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);
113 static void write_variable_display_parameters (struct sfm_writer *w,
114 const struct dictionary *dict);
116 static void write_documents (struct sfm_writer *, const struct dictionary *);
118 static void write_data_file_attributes (struct sfm_writer *,
119 const struct dictionary *);
120 static void write_variable_attributes (struct sfm_writer *,
121 const struct dictionary *);
123 static void write_int (struct sfm_writer *, int32_t);
124 static inline void convert_double_to_output_format (double, uint8_t[8]);
125 static void write_float (struct sfm_writer *, double);
126 static void write_string (struct sfm_writer *, const char *, size_t);
127 static void write_bytes (struct sfm_writer *, const void *, size_t);
128 static void write_zeros (struct sfm_writer *, size_t);
129 static void write_spaces (struct sfm_writer *, size_t);
130 static void write_value (struct sfm_writer *, const union value *, int width);
132 static void write_case_uncompressed (struct sfm_writer *,
133 const struct ccase *);
134 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
135 static void flush_compressed (struct sfm_writer *);
136 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
137 static void put_cmp_number (struct sfm_writer *, double);
138 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
140 bool write_error (const struct sfm_writer *);
141 bool close_writer (struct sfm_writer *);
143 /* Returns default options for writing a system file. */
144 struct sfm_write_options
145 sfm_writer_default_options (void)
147 struct sfm_write_options opts;
148 opts.create_writeable = true;
149 opts.compress = settings_get_scompression ();
154 /* Opens the system file designated by file handle FH for writing
155 cases from dictionary D according to the given OPTS. If
156 COMPRESS is nonzero, the system file will be compressed.
158 No reference to D is retained, so it may be modified or
159 destroyed at will after this function returns. D is not
160 modified by this function, except to assign short names. */
162 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
163 struct sfm_write_options opts)
165 struct sfm_writer *w;
171 if (opts.version != 2 && opts.version != 3)
173 msg (ME, _("Unknown system file version %d. Treating as version %d."),
178 /* Create and initialize writer. */
179 w = xmalloc (sizeof *w);
185 w->compress = opts.compress;
188 w->opcode_cnt = w->data_cnt = 0;
190 /* Figure out how to map in-memory case data to on-disk case
191 data. Also count the number of segments. Very long strings
192 occupy multiple segments, otherwise each variable only takes
194 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
197 /* Open file handle as an exclusive writer. */
198 /* TRANSLATORS: this fragment will be interpolated into
199 messages in fh_lock() that identify types of files. */
200 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
204 /* Create the file on disk. */
205 mode = S_IRUSR | S_IRGRP | S_IROTH;
206 if (opts.create_writeable)
207 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
208 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
212 msg (ME, _("Error opening \"%s\" for writing as a system file: %s."),
213 fh_get_file_name (fh), strerror (errno));
217 /* Write the file header. */
220 /* Write basic variable info. */
221 short_names_assign (d);
222 for (i = 0; i < dict_get_var_cnt (d); i++)
223 write_variable (w, dict_get_var (d, i));
225 /* Write out value labels. */
227 for (i = 0; i < dict_get_var_cnt (d); i++)
229 struct variable *v = dict_get_var (d, i);
231 write_value_labels (w, v, idx);
232 idx += sfm_width_to_octs (var_get_width (v));
235 if (dict_get_documents (d) != NULL)
236 write_documents (w, d);
238 write_integer_info_record (w);
239 write_float_info_record (w);
241 write_variable_display_parameters (w, d);
243 if (opts.version >= 3)
244 write_longvar_table (w, d);
246 write_vls_length_table (w, d);
248 if (attrset_count (dict_get_attributes (d)))
249 write_data_file_attributes (w, d);
250 write_variable_attributes (w, d);
252 write_encoding_record (w, d);
254 /* Write end-of-headers record. */
264 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
271 /* Returns value of X truncated to two least-significant digits. */
282 /* Calculates the offset of data for TARGET_VAR from the
283 beginning of each case's data for dictionary D. The return
284 value is in "octs" (8-byte units). */
286 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
292 for (i = 0; i < dict_get_var_cnt (d); i++)
294 struct variable *var = dict_get_var (d, i);
295 if (var == target_var)
297 oct_idx += sfm_width_to_octs (var_get_width (var));
302 /* Write the sysfile_header header to system file W. */
304 write_header (struct sfm_writer *w, const struct dictionary *d)
307 char creation_date[10];
308 char creation_time[9];
309 const char *file_label;
310 struct variable *weight;
314 /* Record-type code. */
315 write_string (w, "$FL2", 4);
317 /* Product identification. */
318 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
319 version, host_system);
320 write_string (w, prod_name, 60);
325 /* Number of `union value's per case. */
326 write_int (w, calc_oct_idx (d, NULL));
329 write_int (w, w->compress);
331 /* Weight variable. */
332 weight = dict_get_weight (d);
333 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
335 /* Number of cases. We don't know this in advance, so we write
336 -1 to indicate an unknown number of cases. Later we can
337 come back and overwrite it with the true value. */
340 /* Compression bias. */
341 write_float (w, COMPRESSION_BIAS);
343 /* Creation date and time. */
344 if (time (&t) == (time_t) -1)
346 strcpy (creation_date, "01 Jan 70");
347 strcpy (creation_time, "00:00:00");
351 static const char *const month_name[12] =
353 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
354 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
356 struct tm *tmp = localtime (&t);
357 int day = rerange (tmp->tm_mday);
358 int mon = rerange (tmp->tm_mon + 1);
359 int year = rerange (tmp->tm_year);
360 int hour = rerange (tmp->tm_hour + 1);
361 int min = rerange (tmp->tm_min + 1);
362 int sec = rerange (tmp->tm_sec + 1);
364 snprintf (creation_date, sizeof creation_date,
365 "%02d %s %02d", day, month_name[mon - 1], year);
366 snprintf (creation_time, sizeof creation_time,
367 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
369 write_string (w, creation_date, 9);
370 write_string (w, creation_time, 8);
373 file_label = dict_get_label (d);
374 if (file_label == NULL)
376 write_string (w, file_label, 64);
382 /* Write format spec FMT to W, after adjusting it to be
383 compatible with the given WIDTH. */
385 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
387 assert (fmt_check_output (&fmt));
388 assert (sfm_width_to_segments (width) == 1);
391 fmt_resize (&fmt, width);
392 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
395 /* Write a string continuation variable record for each 8-byte
396 section beyond the initial 8 bytes, for a variable of the
399 write_variable_continuation_records (struct sfm_writer *w, int width)
403 assert (sfm_width_to_segments (width) == 1);
404 for (position = 8; position < width; position += 8)
406 write_int (w, 2); /* Record type. */
407 write_int (w, -1); /* Width. */
408 write_int (w, 0); /* No variable label. */
409 write_int (w, 0); /* No missing values. */
410 write_int (w, 0); /* Print format. */
411 write_int (w, 0); /* Write format. */
412 write_zeros (w, 8); /* Name. */
416 /* Write the variable record(s) for variable V to system file
419 write_variable (struct sfm_writer *w, const struct variable *v)
421 int width = var_get_width (v);
422 int segment_cnt = sfm_width_to_segments (width);
423 int seg0_width = sfm_segment_alloc_width (width, 0);
424 const struct missing_values *mv = var_get_missing_values (v);
431 write_int (w, seg0_width);
433 /* Variable has a variable label? */
434 write_int (w, var_has_label (v));
436 /* Number of missing values. If there is a range, then the
437 range counts as 2 missing values and causes the number to be
439 write_int (w, mv_has_range (mv) ? -2 - mv_n_values (mv) : mv_n_values (mv));
441 /* Print and write formats. */
442 write_format (w, *var_get_print_format (v), seg0_width);
443 write_format (w, *var_get_write_format (v), seg0_width);
446 The full name is in a translation table written
448 write_string (w, var_get_short_name (v, 0), 8);
451 if (var_has_label (v))
453 const char *label = var_get_label (v);
454 size_t padded_len = ROUND_UP (MIN (strlen (label), 255), 4);
455 write_int (w, padded_len);
456 write_string (w, label, padded_len);
459 /* Write the missing values, if any, range first. */
460 if (mv_has_range (mv))
463 mv_get_range (mv, &x, &y);
467 for (i = 0; i < mv_n_values (mv); i++)
470 mv_get_value (mv, &value, i);
471 write_value (w, &value, seg0_width);
474 write_variable_continuation_records (w, seg0_width);
476 /* Write additional segments for very long string variables. */
477 for (i = 1; i < segment_cnt; i++)
479 int seg_width = sfm_segment_alloc_width (width, i);
480 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
482 write_int (w, 2); /* Variable record. */
483 write_int (w, seg_width); /* Width. */
484 write_int (w, 0); /* No variable label. */
485 write_int (w, 0); /* No missing values. */
486 write_format (w, fmt, seg_width); /* Print format. */
487 write_format (w, fmt, seg_width); /* Write format. */
488 write_string (w, var_get_short_name (v, i), 8);
490 write_variable_continuation_records (w, seg_width);
494 /* Writes the value labels for variable V having system file
495 variable index IDX to system file W. */
497 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
499 const struct val_labs *val_labs;
500 const struct val_lab **labels;
504 val_labs = var_get_value_labels (v);
505 n_labels = val_labs_count (val_labs);
509 /* Value label record. */
510 write_int (w, 3); /* Record type. */
511 write_int (w, val_labs_count (val_labs));
512 labels = val_labs_sorted (val_labs);
513 for (i = 0; i < n_labels; i++)
515 const struct val_lab *vl = labels[i];
516 const char *label = val_lab_get_label (vl);
517 uint8_t len = MIN (strlen (label), 255);
519 write_value (w, val_lab_get_value (vl), var_get_width (v));
520 write_bytes (w, &len, 1);
521 write_bytes (w, label, len);
522 write_zeros (w, REM_RND_UP (len + 1, 8));
526 /* Value label variable record. */
527 write_int (w, 4); /* Record type. */
528 write_int (w, 1); /* Number of variables. */
529 write_int (w, idx + 1); /* Variable's dictionary index. */
532 /* Writes record type 6, document record. */
534 write_documents (struct sfm_writer *w, const struct dictionary *d)
536 size_t line_cnt = dict_get_document_line_cnt (d);
538 write_int (w, 6); /* Record type. */
539 write_int (w, line_cnt);
540 write_bytes (w, dict_get_documents (d), line_cnt * DOC_LINE_LENGTH);
544 put_attrset (struct string *string, const struct attrset *attrs)
546 const struct attribute *attr;
547 struct attrset_iterator i;
549 for (attr = attrset_first (attrs, &i); attr != NULL;
550 attr = attrset_next (attrs, &i))
552 size_t n_values = attribute_get_n_values (attr);
555 ds_put_cstr (string, attribute_get_name (attr));
556 ds_put_char (string, '(');
557 for (j = 0; j < n_values; j++)
558 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
559 ds_put_char (string, ')');
564 write_attribute_record (struct sfm_writer *w, const struct string *content,
568 write_int (w, subtype);
570 write_int (w, ds_length (content));
571 write_bytes (w, ds_data (content), ds_length (content));
575 write_data_file_attributes (struct sfm_writer *w,
576 const struct dictionary *d)
578 struct string s = DS_EMPTY_INITIALIZER;
579 put_attrset (&s, dict_get_attributes (d));
580 write_attribute_record (w, &s, 17);
585 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
587 struct string s = DS_EMPTY_INITIALIZER;
588 size_t n_vars = dict_get_var_cnt (d);
589 size_t n_attrsets = 0;
592 for (i = 0; i < n_vars; i++)
594 struct variable *v = dict_get_var (d, i);
595 struct attrset *attrs = var_get_attributes (v);
596 if (attrset_count (attrs))
599 ds_put_char (&s, '/');
600 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
601 put_attrset (&s, attrs);
605 write_attribute_record (w, &s, 18);
609 /* Write the alignment, width and scale values. */
611 write_variable_display_parameters (struct sfm_writer *w,
612 const struct dictionary *dict)
616 write_int (w, 7); /* Record type. */
617 write_int (w, 11); /* Record subtype. */
618 write_int (w, 4); /* Data item (int32) size. */
619 write_int (w, w->segment_cnt * 3); /* Number of data items. */
621 for (i = 0; i < dict_get_var_cnt (dict); ++i)
623 struct variable *v = dict_get_var (dict, i);
624 int width = var_get_width (v);
625 int segment_cnt = sfm_width_to_segments (width);
626 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
627 : var_get_measure (v) == MEASURE_ORDINAL ? 2
629 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
630 : var_get_alignment (v) == ALIGN_RIGHT ? 1
634 for (i = 0; i < segment_cnt; i++)
636 int width_left = width - sfm_segment_effective_offset (width, i);
637 write_int (w, measure);
638 write_int (w, (i == 0 ? var_get_display_width (v)
639 : var_default_display_width (width_left)));
640 write_int (w, alignment);
645 /* Writes the table of lengths for very long string variables. */
647 write_vls_length_table (struct sfm_writer *w,
648 const struct dictionary *dict)
653 ds_init_empty (&map);
654 for (i = 0; i < dict_get_var_cnt (dict); ++i)
656 const struct variable *v = dict_get_var (dict, i);
657 if (sfm_width_to_segments (var_get_width (v)) > 1)
658 ds_put_format (&map, "%s=%05d%c\t",
659 var_get_short_name (v, 0), var_get_width (v), 0);
661 if (!ds_is_empty (&map))
663 write_int (w, 7); /* Record type. */
664 write_int (w, 14); /* Record subtype. */
665 write_int (w, 1); /* Data item (char) size. */
666 write_int (w, ds_length (&map)); /* Number of data items. */
667 write_bytes (w, ds_data (&map), ds_length (&map));
674 write_encoding_record (struct sfm_writer *w,
675 const struct dictionary *d)
677 const char *enc = dict_get_encoding (d);
682 write_int (w, 7); /* Record type. */
683 write_int (w, 20); /* Record subtype. */
684 write_int (w, 1); /* Data item (char) size. */
685 write_int (w, strlen (enc)); /* Number of data items. */
686 write_string (w, enc, strlen (enc));
690 /* Writes the long variable name table. */
692 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
697 ds_init_empty (&map);
698 for (i = 0; i < dict_get_var_cnt (dict); i++)
700 struct variable *v = dict_get_var (dict, i);
703 ds_put_char (&map, '\t');
704 ds_put_format (&map, "%s=%s",
705 var_get_short_name (v, 0), var_get_name (v));
708 write_int (w, 7); /* Record type. */
709 write_int (w, 13); /* Record subtype. */
710 write_int (w, 1); /* Data item (char) size. */
711 write_int (w, ds_length (&map)); /* Number of data items. */
712 write_bytes (w, ds_data (&map), ds_length (&map));
717 /* Write integer information record. */
719 write_integer_info_record (struct sfm_writer *w)
721 int version_component[3];
724 /* Parse the version string. */
725 memset (version_component, 0, sizeof version_component);
726 sscanf (bare_version, "%d.%d.%d",
727 &version_component[0], &version_component[1], &version_component[2]);
729 /* Figure out the floating-point format. */
730 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
731 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
733 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
735 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
741 write_int (w, 7); /* Record type. */
742 write_int (w, 3); /* Record subtype. */
743 write_int (w, 4); /* Data item (int32) size. */
744 write_int (w, 8); /* Number of data items. */
745 write_int (w, version_component[0]);
746 write_int (w, version_component[1]);
747 write_int (w, version_component[2]);
748 write_int (w, -1); /* Machine code. */
749 write_int (w, float_format);
750 write_int (w, 1); /* Compression code. */
751 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
752 write_int (w, 2); /* 7-bit ASCII. */
755 /* Write floating-point information record. */
757 write_float_info_record (struct sfm_writer *w)
759 write_int (w, 7); /* Record type. */
760 write_int (w, 4); /* Record subtype. */
761 write_int (w, 8); /* Data item (flt64) size. */
762 write_int (w, 3); /* Number of data items. */
763 write_float (w, SYSMIS); /* System-missing value. */
764 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
765 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
768 /* Writes case C to system file W. */
770 sys_file_casewriter_write (struct casewriter *writer, void *w_,
773 struct sfm_writer *w = w_;
775 if (ferror (w->file))
777 casewriter_force_error (writer);
785 write_case_uncompressed (w, c);
787 write_case_compressed (w, c);
792 /* Destroys system file writer W. */
794 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
796 struct sfm_writer *w = w_;
797 if (!close_writer (w))
798 casewriter_force_error (writer);
801 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
803 write_error (const struct sfm_writer *writer)
805 return ferror (writer->file);
808 /* Closes a system file after we're done with it.
809 Returns true if successful, false if an I/O error occurred. */
811 close_writer (struct sfm_writer *w)
822 if (w->opcode_cnt > 0)
823 flush_compressed (w);
826 ok = !write_error (w);
828 /* Seek back to the beginning and update the number of cases.
829 This is just a courtesy to later readers, so there's no need
830 to check return values or report errors. */
831 if (ok && w->case_cnt <= INT32_MAX && !fseek (w->file, 80, SEEK_SET))
833 write_int (w, w->case_cnt);
837 if (fclose (w->file) == EOF)
841 msg (ME, _("An I/O error occurred writing system file \"%s\"."),
842 fh_get_file_name (w->fh));
844 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
857 /* System file writer casewriter class. */
858 static const struct casewriter_class sys_file_casewriter_class =
860 sys_file_casewriter_write,
861 sys_file_casewriter_destroy,
865 /* Writes case C to system file W, without compressing it. */
867 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
871 for (i = 0; i < w->sfm_var_cnt; i++)
873 struct sfm_var *v = &w->sfm_vars[i];
875 if (v->var_width == 0)
876 write_float (w, case_num_idx (c, v->case_index));
879 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
881 write_spaces (w, v->padding);
886 /* Writes case C to system file W, with compression. */
888 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
892 for (i = 0; i < w->sfm_var_cnt; i++)
894 struct sfm_var *v = &w->sfm_vars[i];
896 if (v->var_width == 0)
898 double d = case_num_idx (c, v->case_index);
900 put_cmp_opcode (w, 255);
901 else if (d >= 1 - COMPRESSION_BIAS
902 && d <= 251 - COMPRESSION_BIAS
904 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
907 put_cmp_opcode (w, 253);
908 put_cmp_number (w, d);
913 int offset = v->offset;
916 /* This code properly deals with a width that is not a
917 multiple of 8, by ensuring that the final partial
918 oct (8 byte unit) is treated as padded with spaces
920 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
922 const void *data = case_str_idx (c, v->case_index) + offset;
923 int chunk_size = MIN (width, 8);
924 if (!memcmp (data, " ", chunk_size))
925 put_cmp_opcode (w, 254);
928 put_cmp_opcode (w, 253);
929 put_cmp_string (w, data, chunk_size);
933 /* This code deals properly with padding that is not a
934 multiple of 8 bytes, by discarding the remainder,
935 which was already effectively padded with spaces in
936 the previous loop. (Note that v->width + v->padding
937 is always a multiple of 8.) */
938 for (padding = v->padding / 8; padding > 0; padding--)
939 put_cmp_opcode (w, 254);
944 /* Flushes buffered compressed opcodes and data to W.
945 The compression buffer must not be empty. */
947 flush_compressed (struct sfm_writer *w)
949 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
951 write_bytes (w, w->opcodes, w->opcode_cnt);
952 write_zeros (w, 8 - w->opcode_cnt);
954 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
956 w->opcode_cnt = w->data_cnt = 0;
959 /* Appends OPCODE to the buffered set of compression opcodes in
960 W. Flushes the compression buffer beforehand if necessary. */
962 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
964 if (w->opcode_cnt >= 8)
965 flush_compressed (w);
967 w->opcodes[w->opcode_cnt++] = opcode;
970 /* Appends NUMBER to the buffered compression data in W. The
971 buffer must not be full; the way to assure that is to call
972 this function only just after a call to put_cmp_opcode, which
973 will flush the buffer as necessary. */
975 put_cmp_number (struct sfm_writer *w, double number)
977 assert (w->opcode_cnt > 0);
978 assert (w->data_cnt < 8);
980 convert_double_to_output_format (number, w->data[w->data_cnt++]);
983 /* Appends SIZE bytes of DATA to the buffered compression data in
984 W, followed by enough spaces to pad the output data to exactly
985 8 bytes (thus, SIZE must be no greater than 8). The buffer
986 must not be full; the way to assure that is to call this
987 function only just after a call to put_cmp_opcode, which will
988 flush the buffer as necessary. */
990 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
992 assert (w->opcode_cnt > 0);
993 assert (w->data_cnt < 8);
996 memset (w->data[w->data_cnt], ' ', 8);
997 memcpy (w->data[w->data_cnt], data, size);
1001 /* Writes 32-bit integer X to the output file for writer W. */
1003 write_int (struct sfm_writer *w, int32_t x)
1005 write_bytes (w, &x, sizeof x);
1008 /* Converts NATIVE to the 64-bit format used in output files in
1011 convert_double_to_output_format (double native, uint8_t output[8])
1013 /* If "double" is not a 64-bit type, then convert it to a
1014 64-bit type. Otherwise just copy it. */
1015 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1016 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1018 memcpy (output, &native, sizeof native);
1021 /* Writes floating-point number X to the output file for writer
1024 write_float (struct sfm_writer *w, double x)
1027 convert_double_to_output_format (x, output);
1028 write_bytes (w, output, sizeof output);
1031 /* Writes contents of VALUE with the given WIDTH to W, padding
1032 with zeros to a multiple of 8 bytes.
1033 To avoid a branch, and because we don't actually need to
1034 support it, WIDTH must be no bigger than 8. */
1036 write_value (struct sfm_writer *w, const union value *value, int width)
1038 assert (width <= 8);
1040 write_float (w, value->f);
1043 write_bytes (w, value_str (value, width), width);
1044 write_zeros (w, 8 - width);
1048 /* Writes null-terminated STRING in a field of the given WIDTH to
1049 W. If STRING is longer than WIDTH, it is truncated; if WIDTH
1050 is narrowed, it is padded on the right with spaces. */
1052 write_string (struct sfm_writer *w, const char *string, size_t width)
1054 size_t data_bytes = MIN (strlen (string), width);
1055 size_t pad_bytes = width - data_bytes;
1056 write_bytes (w, string, data_bytes);
1057 while (pad_bytes-- > 0)
1058 putc (' ', w->file);
1061 /* Writes SIZE bytes of DATA to W's output file. */
1063 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1065 fwrite (data, 1, size, w->file);
1068 /* Writes N zeros to W's output file. */
1070 write_zeros (struct sfm_writer *w, size_t n)
1076 /* Writes N spaces to W's output file. */
1078 write_spaces (struct sfm_writer *w, size_t n)
1081 putc (' ', w->file);