels.h>
49 #include <data/variable.h>
52 #include "unlocked-io.h"
56 #define _(msgid) gettext (msgid)
57 #define N_(msgid) (msgid)
59 /* Compression bias used by PSPP. Values between (1 -
60 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
62 #define COMPRESSION_BIAS 100
64 /* System file writer. */
67 struct file_handle *fh; /* File handle. */
68 struct fh_lock *lock; /* Mutual exclusion for file. */
69 FILE *file; /* File stream. */
70 struct replace_file *rf; /* Ticket for replacing output file. */
72 bool compress; /* 1=compressed, 0=not compressed. */
73 casenumber case_cnt; /* Number of cases written so far. */
75 /* Compression buffering.
77 Compressed data is output as groups of 8 1-byte opcodes
78 followed by up to 8 (depending on the opcodes) 8-byte data
79 items. Data items and opcodes arrive at the same time but
80 must be reordered for writing to disk, thus a small amount
82 uint8_t opcodes[8]; /* Buffered opcodes. */
83 int opcode_cnt; /* Number of buffered opcodes. */
84 uint8_t data[8][8]; /* Buffered data. */
85 int data_cnt; /* Number of buffered data items. */
88 struct sfm_var *sfm_vars; /* Variables. */
89 size_t sfm_var_cnt; /* Number of variables. */
90 size_t segment_cnt; /* Number of variables including extra segments
91 for long string variables. */
94 static const struct casewriter_class sys_file_casewriter_class;
96 static void write_header (struct sfm_writer *, const struct dictionary *);
97 static void write_variable (struct sfm_writer *, const struct variable *);
98 static void write_value_labels (struct sfm_writer *,
99 struct variable *, int idx);
100 static void write_integer_info_record (struct sfm_writer *);
101 static void write_float_info_record (struct sfm_writer *);
103 static void write_longvar_table (struct sfm_writer *w,
104 const struct dictionary *dict);
106 static void write_encoding_record (struct sfm_writer *w,
107 const struct dictionary *);
109 static void write_vls_length_table (struct sfm_writer *w,
110 const struct dictionary *dict);
112 static void write_long_string_value_labels (struct sfm_writer *,
113 const struct dictionary *);
115 static void write_variable_display_parameters (struct sfm_writer *w,
116 const struct dictionary *dict);
118 static void write_documents (struct sfm_writer *, const struct dictionary *);
120 static void write_data_file_attributes (struct sfm_writer *,
121 const struct dictionary *);
122 static void write_variable_attributes (struct sfm_writer *,
123 const struct dictionary *);
125 static void write_int (struct sfm_writer *, int32_t);
126 static inline void convert_double_to_output_format (double, uint8_t[8]);
127 static void write_float (struct sfm_writer *, double);
128 static void write_string (struct sfm_writer *, const char *, size_t);
129 static void write_bytes (struct sfm_writer *, const void *, size_t);
130 static void write_zeros (struct sfm_writer *, size_t);
131 static void write_spaces (struct sfm_writer *, size_t);
132 static void write_value (struct sfm_writer *, const union value *, int width);
134 static void write_case_uncompressed (struct sfm_writer *,
135 const struct ccase *);
136 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
137 static void flush_compressed (struct sfm_writer *);
138 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
139 static void put_cmp_number (struct sfm_writer *, double);
140 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
142 bool write_error (const struct sfm_writer *);
143 bool close_writer (struct sfm_writer *);
145 /* Returns default options for writing a system file. */
146 struct sfm_write_options
147 sfm_writer_default_options (void)
149 struct sfm_write_options opts;
150 opts.create_writeable = true;
151 opts.compress = settings_get_scompression ();
156 /* Opens the system file designated by file handle FH for writing
157 cases from dictionary D according to the given OPTS. If
158 COMPRESS is nonzero, the system file will be compressed.
160 No reference to D is retained, so it may be modified or
161 destroyed at will after this function returns. D is not
162 modified by this function, except to assign short names. */
164 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
165 struct sfm_write_options opts)
167 struct sfm_writer *w;
173 if (opts.version != 2 && opts.version != 3)
175 msg (ME, _("Unknown system file version %d. Treating as version %d."),
180 /* Create and initialize writer. */
181 w = xmalloc (sizeof *w);
187 w->compress = opts.compress;
190 w->opcode_cnt = w->data_cnt = 0;
192 /* Figure out how to map in-memory case data to on-disk case
193 data. Also count the number of segments. Very long strings
194 occupy multiple segments, otherwise each variable only takes
196 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
199 /* Open file handle as an exclusive writer. */
200 /* TRANSLATORS: this fragment will be interpolated into
201 messages in fh_lock() that identify types of files. */
202 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
206 /* Create the file on disk. */
207 mode = S_IRUSR | S_IRGRP | S_IROTH;
208 if (opts.create_writeable)
209 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
210 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
214 msg (ME, _("Error opening \"%s\" for writing as a system file: %s."),
215 fh_get_file_name (fh), strerror (errno));
219 /* Write the file header. */
222 /* Write basic variable info. */
223 short_names_assign (d);
224 for (i = 0; i < dict_get_var_cnt (d); i++)
225 write_variable (w, dict_get_var (d, i));
227 /* Write out value labels. */
229 for (i = 0; i < dict_get_var_cnt (d); i++)
231 struct variable *v = dict_get_var (d, i);
233 write_value_labels (w, v, idx);
234 idx += sfm_width_to_octs (var_get_width (v));
237 if (dict_get_documents (d) != NULL)
238 write_documents (w, d);
240 write_integer_info_record (w);
241 write_float_info_record (w);
243 write_variable_display_parameters (w, d);
245 if (opts.version >= 3)
246 write_longvar_table (w, d);
248 write_vls_length_table (w, d);
250 write_long_string_value_labels (w, d);
252 if (attrset_count (dict_get_attributes (d)))
253 write_data_file_attributes (w, d);
254 write_variable_attributes (w, d);
256 write_encoding_record (w, d);
258 /* Write end-of-headers record. */
268 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
275 /* Returns value of X truncated to two least-significant digits. */
286 /* Calculates the offset of data for TARGET_VAR from the
287 beginning of each case's data for dictionary D. The return
288 value is in "octs" (8-byte units). */
290 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
296 for (i = 0; i < dict_get_var_cnt (d); i++)
298 struct variable *var = dict_get_var (d, i);
299 if (var == target_var)
301 oct_idx += sfm_width_to_octs (var_get_width (var));
306 /* Write the sysfile_header header to system file W. */
308 write_header (struct sfm_writer *w, const struct dictionary *d)
311 char creation_date[10];
312 char creation_time[9];
313 const char *file_label;
314 struct variable *weight;
318 /* Record-type code. */
319 write_string (w, "$FL2", 4);
321 /* Product identification. */
322 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
323 version, host_system);
324 write_string (w, prod_name, 60);
329 /* Number of `union value's per case. */
330 write_int (w, calc_oct_idx (d, NULL));
333 write_int (w, w->compress);
335 /* Weight variable. */
336 weight = dict_get_weight (d);
337 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
339 /* Number of cases. We don't know this in advance, so we write
340 -1 to indicate an unknown number of cases. Later we can
341 come back and overwrite it with the true value. */
344 /* Compression bias. */
345 write_float (w, COMPRESSION_BIAS);
347 /* Creation date and time. */
348 if (time (&t) == (time_t) -1)
350 strcpy (creation_date, "01 Jan 70");
351 strcpy (creation_time, "00:00:00");
355 static const char *const month_name[12] =
357 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
358 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
360 struct tm *tmp = localtime (&t);
361 int day = rerange (tmp->tm_mday);
362 int mon = rerange (tmp->tm_mon + 1);
363 int year = rerange (tmp->tm_year);
364 int hour = rerange (tmp->tm_hour + 1);
365 int min = rerange (tmp->tm_min + 1);
366 int sec = rerange (tmp->tm_sec + 1);
368 snprintf (creation_date, sizeof creation_date,
369 "%02d %s %02d", day, month_name[mon - 1], year);
370 snprintf (creation_time, sizeof creation_time,
371 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
373 write_string (w, creation_date, 9);
374 write_string (w, creation_time, 8);
377 file_label = dict_get_label (d);
378 if (file_label == NULL)
380 write_string (w, file_label, 64);
386 /* Write format spec FMT to W, after adjusting it to be
387 compatible with the given WIDTH. */
389 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
391 assert (fmt_check_output (&fmt));
392 assert (sfm_width_to_segments (width) == 1);
395 fmt_resize (&fmt, width);
396 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
399 /* Write a string continuation variable record for each 8-byte
400 section beyond the initial 8 bytes, for a variable of the
403 write_variable_continuation_records (struct sfm_writer *w, int width)
407 assert (sfm_width_to_segments (width) == 1);
408 for (position = 8; position < width; position += 8)
410 write_int (w, 2); /* Record type. */
411 write_int (w, -1); /* Width. */
412 write_int (w, 0); /* No variable label. */
413 write_int (w, 0); /* No missing values. */
414 write_int (w, 0); /* Print format. */
415 write_int (w, 0); /* Write format. */
416 write_zeros (w, 8); /* Name. */
420 /* Write the variable record(s) for variable V to system file
423 write_variable (struct sfm_writer *w, const struct variable *v)
425 int width = var_get_width (v);
426 int segment_cnt = sfm_width_to_segments (width);
427 int seg0_width = sfm_segment_alloc_width (width, 0);
428 struct missing_values mv;
435 write_int (w, seg0_width);
437 /* Variable has a variable label? */
438 write_int (w, var_has_label (v));
440 /* Number of missing values. If there is a range, then the
441 range counts as 2 missing values and causes the number to be
443 mv_copy (&mv, var_get_missing_values (v));
444 if (mv_get_width (&mv) > 8)
446 if (mv_has_range (&mv))
447 write_int (w, -2 - mv_n_values (&mv));
449 write_int (w, mv_n_values (&mv));
451 /* Print and write formats. */
452 write_format (w, *var_get_print_format (v), seg0_width);
453 write_format (w, *var_get_write_format (v), seg0_width);
456 The full name is in a translation table written
458 write_string (w, var_get_short_name (v, 0), 8);
461 if (var_has_label (v))
463 const char *label = var_get_label (v);
464 size_t padded_len = ROUND_UP (MIN (strlen (label), 255), 4);
465 write_int (w, padded_len);
466 write_string (w, label, padded_len);
469 /* Write the missing values, if any, range first. */
470 if (mv_has_range (&mv))
473 mv_get_range (&mv, &x, &y);
477 for (i = 0; i < mv_n_values (&mv); i++)
478 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
480 write_variable_continuation_records (w, seg0_width);
482 /* Write additional segments for very long string variables. */
483 for (i = 1; i < segment_cnt; i++)
485 int seg_width = sfm_segment_alloc_width (width, i);
486 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
488 write_int (w, 2); /* Variable record. */
489 write_int (w, seg_width); /* Width. */
490 write_int (w, 0); /* No variable label. */
491 write_int (w, 0); /* No missing values. */
492 write_format (w, fmt, seg_width); /* Print format. */
493 write_format (w, fmt, seg_width); /* Write format. */
494 write_string (w, var_get_short_name (v, i), 8);
496 write_variable_continuation_records (w, seg_width);
502 /* Writes the value labels for variable V having system file
503 variable index IDX to system file W.
505 Value labels for long string variables are written separately,
506 by write_long_string_value_labels. */
508 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
510 const struct val_labs *val_labs;
511 const struct val_lab **labels;
515 val_labs = var_get_value_labels (v);
516 n_labels = val_labs_count (val_labs);
517 if (n_labels == 0 || var_get_width (v) > 8)
520 /* Value label record. */
521 write_int (w, 3); /* Record type. */
522 write_int (w, val_labs_count (val_labs));
523 labels = val_labs_sorted (val_labs);
524 for (i = 0; i < n_labels; i++)
526 const struct val_lab *vl = labels[i];
527 const char *label = val_lab_get_label (vl);
528 uint8_t len = MIN (strlen (label), 255);
530 write_value (w, val_lab_get_value (vl), var_get_width (v));
531 write_bytes (w, &len, 1);
532 write_bytes (w, label, len);
533 write_zeros (w, REM_RND_UP (len + 1, 8));
537 /* Value label variable record. */
538 write_int (w, 4); /* Record type. */
539 write_int (w, 1); /* Number of variables. */
540 write_int (w, idx + 1); /* Variable's dictionary index. */
543 /* Writes record type 6, document record. */
545 write_documents (struct sfm_writer *w, const struct dictionary *d)
547 size_t line_cnt = dict_get_document_line_cnt (d);
549 write_int (w, 6); /* Record type. */
550 write_int (w, line_cnt);
551 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
555 put_attrset (struct string *string, const struct attrset *attrs)
557 const struct attribute *attr;
558 struct attrset_iterator i;
560 for (attr = attrset_first (attrs, &i); attr != NULL;
561 attr = attrset_next (attrs, &i))
563 size_t n_values = attribute_get_n_values (attr);
566 ds_put_cstr (string, attribute_get_name (attr));
567 ds_put_char (string, '(');
568 for (j = 0; j < n_values; j++)
569 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
570 ds_put_char (string, ')');
575 write_attribute_record (struct sfm_writer *w, const struct string *content,
579 write_int (w, subtype);
581 write_int (w, ds_length (content));
582 write_bytes (w, ds_data (content), ds_length (content));
586 write_data_file_attributes (struct sfm_writer *w,
587 const struct dictionary *d)
589 struct string s = DS_EMPTY_INITIALIZER;
590 put_attrset (&s, dict_get_attributes (d));
591 write_attribute_record (w, &s, 17);
596 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
598 struct string s = DS_EMPTY_INITIALIZER;
599 size_t n_vars = dict_get_var_cnt (d);
600 size_t n_attrsets = 0;
603 for (i = 0; i < n_vars; i++)
605 struct variable *v = dict_get_var (d, i);
606 struct attrset *attrs = var_get_attributes (v);
607 if (attrset_count (attrs))
610 ds_put_char (&s, '/');
611 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
612 put_attrset (&s, attrs);
616 write_attribute_record (w, &s, 18);
620 /* Write the alignment, width and scale values. */
622 write_variable_display_parameters (struct sfm_writer *w,
623 const struct dictionary *dict)
627 write_int (w, 7); /* Record type. */
628 write_int (w, 11); /* Record subtype. */
629 write_int (w, 4); /* Data item (int32) size. */
630 write_int (w, w->segment_cnt * 3); /* Number of data items. */
632 for (i = 0; i < dict_get_var_cnt (dict); ++i)
634 struct variable *v = dict_get_var (dict, i);
635 int width = var_get_width (v);
636 int segment_cnt = sfm_width_to_segments (width);
637 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
638 : var_get_measure (v) == MEASURE_ORDINAL ? 2
640 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
641 : var_get_alignment (v) == ALIGN_RIGHT ? 1
645 for (i = 0; i < segment_cnt; i++)
647 int width_left = width - sfm_segment_effective_offset (width, i);
648 write_int (w, measure);
649 write_int (w, (i == 0 ? var_get_display_width (v)
650 : var_default_display_width (width_left)));
651 write_int (w, alignment);
656 /* Writes the table of lengths for very long string variables. */
658 write_vls_length_table (struct sfm_writer *w,
659 const struct dictionary *dict)
664 ds_init_empty (&map);
665 for (i = 0; i < dict_get_var_cnt (dict); ++i)
667 const struct variable *v = dict_get_var (dict, i);
668 if (sfm_width_to_segments (var_get_width (v)) > 1)
669 ds_put_format (&map, "%s=%05d%c\t",
670 var_get_short_name (v, 0), var_get_width (v), 0);
672 if (!ds_is_empty (&map))
674 write_int (w, 7); /* Record type. */
675 write_int (w, 14); /* Record subtype. */
676 write_int (w, 1); /* Data item (char) size. */
677 write_int (w, ds_length (&map)); /* Number of data items. */
678 write_bytes (w, ds_data (&map), ds_length (&map));
685 write_long_string_value_labels (struct sfm_writer *w,
686 const struct dictionary *dict)
688 size_t n_vars = dict_get_var_cnt (dict);
692 /* Figure out the size in advance. */
694 for (i = 0; i < n_vars; i++)
696 struct variable *var = dict_get_var (dict, i);
697 const struct val_labs *val_labs = var_get_value_labels (var);
698 int width = var_get_width (var);
699 const struct val_lab *val_lab;
701 if (val_labs_count (val_labs) == 0 || width < 9)
704 size += 12 + strlen (var_get_name (var));
705 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
706 val_lab = val_labs_next (val_labs, val_lab))
707 size += 8 + width + strlen (val_lab_get_label (val_lab));
712 write_int (w, 7); /* Record type. */
713 write_int (w, 21); /* Record subtype */
714 write_int (w, 1); /* Data item (byte) size. */
715 write_int (w, size); /* Number of data items. */
717 start = ftello (w->file);
718 for (i = 0; i < n_vars; i++)
720 struct variable *var = dict_get_var (dict, i);
721 const struct val_labs *val_labs = var_get_value_labels (var);
722 const char *var_name = var_get_name (var);
723 int width = var_get_width (var);
724 const struct val_lab *val_lab;
726 if (val_labs_count (val_labs) == 0 || width < 9)
729 write_int (w, strlen (var_name));
730 write_bytes (w, var_name, strlen (var_name));
731 write_int (w, width);
732 write_int (w, val_labs_count (val_labs));
733 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
734 val_lab = val_labs_next (val_labs, val_lab))
736 const char *label = val_lab_get_label (val_lab);
737 size_t label_length = strlen (label);
739 write_int (w, width);
740 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
742 write_int (w, label_length);
743 write_bytes (w, label, label_length);
746 assert (ftello (w->file) == start + size);
750 write_encoding_record (struct sfm_writer *w,
751 const struct dictionary *d)
753 const char *enc = dict_get_encoding (d);
758 write_int (w, 7); /* Record type. */
759 write_int (w, 20); /* Record subtype. */
760 write_int (w, 1); /* Data item (char) size. */
761 write_int (w, strlen (enc)); /* Number of data items. */
762 write_string (w, enc, strlen (enc));
766 /* Writes the long variable name table. */
768 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
773 ds_init_empty (&map);
774 for (i = 0; i < dict_get_var_cnt (dict); i++)
776 struct variable *v = dict_get_var (dict, i);
779 ds_put_char (&map, '\t');
780 ds_put_format (&map, "%s=%s",
781 var_get_short_name (v, 0), var_get_name (v));
784 write_int (w, 7); /* Record type. */
785 write_int (w, 13); /* Record subtype. */
786 write_int (w, 1); /* Data item (char) size. */
787 write_int (w, ds_length (&map)); /* Number of data items. */
788 write_bytes (w, ds_data (&map), ds_length (&map));
793 /* Write integer information record. */
795 write_integer_info_record (struct sfm_writer *w)
797 int version_component[3];
800 /* Parse the version string. */
801 memset (version_component, 0, sizeof version_component);
802 sscanf (bare_version, "%d.%d.%d",
803 &version_component[0], &version_component[1], &version_component[2]);
805 /* Figure out the floating-point format. */
806 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
807 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
809 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
811 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
817 write_int (w, 7); /* Record type. */
818 write_int (w, 3); /* Record subtype. */
819 write_int (w, 4); /* Data item (int32) size. */
820 write_int (w, 8); /* Number of data items. */
821 write_int (w, version_component[0]);
822 write_int (w, version_component[1]);
823 write_int (w, version_component[2]);
824 write_int (w, -1); /* Machine code. */
825 write_int (w, float_format);
826 write_int (w, 1); /* Compression code. */
827 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
828 write_int (w, 2); /* 7-bit ASCII. */
831 /* Write floating-point information record. */
833 write_float_info_record (struct sfm_writer *w)
835 write_int (w, 7); /* Record type. */
836 write_int (w, 4); /* Record subtype. */
837 write_int (w, 8); /* Data item (flt64) size. */
838 write_int (w, 3); /* Number of data items. */
839 write_float (w, SYSMIS); /* System-missing value. */
840 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
841 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
844 /* Writes case C to system file W. */
846 sys_file_casewriter_write (struct casewriter *writer, void *w_,
849 struct sfm_writer *w = w_;
851 if (ferror (w->file))
853 casewriter_force_error (writer);
861 write_case_uncompressed (w, c);
863 write_case_compressed (w, c);
868 /* Destroys system file writer W. */
870 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
872 struct sfm_writer *w = w_;
873 if (!close_writer (w))
874 casewriter_force_error (writer);
877 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
879 write_error (const struct sfm_writer *writer)
881 return ferror (writer->file);
884 /* Closes a system file after we're done with it.
885 Returns true if successful, false if an I/O error occurred. */
887 close_writer (struct sfm_writer *w)
898 if (w->opcode_cnt > 0)
899 flush_compressed (w);
902 ok = !write_error (w);
904 /* Seek back to the beginning and update the number of cases.
905 This is just a courtesy to later readers, so there's no need
906 to check return values or report errors. */
907 if (ok && w->case_cnt <= INT32_MAX && !fseek (w->file, 80, SEEK_SET))
909 write_int (w, w->case_cnt);
913 if (fclose (w->file) == EOF)
917 msg (ME, _("An I/O error occurred writing system file \"%s\"."),
918 fh_get_file_name (w->fh));
920 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
933 /* System file writer casewriter class. */
934 static const struct casewriter_class sys_file_casewriter_class =
936 sys_file_casewriter_write,
937 sys_file_casewriter_destroy,
941 /* Writes case C to system file W, without compressing it. */
943 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
947 for (i = 0; i < w->sfm_var_cnt; i++)
949 struct sfm_var *v = &w->sfm_vars[i];
951 if (v->var_width == 0)
952 write_float (w, case_num_idx (c, v->case_index));
955 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
957 write_spaces (w, v->padding);
962 /* Writes case C to system file W, with compression. */
964 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
968 for (i = 0; i < w->sfm_var_cnt; i++)
970 struct sfm_var *v = &w->sfm_vars[i];
972 if (v->var_width == 0)
974 double d = case_num_idx (c, v->case_index);
976 put_cmp_opcode (w, 255);
977 else if (d >= 1 - COMPRESSION_BIAS
978 && d <= 251 - COMPRESSION_BIAS
980 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
983 put_cmp_opcode (w, 253);
984 put_cmp_number (w, d);
989 int offset = v->offset;
992 /* This code properly deals with a width that is not a
993 multiple of 8, by ensuring that the final partial
994 oct (8 byte unit) is treated as padded with spaces
996 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
998 const void *data = case_str_idx (c, v->case_index) + offset;
999 int chunk_size = MIN (width, 8);
1000 if (!memcmp (data, " ", chunk_size))
1001 put_cmp_opcode (w, 254);
1004 put_cmp_opcode (w, 253);
1005 put_cmp_string (w, data, chunk_size);
1009 /* This code deals properly with padding that is not a
1010 multiple of 8 bytes, by discarding the remainder,
1011 which was already effectively padded with spaces in
1012 the previous loop. (Note that v->width + v->padding
1013 is always a multiple of 8.) */
1014 for (padding = v->padding / 8; padding > 0; padding--)
1015 put_cmp_opcode (w, 254);
1020 /* Flushes buffered compressed opcodes and data to W.
1021 The compression buffer must not be empty. */
1023 flush_compressed (struct sfm_writer *w)
1025 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1027 write_bytes (w, w->opcodes, w->opcode_cnt);
1028 write_zeros (w, 8 - w->opcode_cnt);
1030 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1032 w->opcode_cnt = w->data_cnt = 0;
1035 /* Appends OPCODE to the buffered set of compression opcodes in
1036 W. Flushes the compression buffer beforehand if necessary. */
1038 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1040 if (w->opcode_cnt >= 8)
1041 flush_compressed (w);
1043 w->opcodes[w->opcode_cnt++] = opcode;
1046 /* Appends NUMBER to the buffered compression data in W. The
1047 buffer must not be full; the way to assure that is to call
1048 this function only just after a call to put_cmp_opcode, which
1049 will flush the buffer as necessary. */
1051 put_cmp_number (struct sfm_writer *w, double number)
1053 assert (w->opcode_cnt > 0);
1054 assert (w->data_cnt < 8);
1056 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1059 /* Appends SIZE bytes of DATA to the buffered compression data in
1060 W, followed by enough spaces to pad the output data to exactly
1061 8 bytes (thus, SIZE must be no greater than 8). The buffer
1062 must not be full; the way to assure that is to call this
1063 function only just after a call to put_cmp_opcode, which will
1064 flush the buffer as necessary. */
1066 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1068 assert (w->opcode_cnt > 0);
1069 assert (w->data_cnt < 8);
1072 memset (w->data[w->data_cnt], ' ', 8);
1073 memcpy (w->data[w->data_cnt], data, size);
1077 /* Writes 32-bit integer X to the output file for writer W. */
1079 write_int (struct sfm_writer *w, int32_t x)
1081 write_bytes (w, &x, sizeof x);
1084 /* Converts NATIVE to the 64-bit format used in output files in
1087 convert_double_to_output_format (double native, uint8_t output[8])
1089 /* If "double" is not a 64-bit type, then convert it to a
1090 64-bit type. Otherwise just copy it. */
1091 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1092 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1094 memcpy (output, &native, sizeof native);
1097 /* Writes floating-point number X to the output file for writer
1100 write_float (struct sfm_writer *w, double x)
1103 convert_double_to_output_format (x, output);
1104 write_bytes (w, output, sizeof output);
1107 /* Writes contents of VALUE with the given WIDTH to W, padding
1108 with zeros to a multiple of 8 bytes.
1109 To avoid a branch, and because we don't actually need to
1110 support it, WIDTH must be no bigger than 8. */
1112 write_value (struct sfm_writer *w, const union value *value, int width)
1114 assert (width <= 8);
1116 write_float (w, value->f);
1119 write_bytes (w, value_str (value, width), width);
1120 write_zeros (w, 8 - width);
1124 /* Writes null-terminated STRING in a field of the given WIDTH to
1125 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1126 is narrowed, it is padded on the right with spaces. */
1128 write_string (struct sfm_writer *w, const char *string, size_t width)
1130 size_t data_bytes = MIN (strlen (string), width);
1131 size_t pad_bytes = width - data_bytes;
1132 write_bytes (w, string, data_bytes);
1133 while (pad_bytes-- > 0)
1134 putc (' ', w->file);
1137 /* Writes SIZE bytes of DATA to W's output file. */
1139 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1141 fwrite (data, 1, size, w->file);
1144 /* Writes N zeros to W's output file. */
1146 write_zeros (struct sfm_writer *w, size_t n)
1152 /* Writes N spaces to W's output file. */
1154 write_spaces (struct sfm_writer *w, size_t n)
1157 putc (' ', w->file);