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/i18n.h>
35 #include <libpspp/version.h>
37 #include <data/attributes.h>
38 #include <data/case.h>
39 #include <data/casewriter-provider.h>
40 #include <data/casewriter.h>
41 #include <data/dictionary.h>
42 #include <data/file-handle-def.h>
43 #include <data/file-name.h>
44 #include <data/format.h>
45 #include <data/make-file.h>
46 #include <data/missing-values.h>
47 #include <data/settings.h>
48 #include <data/short-names.h>
49 #include <data/value-labels.h>
50 #include <data/variable.h>
53 #include "unlocked-io.h"
57 #define _(msgid) gettext (msgid)
58 #define N_(msgid) (msgid)
60 /* Compression bias used by PSPP. Values between (1 -
61 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
63 #define COMPRESSION_BIAS 100
65 /* System file writer. */
68 struct file_handle *fh; /* File handle. */
69 struct fh_lock *lock; /* Mutual exclusion for file. */
70 FILE *file; /* File stream. */
71 struct replace_file *rf; /* Ticket for replacing output file. */
73 bool compress; /* 1=compressed, 0=not compressed. */
74 casenumber case_cnt; /* Number of cases written so far. */
76 /* Compression buffering.
78 Compressed data is output as groups of 8 1-byte opcodes
79 followed by up to 8 (depending on the opcodes) 8-byte data
80 items. Data items and opcodes arrive at the same time but
81 must be reordered for writing to disk, thus a small amount
83 uint8_t opcodes[8]; /* Buffered opcodes. */
84 int opcode_cnt; /* Number of buffered opcodes. */
85 uint8_t data[8][8]; /* Buffered data. */
86 int data_cnt; /* Number of buffered data items. */
89 struct sfm_var *sfm_vars; /* Variables. */
90 size_t sfm_var_cnt; /* Number of variables. */
91 size_t segment_cnt; /* Number of variables including extra segments
92 for long string variables. */
95 static const struct casewriter_class sys_file_casewriter_class;
97 static void write_header (struct sfm_writer *, const struct dictionary *);
98 static void write_variable (struct sfm_writer *, const struct variable *, const struct dictionary *);
99 static void write_value_labels (struct sfm_writer *,
100 struct variable *, int idx, const struct dictionary *);
101 static void write_integer_info_record (struct sfm_writer *);
102 static void write_float_info_record (struct sfm_writer *);
104 static void write_longvar_table (struct sfm_writer *w,
105 const struct dictionary *dict);
107 static void write_encoding_record (struct sfm_writer *w,
108 const struct dictionary *);
110 static void write_vls_length_table (struct sfm_writer *w,
111 const struct dictionary *dict);
113 static void write_long_string_value_labels (struct sfm_writer *,
114 const struct dictionary *);
116 static void write_variable_display_parameters (struct sfm_writer *w,
117 const struct dictionary *dict);
119 static void write_documents (struct sfm_writer *, const struct dictionary *);
121 static void write_data_file_attributes (struct sfm_writer *,
122 const struct dictionary *);
123 static void write_variable_attributes (struct sfm_writer *,
124 const struct dictionary *);
126 static void write_int (struct sfm_writer *, int32_t);
127 static inline void convert_double_to_output_format (double, uint8_t[8]);
128 static void write_float (struct sfm_writer *, double);
129 static void write_string (struct sfm_writer *, const char *, size_t);
130 static void write_bytes (struct sfm_writer *, const void *, size_t);
131 static void write_zeros (struct sfm_writer *, size_t);
132 static void write_spaces (struct sfm_writer *, size_t);
133 static void write_value (struct sfm_writer *, const union value *, int width);
135 static void write_case_uncompressed (struct sfm_writer *,
136 const struct ccase *);
137 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
138 static void flush_compressed (struct sfm_writer *);
139 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
140 static void put_cmp_number (struct sfm_writer *, double);
141 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
143 bool write_error (const struct sfm_writer *);
144 bool close_writer (struct sfm_writer *);
146 /* Returns default options for writing a system file. */
147 struct sfm_write_options
148 sfm_writer_default_options (void)
150 struct sfm_write_options opts;
151 opts.create_writeable = true;
152 opts.compress = settings_get_scompression ();
157 /* Opens the system file designated by file handle FH for writing
158 cases from dictionary D according to the given OPTS. If
159 COMPRESS is nonzero, the system file will be compressed.
161 No reference to D is retained, so it may be modified or
162 destroyed at will after this function returns. D is not
163 modified by this function, except to assign short names. */
165 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
166 struct sfm_write_options opts)
168 struct sfm_writer *w;
174 if (opts.version != 2 && opts.version != 3)
176 msg (ME, _("Unknown system file version %d. Treating as version %d."),
181 /* Create and initialize writer. */
182 w = xmalloc (sizeof *w);
188 w->compress = opts.compress;
191 w->opcode_cnt = w->data_cnt = 0;
193 /* Figure out how to map in-memory case data to on-disk case
194 data. Also count the number of segments. Very long strings
195 occupy multiple segments, otherwise each variable only takes
197 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
200 /* Open file handle as an exclusive writer. */
201 /* TRANSLATORS: this fragment will be interpolated into
202 messages in fh_lock() that identify types of files. */
203 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
207 /* Create the file on disk. */
208 mode = S_IRUSR | S_IRGRP | S_IROTH;
209 if (opts.create_writeable)
210 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
211 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
215 msg (ME, _("Error opening \"%s\" for writing as a system file: %s."),
216 fh_get_file_name (fh), strerror (errno));
220 /* Write the file header. */
223 /* Write basic variable info. */
224 short_names_assign (d);
225 for (i = 0; i < dict_get_var_cnt (d); i++)
226 write_variable (w, dict_get_var (d, i), d);
228 /* Write out value labels. */
230 for (i = 0; i < dict_get_var_cnt (d); i++)
232 struct variable *v = dict_get_var (d, i);
234 write_value_labels (w, v, idx, d);
235 idx += sfm_width_to_octs (var_get_width (v));
238 if (dict_get_documents (d) != NULL)
239 write_documents (w, d);
241 write_integer_info_record (w);
242 write_float_info_record (w);
244 write_variable_display_parameters (w, d);
246 if (opts.version >= 3)
247 write_longvar_table (w, d);
249 write_vls_length_table (w, d);
251 write_long_string_value_labels (w, d);
253 if (attrset_count (dict_get_attributes (d)))
254 write_data_file_attributes (w, d);
255 write_variable_attributes (w, d);
257 write_encoding_record (w, d);
259 /* Write end-of-headers record. */
269 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
276 /* Returns value of X truncated to two least-significant digits. */
287 /* Calculates the offset of data for TARGET_VAR from the
288 beginning of each case's data for dictionary D. The return
289 value is in "octs" (8-byte units). */
291 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
297 for (i = 0; i < dict_get_var_cnt (d); i++)
299 struct variable *var = dict_get_var (d, i);
300 if (var == target_var)
302 oct_idx += sfm_width_to_octs (var_get_width (var));
307 /* Write the sysfile_header header to system file W. */
309 write_header (struct sfm_writer *w, const struct dictionary *d)
312 char creation_date[10];
313 char creation_time[9];
314 const char *file_label;
315 struct variable *weight;
319 /* Record-type code. */
320 write_string (w, "$FL2", 4);
322 /* Product identification. */
323 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
324 version, host_system);
325 write_string (w, prod_name, 60);
330 /* Number of `union value's per case. */
331 write_int (w, calc_oct_idx (d, NULL));
334 write_int (w, w->compress);
336 /* Weight variable. */
337 weight = dict_get_weight (d);
338 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
340 /* Number of cases. We don't know this in advance, so we write
341 -1 to indicate an unknown number of cases. Later we can
342 come back and overwrite it with the true value. */
345 /* Compression bias. */
346 write_float (w, COMPRESSION_BIAS);
348 /* Creation date and time. */
349 if (time (&t) == (time_t) -1)
351 strcpy (creation_date, "01 Jan 70");
352 strcpy (creation_time, "00:00:00");
356 static const char *const month_name[12] =
358 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
359 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
361 struct tm *tmp = localtime (&t);
362 int day = rerange (tmp->tm_mday);
363 int mon = rerange (tmp->tm_mon + 1);
364 int year = rerange (tmp->tm_year);
365 int hour = rerange (tmp->tm_hour + 1);
366 int min = rerange (tmp->tm_min + 1);
367 int sec = rerange (tmp->tm_sec + 1);
369 snprintf (creation_date, sizeof creation_date,
370 "%02d %s %02d", day, month_name[mon - 1], year);
371 snprintf (creation_time, sizeof creation_time,
372 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
374 write_string (w, creation_date, 9);
375 write_string (w, creation_time, 8);
378 file_label = dict_get_label (d);
379 if (file_label == NULL)
381 write_string (w, file_label, 64);
387 /* Write format spec FMT to W, after adjusting it to be
388 compatible with the given WIDTH. */
390 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
392 assert (fmt_check_output (&fmt));
393 assert (sfm_width_to_segments (width) == 1);
396 fmt_resize (&fmt, width);
397 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
400 /* Write a string continuation variable record for each 8-byte
401 section beyond the initial 8 bytes, for a variable of the
404 write_variable_continuation_records (struct sfm_writer *w, int width)
408 assert (sfm_width_to_segments (width) == 1);
409 for (position = 8; position < width; position += 8)
411 write_int (w, 2); /* Record type. */
412 write_int (w, -1); /* Width. */
413 write_int (w, 0); /* No variable label. */
414 write_int (w, 0); /* No missing values. */
415 write_int (w, 0); /* Print format. */
416 write_int (w, 0); /* Write format. */
417 write_zeros (w, 8); /* Name. */
421 /* Write the variable record(s) for variable V to system file
424 write_variable (struct sfm_writer *w, const struct variable *v, const struct dictionary *dict)
426 int width = var_get_width (v);
427 int segment_cnt = sfm_width_to_segments (width);
428 int seg0_width = sfm_segment_alloc_width (width, 0);
429 struct missing_values mv;
436 write_int (w, seg0_width);
438 /* Variable has a variable label? */
439 write_int (w, var_has_label (v));
441 /* Number of missing values. If there is a range, then the
442 range counts as 2 missing values and causes the number to be
444 mv_copy (&mv, var_get_missing_values (v));
445 if (mv_get_width (&mv) > 8)
447 if (mv_has_range (&mv))
448 write_int (w, -2 - mv_n_values (&mv));
450 write_int (w, mv_n_values (&mv));
452 /* Print and write formats. */
453 write_format (w, *var_get_print_format (v), seg0_width);
454 write_format (w, *var_get_write_format (v), seg0_width);
457 The full name is in a translation table written
459 write_string (w, var_get_short_name (v, 0), 8);
462 if (var_has_label (v))
464 char *label = recode_string (dict_get_encoding (dict), UTF8, var_get_label (v), -1);
465 size_t label_len = MIN (strlen (label), 255);
466 size_t padded_len = ROUND_UP (label_len, 4);
467 write_int (w, label_len);
468 write_string (w, label, padded_len);
472 /* Write the missing values, if any, range first. */
473 if (mv_has_range (&mv))
476 mv_get_range (&mv, &x, &y);
480 for (i = 0; i < mv_n_values (&mv); i++)
481 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
483 write_variable_continuation_records (w, seg0_width);
485 /* Write additional segments for very long string variables. */
486 for (i = 1; i < segment_cnt; i++)
488 int seg_width = sfm_segment_alloc_width (width, i);
489 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
491 write_int (w, 2); /* Variable record. */
492 write_int (w, seg_width); /* Width. */
493 write_int (w, 0); /* No variable label. */
494 write_int (w, 0); /* No missing values. */
495 write_format (w, fmt, seg_width); /* Print format. */
496 write_format (w, fmt, seg_width); /* Write format. */
497 write_string (w, var_get_short_name (v, i), 8);
499 write_variable_continuation_records (w, seg_width);
505 /* Writes the value labels for variable V having system file
506 variable index IDX to system file W.
508 Value labels for long string variables are written separately,
509 by write_long_string_value_labels. */
511 write_value_labels (struct sfm_writer *w, struct variable *v, int idx, const struct dictionary *dict)
513 const struct val_labs *val_labs;
514 const struct val_lab **labels;
518 val_labs = var_get_value_labels (v);
519 n_labels = val_labs_count (val_labs);
520 if (n_labels == 0 || var_get_width (v) > 8)
523 /* Value label record. */
524 write_int (w, 3); /* Record type. */
525 write_int (w, val_labs_count (val_labs));
526 labels = val_labs_sorted (val_labs);
527 for (i = 0; i < n_labels; i++)
529 const struct val_lab *vl = labels[i];
530 char *label = recode_string (dict_get_encoding (dict), UTF8, val_lab_get_label (vl), -1);
531 uint8_t len = MIN (strlen (label), 255);
533 write_value (w, val_lab_get_value (vl), var_get_width (v));
534 write_bytes (w, &len, 1);
535 write_bytes (w, label, len);
536 write_zeros (w, REM_RND_UP (len + 1, 8));
541 /* Value label variable record. */
542 write_int (w, 4); /* Record type. */
543 write_int (w, 1); /* Number of variables. */
544 write_int (w, idx + 1); /* Variable's dictionary index. */
547 /* Writes record type 6, document record. */
549 write_documents (struct sfm_writer *w, const struct dictionary *d)
551 size_t line_cnt = dict_get_document_line_cnt (d);
553 write_int (w, 6); /* Record type. */
554 write_int (w, line_cnt);
555 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
559 put_attrset (struct string *string, const struct attrset *attrs)
561 const struct attribute *attr;
562 struct attrset_iterator i;
564 for (attr = attrset_first (attrs, &i); attr != NULL;
565 attr = attrset_next (attrs, &i))
567 size_t n_values = attribute_get_n_values (attr);
570 ds_put_cstr (string, attribute_get_name (attr));
571 ds_put_char (string, '(');
572 for (j = 0; j < n_values; j++)
573 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
574 ds_put_char (string, ')');
579 write_attribute_record (struct sfm_writer *w, const struct string *content,
583 write_int (w, subtype);
585 write_int (w, ds_length (content));
586 write_bytes (w, ds_data (content), ds_length (content));
590 write_data_file_attributes (struct sfm_writer *w,
591 const struct dictionary *d)
593 struct string s = DS_EMPTY_INITIALIZER;
594 put_attrset (&s, dict_get_attributes (d));
595 write_attribute_record (w, &s, 17);
600 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
602 struct string s = DS_EMPTY_INITIALIZER;
603 size_t n_vars = dict_get_var_cnt (d);
604 size_t n_attrsets = 0;
607 for (i = 0; i < n_vars; i++)
609 struct variable *v = dict_get_var (d, i);
610 struct attrset *attrs = var_get_attributes (v);
611 if (attrset_count (attrs))
614 ds_put_char (&s, '/');
615 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
616 put_attrset (&s, attrs);
620 write_attribute_record (w, &s, 18);
624 /* Write the alignment, width and scale values. */
626 write_variable_display_parameters (struct sfm_writer *w,
627 const struct dictionary *dict)
631 write_int (w, 7); /* Record type. */
632 write_int (w, 11); /* Record subtype. */
633 write_int (w, 4); /* Data item (int32) size. */
634 write_int (w, w->segment_cnt * 3); /* Number of data items. */
636 for (i = 0; i < dict_get_var_cnt (dict); ++i)
638 struct variable *v = dict_get_var (dict, i);
639 int width = var_get_width (v);
640 int segment_cnt = sfm_width_to_segments (width);
641 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
642 : var_get_measure (v) == MEASURE_ORDINAL ? 2
644 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
645 : var_get_alignment (v) == ALIGN_RIGHT ? 1
649 for (i = 0; i < segment_cnt; i++)
651 int width_left = width - sfm_segment_effective_offset (width, i);
652 write_int (w, measure);
653 write_int (w, (i == 0 ? var_get_display_width (v)
654 : var_default_display_width (width_left)));
655 write_int (w, alignment);
660 /* Writes the table of lengths for very long string variables. */
662 write_vls_length_table (struct sfm_writer *w,
663 const struct dictionary *dict)
668 ds_init_empty (&map);
669 for (i = 0; i < dict_get_var_cnt (dict); ++i)
671 const struct variable *v = dict_get_var (dict, i);
672 if (sfm_width_to_segments (var_get_width (v)) > 1)
673 ds_put_format (&map, "%s=%05d%c\t",
674 var_get_short_name (v, 0), var_get_width (v), 0);
676 if (!ds_is_empty (&map))
678 write_int (w, 7); /* Record type. */
679 write_int (w, 14); /* Record subtype. */
680 write_int (w, 1); /* Data item (char) size. */
681 write_int (w, ds_length (&map)); /* Number of data items. */
682 write_bytes (w, ds_data (&map), ds_length (&map));
689 write_long_string_value_labels (struct sfm_writer *w,
690 const struct dictionary *dict)
692 size_t n_vars = dict_get_var_cnt (dict);
696 /* Figure out the size in advance. */
698 for (i = 0; i < n_vars; i++)
700 struct variable *var = dict_get_var (dict, i);
701 const struct val_labs *val_labs = var_get_value_labels (var);
702 int width = var_get_width (var);
703 const struct val_lab *val_lab;
705 if (val_labs_count (val_labs) == 0 || width < 9)
708 size += 12 + strlen (var_get_name (var));
709 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
710 val_lab = val_labs_next (val_labs, val_lab))
711 size += 8 + width + strlen (val_lab_get_label (val_lab));
716 write_int (w, 7); /* Record type. */
717 write_int (w, 21); /* Record subtype */
718 write_int (w, 1); /* Data item (byte) size. */
719 write_int (w, size); /* Number of data items. */
721 start = ftello (w->file);
722 for (i = 0; i < n_vars; i++)
724 struct variable *var = dict_get_var (dict, i);
725 const struct val_labs *val_labs = var_get_value_labels (var);
726 const char *var_name = var_get_name (var);
727 int width = var_get_width (var);
728 const struct val_lab *val_lab;
730 if (val_labs_count (val_labs) == 0 || width < 9)
733 write_int (w, strlen (var_name));
734 write_bytes (w, var_name, strlen (var_name));
735 write_int (w, width);
736 write_int (w, val_labs_count (val_labs));
737 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
738 val_lab = val_labs_next (val_labs, val_lab))
740 const char *label = val_lab_get_label (val_lab);
741 size_t label_length = strlen (label);
743 write_int (w, width);
744 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
746 write_int (w, label_length);
747 write_bytes (w, label, label_length);
750 assert (ftello (w->file) == start + size);
754 write_encoding_record (struct sfm_writer *w,
755 const struct dictionary *d)
757 const char *enc = dict_get_encoding (d);
762 write_int (w, 7); /* Record type. */
763 write_int (w, 20); /* Record subtype. */
764 write_int (w, 1); /* Data item (char) size. */
765 write_int (w, strlen (enc)); /* Number of data items. */
766 write_string (w, enc, strlen (enc));
770 /* Writes the long variable name table. */
772 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
777 ds_init_empty (&map);
778 for (i = 0; i < dict_get_var_cnt (dict); i++)
780 struct variable *v = dict_get_var (dict, i);
781 char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1);
784 ds_put_char (&map, '\t');
785 ds_put_format (&map, "%s=%s",
786 var_get_short_name (v, 0), longname);
790 write_int (w, 7); /* Record type. */
791 write_int (w, 13); /* Record subtype. */
792 write_int (w, 1); /* Data item (char) size. */
793 write_int (w, ds_length (&map)); /* Number of data items. */
794 write_bytes (w, ds_data (&map), ds_length (&map));
799 /* Write integer information record. */
801 write_integer_info_record (struct sfm_writer *w)
803 int version_component[3];
806 /* Parse the version string. */
807 memset (version_component, 0, sizeof version_component);
808 sscanf (bare_version, "%d.%d.%d",
809 &version_component[0], &version_component[1], &version_component[2]);
811 /* Figure out the floating-point format. */
812 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
813 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
815 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
817 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
823 write_int (w, 7); /* Record type. */
824 write_int (w, 3); /* Record subtype. */
825 write_int (w, 4); /* Data item (int32) size. */
826 write_int (w, 8); /* Number of data items. */
827 write_int (w, version_component[0]);
828 write_int (w, version_component[1]);
829 write_int (w, version_component[2]);
830 write_int (w, -1); /* Machine code. */
831 write_int (w, float_format);
832 write_int (w, 1); /* Compression code. */
833 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
834 write_int (w, 2); /* 7-bit ASCII. */
837 /* Write floating-point information record. */
839 write_float_info_record (struct sfm_writer *w)
841 write_int (w, 7); /* Record type. */
842 write_int (w, 4); /* Record subtype. */
843 write_int (w, 8); /* Data item (flt64) size. */
844 write_int (w, 3); /* Number of data items. */
845 write_float (w, SYSMIS); /* System-missing value. */
846 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
847 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
850 /* Writes case C to system file W. */
852 sys_file_casewriter_write (struct casewriter *writer, void *w_,
855 struct sfm_writer *w = w_;
857 if (ferror (w->file))
859 casewriter_force_error (writer);
867 write_case_uncompressed (w, c);
869 write_case_compressed (w, c);
874 /* Destroys system file writer W. */
876 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
878 struct sfm_writer *w = w_;
879 if (!close_writer (w))
880 casewriter_force_error (writer);
883 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
885 write_error (const struct sfm_writer *writer)
887 return ferror (writer->file);
890 /* Closes a system file after we're done with it.
891 Returns true if successful, false if an I/O error occurred. */
893 close_writer (struct sfm_writer *w)
904 if (w->opcode_cnt > 0)
905 flush_compressed (w);
908 ok = !write_error (w);
910 /* Seek back to the beginning and update the number of cases.
911 This is just a courtesy to later readers, so there's no need
912 to check return values or report errors. */
913 if (ok && w->case_cnt <= INT32_MAX && !fseek (w->file, 80, SEEK_SET))
915 write_int (w, w->case_cnt);
919 if (fclose (w->file) == EOF)
923 msg (ME, _("An I/O error occurred writing system file \"%s\"."),
924 fh_get_file_name (w->fh));
926 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
939 /* System file writer casewriter class. */
940 static const struct casewriter_class sys_file_casewriter_class =
942 sys_file_casewriter_write,
943 sys_file_casewriter_destroy,
947 /* Writes case C to system file W, without compressing it. */
949 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
953 for (i = 0; i < w->sfm_var_cnt; i++)
955 struct sfm_var *v = &w->sfm_vars[i];
957 if (v->var_width == 0)
958 write_float (w, case_num_idx (c, v->case_index));
961 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
963 write_spaces (w, v->padding);
968 /* Writes case C to system file W, with compression. */
970 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
974 for (i = 0; i < w->sfm_var_cnt; i++)
976 struct sfm_var *v = &w->sfm_vars[i];
978 if (v->var_width == 0)
980 double d = case_num_idx (c, v->case_index);
982 put_cmp_opcode (w, 255);
983 else if (d >= 1 - COMPRESSION_BIAS
984 && d <= 251 - COMPRESSION_BIAS
986 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
989 put_cmp_opcode (w, 253);
990 put_cmp_number (w, d);
995 int offset = v->offset;
998 /* This code properly deals with a width that is not a
999 multiple of 8, by ensuring that the final partial
1000 oct (8 byte unit) is treated as padded with spaces
1002 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1004 const void *data = case_str_idx (c, v->case_index) + offset;
1005 int chunk_size = MIN (width, 8);
1006 if (!memcmp (data, " ", chunk_size))
1007 put_cmp_opcode (w, 254);
1010 put_cmp_opcode (w, 253);
1011 put_cmp_string (w, data, chunk_size);
1015 /* This code deals properly with padding that is not a
1016 multiple of 8 bytes, by discarding the remainder,
1017 which was already effectively padded with spaces in
1018 the previous loop. (Note that v->width + v->padding
1019 is always a multiple of 8.) */
1020 for (padding = v->padding / 8; padding > 0; padding--)
1021 put_cmp_opcode (w, 254);
1026 /* Flushes buffered compressed opcodes and data to W.
1027 The compression buffer must not be empty. */
1029 flush_compressed (struct sfm_writer *w)
1031 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1033 write_bytes (w, w->opcodes, w->opcode_cnt);
1034 write_zeros (w, 8 - w->opcode_cnt);
1036 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1038 w->opcode_cnt = w->data_cnt = 0;
1041 /* Appends OPCODE to the buffered set of compression opcodes in
1042 W. Flushes the compression buffer beforehand if necessary. */
1044 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1046 if (w->opcode_cnt >= 8)
1047 flush_compressed (w);
1049 w->opcodes[w->opcode_cnt++] = opcode;
1052 /* Appends NUMBER to the buffered compression data in W. The
1053 buffer must not be full; the way to assure that is to call
1054 this function only just after a call to put_cmp_opcode, which
1055 will flush the buffer as necessary. */
1057 put_cmp_number (struct sfm_writer *w, double number)
1059 assert (w->opcode_cnt > 0);
1060 assert (w->data_cnt < 8);
1062 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1065 /* Appends SIZE bytes of DATA to the buffered compression data in
1066 W, followed by enough spaces to pad the output data to exactly
1067 8 bytes (thus, SIZE must be no greater than 8). The buffer
1068 must not be full; the way to assure that is to call this
1069 function only just after a call to put_cmp_opcode, which will
1070 flush the buffer as necessary. */
1072 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1074 assert (w->opcode_cnt > 0);
1075 assert (w->data_cnt < 8);
1078 memset (w->data[w->data_cnt], ' ', 8);
1079 memcpy (w->data[w->data_cnt], data, size);
1083 /* Writes 32-bit integer X to the output file for writer W. */
1085 write_int (struct sfm_writer *w, int32_t x)
1087 write_bytes (w, &x, sizeof x);
1090 /* Converts NATIVE to the 64-bit format used in output files in
1093 convert_double_to_output_format (double native, uint8_t output[8])
1095 /* If "double" is not a 64-bit type, then convert it to a
1096 64-bit type. Otherwise just copy it. */
1097 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1098 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1100 memcpy (output, &native, sizeof native);
1103 /* Writes floating-point number X to the output file for writer
1106 write_float (struct sfm_writer *w, double x)
1109 convert_double_to_output_format (x, output);
1110 write_bytes (w, output, sizeof output);
1113 /* Writes contents of VALUE with the given WIDTH to W, padding
1114 with zeros to a multiple of 8 bytes.
1115 To avoid a branch, and because we don't actually need to
1116 support it, WIDTH must be no bigger than 8. */
1118 write_value (struct sfm_writer *w, const union value *value, int width)
1120 assert (width <= 8);
1122 write_float (w, value->f);
1125 write_bytes (w, value_str (value, width), width);
1126 write_zeros (w, 8 - width);
1130 /* Writes null-terminated STRING in a field of the given WIDTH to
1131 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1132 is narrowed, it is padded on the right with spaces. */
1134 write_string (struct sfm_writer *w, const char *string, size_t width)
1136 size_t data_bytes = MIN (strlen (string), width);
1137 size_t pad_bytes = width - data_bytes;
1138 write_bytes (w, string, data_bytes);
1139 while (pad_bytes-- > 0)
1140 putc (' ', w->file);
1143 /* Writes SIZE bytes of DATA to W's output file. */
1145 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1147 fwrite (data, 1, size, w->file);
1150 /* Writes N zeros to W's output file. */
1152 write_zeros (struct sfm_writer *w, size_t n)
1158 /* Writes N spaces to W's output file. */
1160 write_spaces (struct sfm_writer *w, size_t n)
1163 putc (' ', w->file);