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/string-array.h"
51 #include "libpspp/version.h"
53 #include "gl/xmemdup0.h"
54 #include "gl/minmax.h"
55 #include "gl/unlocked-io.h"
56 #include "gl/xalloc.h"
59 #define _(msgid) gettext (msgid)
60 #define N_(msgid) (msgid)
62 /* Compression bias used by PSPP. Values between (1 -
63 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
65 #define COMPRESSION_BIAS 100
67 /* System file writer. */
70 struct file_handle *fh; /* File handle. */
71 struct fh_lock *lock; /* Mutual exclusion for file. */
72 FILE *file; /* File stream. */
73 struct replace_file *rf; /* Ticket for replacing output file. */
75 bool compress; /* 1=compressed, 0=not compressed. */
76 casenumber case_cnt; /* Number of cases written so far. */
78 /* Compression buffering.
80 Compressed data is output as groups of 8 1-byte opcodes
81 followed by up to 8 (depending on the opcodes) 8-byte data
82 items. Data items and opcodes arrive at the same time but
83 must be reordered for writing to disk, thus a small amount
85 uint8_t opcodes[8]; /* Buffered opcodes. */
86 int opcode_cnt; /* Number of buffered opcodes. */
87 uint8_t data[8][8]; /* Buffered data. */
88 int data_cnt; /* Number of buffered data items. */
91 struct sfm_var *sfm_vars; /* Variables. */
92 size_t sfm_var_cnt; /* Number of variables. */
93 size_t segment_cnt; /* Number of variables including extra segments
94 for long string variables. */
97 static const struct casewriter_class sys_file_casewriter_class;
99 static void write_header (struct sfm_writer *, const struct dictionary *);
100 static void write_variable (struct sfm_writer *, const struct variable *);
101 static void write_value_labels (struct sfm_writer *, struct variable *,
103 static void write_integer_info_record (struct sfm_writer *,
104 const struct dictionary *);
105 static void write_float_info_record (struct sfm_writer *);
107 static void write_longvar_table (struct sfm_writer *w,
108 const struct dictionary *dict);
110 static void write_encoding_record (struct sfm_writer *w,
111 const struct dictionary *);
113 static void write_vls_length_table (struct sfm_writer *w,
114 const struct dictionary *dict);
116 static void write_long_string_value_labels (struct sfm_writer *,
117 const struct dictionary *);
119 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
122 static void write_variable_display_parameters (struct sfm_writer *w,
123 const struct dictionary *dict);
125 static void write_documents (struct sfm_writer *, const struct dictionary *);
127 static void write_data_file_attributes (struct sfm_writer *,
128 const struct dictionary *);
129 static void write_variable_attributes (struct sfm_writer *,
130 const struct dictionary *);
132 static void write_int (struct sfm_writer *, int32_t);
133 static inline void convert_double_to_output_format (double, uint8_t[8]);
134 static void write_float (struct sfm_writer *, double);
135 static void write_string (struct sfm_writer *, const char *, size_t);
136 static void write_utf8_string (struct sfm_writer *, const char *encoding,
137 const char *string, size_t width);
138 static void write_utf8_record (struct sfm_writer *, const char *encoding,
139 const struct string *content, int subtype);
140 static void write_string_record (struct sfm_writer *,
141 const struct substring content, int subtype);
142 static void write_bytes (struct sfm_writer *, const void *, size_t);
143 static void write_zeros (struct sfm_writer *, size_t);
144 static void write_spaces (struct sfm_writer *, size_t);
145 static void write_value (struct sfm_writer *, const union value *, int width);
147 static void write_case_uncompressed (struct sfm_writer *,
148 const struct ccase *);
149 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
150 static void flush_compressed (struct sfm_writer *);
151 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
152 static void put_cmp_number (struct sfm_writer *, double);
153 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
155 static bool write_error (const struct sfm_writer *);
156 static bool close_writer (struct sfm_writer *);
158 /* Returns default options for writing a system file. */
159 struct sfm_write_options
160 sfm_writer_default_options (void)
162 struct sfm_write_options opts;
163 opts.create_writeable = true;
164 opts.compress = settings_get_scompression ();
169 /* Opens the system file designated by file handle FH for writing
170 cases from dictionary D according to the given OPTS.
172 No reference to D is retained, so it may be modified or
173 destroyed at will after this function returns. D is not
174 modified by this function, except to assign short names. */
176 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
177 struct sfm_write_options opts)
179 struct sfm_writer *w;
185 if (opts.version != 2 && opts.version != 3)
187 msg (ME, _("Unknown system file version %d. Treating as version %d."),
192 /* Create and initialize writer. */
193 w = xmalloc (sizeof *w);
199 w->compress = opts.compress;
202 w->opcode_cnt = w->data_cnt = 0;
204 /* Figure out how to map in-memory case data to on-disk case
205 data. Also count the number of segments. Very long strings
206 occupy multiple segments, otherwise each variable only takes
208 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
211 /* Open file handle as an exclusive writer. */
212 /* TRANSLATORS: this fragment will be interpolated into
213 messages in fh_lock() that identify types of files. */
214 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
218 /* Create the file on disk. */
220 if (opts.create_writeable)
222 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
226 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
227 fh_get_file_name (fh), strerror (errno));
231 /* Write the file header. */
234 /* Write basic variable info. */
235 short_names_assign (d);
236 for (i = 0; i < dict_get_var_cnt (d); i++)
237 write_variable (w, dict_get_var (d, i));
239 /* Write out value labels. */
241 for (i = 0; i < dict_get_var_cnt (d); i++)
243 struct variable *v = dict_get_var (d, i);
245 write_value_labels (w, v, idx);
246 idx += sfm_width_to_octs (var_get_width (v));
249 if (dict_get_document_line_cnt (d) > 0)
250 write_documents (w, d);
252 write_integer_info_record (w, d);
253 write_float_info_record (w);
255 write_mrsets (w, d, true);
257 write_variable_display_parameters (w, d);
259 if (opts.version >= 3)
260 write_longvar_table (w, d);
262 write_vls_length_table (w, d);
264 write_long_string_value_labels (w, d);
266 if (attrset_count (dict_get_attributes (d)))
267 write_data_file_attributes (w, d);
268 write_variable_attributes (w, d);
270 write_mrsets (w, d, false);
272 write_encoding_record (w, d);
274 /* Write end-of-headers record. */
281 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
288 /* Returns value of X truncated to two least-significant digits. */
299 /* Calculates the offset of data for TARGET_VAR from the
300 beginning of each case's data for dictionary D. The return
301 value is in "octs" (8-byte units). */
303 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
309 for (i = 0; i < dict_get_var_cnt (d); i++)
311 struct variable *var = dict_get_var (d, i);
312 if (var == target_var)
314 oct_idx += sfm_width_to_octs (var_get_width (var));
319 /* Write the sysfile_header header to system file W. */
321 write_header (struct sfm_writer *w, const struct dictionary *d)
324 char creation_date[10];
325 char creation_time[9];
326 const char *file_label;
327 struct variable *weight;
331 /* Record-type code. */
332 write_string (w, "$FL2", 4);
334 /* Product identification. */
335 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
336 version, host_system);
337 write_string (w, prod_name, 60);
342 /* Number of `union value's per case. */
343 write_int (w, calc_oct_idx (d, NULL));
346 write_int (w, w->compress);
348 /* Weight variable. */
349 weight = dict_get_weight (d);
350 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
352 /* Number of cases. We don't know this in advance, so we write
353 -1 to indicate an unknown number of cases. Later we can
354 come back and overwrite it with the true value. */
357 /* Compression bias. */
358 write_float (w, COMPRESSION_BIAS);
360 /* Creation date and time. */
361 if (time (&t) == (time_t) -1)
363 strcpy (creation_date, "01 Jan 70");
364 strcpy (creation_time, "00:00:00");
368 static const char *const month_name[12] =
370 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
371 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
373 struct tm *tmp = localtime (&t);
374 int day = rerange (tmp->tm_mday);
375 int mon = rerange (tmp->tm_mon + 1);
376 int year = rerange (tmp->tm_year);
377 int hour = rerange (tmp->tm_hour + 1);
378 int min = rerange (tmp->tm_min + 1);
379 int sec = rerange (tmp->tm_sec + 1);
381 snprintf (creation_date, sizeof creation_date,
382 "%02d %s %02d", day, month_name[mon - 1], year);
383 snprintf (creation_time, sizeof creation_time,
384 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
386 write_string (w, creation_date, 9);
387 write_string (w, creation_time, 8);
390 file_label = dict_get_label (d);
391 if (file_label == NULL)
393 write_utf8_string (w, dict_get_encoding (d), file_label, 64);
399 /* Write format spec FMT to W, after adjusting it to be
400 compatible with the given WIDTH. */
402 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
404 assert (fmt_check_output (&fmt));
405 assert (sfm_width_to_segments (width) == 1);
408 fmt_resize (&fmt, width);
409 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
412 /* Write a string continuation variable record for each 8-byte
413 section beyond the initial 8 bytes, for a variable of the
416 write_variable_continuation_records (struct sfm_writer *w, int width)
420 assert (sfm_width_to_segments (width) == 1);
421 for (position = 8; position < width; position += 8)
423 write_int (w, 2); /* Record type. */
424 write_int (w, -1); /* Width. */
425 write_int (w, 0); /* No variable label. */
426 write_int (w, 0); /* No missing values. */
427 write_int (w, 0); /* Print format. */
428 write_int (w, 0); /* Write format. */
429 write_zeros (w, 8); /* Name. */
433 /* Write the variable record(s) for variable V to system file
436 write_variable (struct sfm_writer *w, const struct variable *v)
438 int width = var_get_width (v);
439 int segment_cnt = sfm_width_to_segments (width);
440 int seg0_width = sfm_segment_alloc_width (width, 0);
441 const char *encoding = var_get_encoding (v);
442 struct missing_values mv;
449 write_int (w, seg0_width);
451 /* Variable has a variable label? */
452 write_int (w, var_has_label (v));
454 /* Number of missing values. If there is a range, then the
455 range counts as 2 missing values and causes the number to be
457 mv_copy (&mv, var_get_missing_values (v));
458 if (mv_get_width (&mv) > 8)
460 if (mv_has_range (&mv))
461 write_int (w, -2 - mv_n_values (&mv));
463 write_int (w, mv_n_values (&mv));
465 /* Print and write formats. */
466 write_format (w, *var_get_print_format (v), seg0_width);
467 write_format (w, *var_get_write_format (v), seg0_width);
470 The full name is in a translation table written
472 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
475 if (var_has_label (v))
477 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
478 size_t label_len = MIN (strlen (label), 255);
479 size_t padded_len = ROUND_UP (label_len, 4);
480 write_int (w, label_len);
481 write_string (w, label, padded_len);
485 /* Write the missing values, if any, range first. */
486 if (mv_has_range (&mv))
489 mv_get_range (&mv, &x, &y);
493 for (i = 0; i < mv_n_values (&mv); i++)
494 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
496 write_variable_continuation_records (w, seg0_width);
498 /* Write additional segments for very long string variables. */
499 for (i = 1; i < segment_cnt; i++)
501 int seg_width = sfm_segment_alloc_width (width, i);
502 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
504 write_int (w, 2); /* Variable record. */
505 write_int (w, seg_width); /* Width. */
506 write_int (w, 0); /* No variable label. */
507 write_int (w, 0); /* No missing values. */
508 write_format (w, fmt, seg_width); /* Print format. */
509 write_format (w, fmt, seg_width); /* Write format. */
510 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
512 write_variable_continuation_records (w, seg_width);
518 /* Writes the value labels for variable V having system file
519 variable index IDX to system file W.
521 Value labels for long string variables are written separately,
522 by write_long_string_value_labels. */
524 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
526 const struct val_labs *val_labs;
527 const struct val_lab **labels;
531 val_labs = var_get_value_labels (v);
532 n_labels = val_labs_count (val_labs);
533 if (n_labels == 0 || var_get_width (v) > 8)
536 /* Value label record. */
537 write_int (w, 3); /* Record type. */
538 write_int (w, val_labs_count (val_labs));
539 labels = val_labs_sorted (val_labs);
540 for (i = 0; i < n_labels; i++)
542 const struct val_lab *vl = labels[i];
543 char *label = recode_string (var_get_encoding (v), UTF8,
544 val_lab_get_escaped_label (vl), -1);
545 uint8_t len = MIN (strlen (label), 255);
547 write_value (w, val_lab_get_value (vl), var_get_width (v));
548 write_bytes (w, &len, 1);
549 write_bytes (w, label, len);
550 write_zeros (w, REM_RND_UP (len + 1, 8));
555 /* Value label variable record. */
556 write_int (w, 4); /* Record type. */
557 write_int (w, 1); /* Number of variables. */
558 write_int (w, idx + 1); /* Variable's dictionary index. */
561 /* Writes record type 6, document record. */
563 write_documents (struct sfm_writer *w, const struct dictionary *d)
565 const struct string_array *docs = dict_get_documents (d);
566 const char *enc = dict_get_encoding (d);
569 write_int (w, 6); /* Record type. */
570 write_int (w, docs->n);
571 for (i = 0; i < docs->n; i++)
573 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
574 size_t s_len = strlen (s);
575 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
577 write_bytes (w, s, write_len);
578 write_spaces (w, DOC_LINE_LENGTH - write_len);
584 put_attrset (struct string *string, const struct attrset *attrs)
586 const struct attribute *attr;
587 struct attrset_iterator i;
589 for (attr = attrset_first (attrs, &i); attr != NULL;
590 attr = attrset_next (attrs, &i))
592 size_t n_values = attribute_get_n_values (attr);
595 ds_put_cstr (string, attribute_get_name (attr));
596 ds_put_byte (string, '(');
597 for (j = 0; j < n_values; j++)
598 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
599 ds_put_byte (string, ')');
604 write_data_file_attributes (struct sfm_writer *w,
605 const struct dictionary *d)
607 struct string s = DS_EMPTY_INITIALIZER;
608 put_attrset (&s, dict_get_attributes (d));
609 write_utf8_record (w, dict_get_encoding (d), &s, 17);
614 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
616 struct string s = DS_EMPTY_INITIALIZER;
617 size_t n_vars = dict_get_var_cnt (d);
618 size_t n_attrsets = 0;
621 for (i = 0; i < n_vars; i++)
623 struct variable *v = dict_get_var (d, i);
624 struct attrset *attrs = var_get_attributes (v);
625 if (attrset_count (attrs))
628 ds_put_byte (&s, '/');
629 ds_put_format (&s, "%s:", var_get_name (v));
630 put_attrset (&s, attrs);
634 write_utf8_record (w, dict_get_encoding (d), &s, 18);
638 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
639 SPSS before release 14, otherwise writes sets supported only by later
642 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
645 const char *encoding = dict_get_encoding (dict);
646 struct string s = DS_EMPTY_INITIALIZER;
650 n_mrsets = dict_get_n_mrsets (dict);
654 for (i = 0; i < n_mrsets; i++)
656 const struct mrset *mrset = dict_get_mrset (dict, i);
660 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
664 name = recode_string (encoding, "UTF-8", mrset->name, -1);
665 ds_put_format (&s, "%s=", name);
668 if (mrset->type == MRSET_MD)
672 if (mrset->cat_source == MRSET_COUNTEDVALUES)
673 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
675 ds_put_byte (&s, 'D');
677 if (mrset->width == 0)
678 counted = xasprintf ("%.0f", mrset->counted.f);
680 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
682 ds_put_format (&s, "%zu %s", strlen (counted), counted);
686 ds_put_byte (&s, 'C');
687 ds_put_byte (&s, ' ');
689 if (mrset->label && !mrset->label_from_var_label)
691 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
692 ds_put_format (&s, "%zu %s", strlen (label), label);
696 ds_put_cstr (&s, "0 ");
698 for (j = 0; j < mrset->n_vars; j++)
700 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
701 char *short_name = recode_string (encoding, "UTF-8",
702 short_name_utf8, -1);
703 str_lowercase (short_name);
704 ds_put_format (&s, " %s", short_name);
707 ds_put_byte (&s, '\n');
710 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
714 /* Write the alignment, width and scale values. */
716 write_variable_display_parameters (struct sfm_writer *w,
717 const struct dictionary *dict)
721 write_int (w, 7); /* Record type. */
722 write_int (w, 11); /* Record subtype. */
723 write_int (w, 4); /* Data item (int32) size. */
724 write_int (w, w->segment_cnt * 3); /* Number of data items. */
726 for (i = 0; i < dict_get_var_cnt (dict); ++i)
728 struct variable *v = dict_get_var (dict, i);
729 int width = var_get_width (v);
730 int segment_cnt = sfm_width_to_segments (width);
731 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
732 : var_get_measure (v) == MEASURE_ORDINAL ? 2
734 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
735 : var_get_alignment (v) == ALIGN_RIGHT ? 1
739 for (i = 0; i < segment_cnt; i++)
741 int width_left = width - sfm_segment_effective_offset (width, i);
742 write_int (w, measure);
743 write_int (w, (i == 0 ? var_get_display_width (v)
744 : var_default_display_width (width_left)));
745 write_int (w, alignment);
750 /* Writes the table of lengths for very long string variables. */
752 write_vls_length_table (struct sfm_writer *w,
753 const struct dictionary *dict)
758 ds_init_empty (&map);
759 for (i = 0; i < dict_get_var_cnt (dict); ++i)
761 const struct variable *v = dict_get_var (dict, i);
762 if (sfm_width_to_segments (var_get_width (v)) > 1)
763 ds_put_format (&map, "%s=%05d%c\t",
764 var_get_short_name (v, 0), var_get_width (v), 0);
766 if (!ds_is_empty (&map))
767 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
773 write_long_string_value_labels (struct sfm_writer *w,
774 const struct dictionary *dict)
776 size_t n_vars = dict_get_var_cnt (dict);
780 /* Figure out the size in advance. */
782 for (i = 0; i < n_vars; i++)
784 struct variable *var = dict_get_var (dict, i);
785 const struct val_labs *val_labs = var_get_value_labels (var);
786 const char *encoding = var_get_encoding (var);
787 int width = var_get_width (var);
788 const struct val_lab *val_lab;
790 if (val_labs_count (val_labs) == 0 || width < 9)
794 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
795 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
796 val_lab = val_labs_next (val_labs, val_lab))
799 size += recode_string_len (encoding, "UTF-8",
800 val_lab_get_escaped_label (val_lab), -1);
806 write_int (w, 7); /* Record type. */
807 write_int (w, 21); /* Record subtype */
808 write_int (w, 1); /* Data item (byte) size. */
809 write_int (w, size); /* Number of data items. */
811 start = ftello (w->file);
812 for (i = 0; i < n_vars; i++)
814 struct variable *var = dict_get_var (dict, i);
815 const struct val_labs *val_labs = var_get_value_labels (var);
816 const char *encoding = var_get_encoding (var);
817 int width = var_get_width (var);
818 const struct val_lab *val_lab;
821 if (val_labs_count (val_labs) == 0 || width < 9)
824 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
825 write_int (w, strlen (var_name));
826 write_bytes (w, var_name, strlen (var_name));
829 write_int (w, width);
830 write_int (w, val_labs_count (val_labs));
831 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
832 val_lab = val_labs_next (val_labs, val_lab))
837 write_int (w, width);
838 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
841 label = recode_string (var_get_encoding (var), "UTF-8",
842 val_lab_get_escaped_label (val_lab), -1);
843 len = strlen (label);
845 write_bytes (w, label, len);
849 assert (ftello (w->file) == start + size);
853 write_encoding_record (struct sfm_writer *w,
854 const struct dictionary *d)
856 /* IANA says "...character set names may be up to 40 characters taken
857 from the printable characters of US-ASCII," so character set names
858 don't need to be recoded to be in UTF-8.
860 We convert encoding names to uppercase because SPSS writes encoding
861 names in uppercase. */
862 char *encoding = xstrdup (dict_get_encoding (d));
863 str_uppercase (encoding);
864 write_string_record (w, ss_cstr (encoding), 20);
868 /* Writes the long variable name table. */
870 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
875 ds_init_empty (&map);
876 for (i = 0; i < dict_get_var_cnt (dict); i++)
878 struct variable *v = dict_get_var (dict, i);
880 ds_put_byte (&map, '\t');
881 ds_put_format (&map, "%s=%s",
882 var_get_short_name (v, 0), var_get_name (v));
884 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
888 /* Write integer information record. */
890 write_integer_info_record (struct sfm_writer *w,
891 const struct dictionary *d)
893 int version_component[3];
897 /* Parse the version string. */
898 memset (version_component, 0, sizeof version_component);
899 sscanf (bare_version, "%d.%d.%d",
900 &version_component[0], &version_component[1], &version_component[2]);
902 /* Figure out the floating-point format. */
903 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
904 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
906 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
908 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
913 /* Choose codepage. */
914 codepage = sys_get_codepage_from_encoding (dict_get_encoding (d));
917 /* Default to "7-bit ASCII" if the codepage number is unknown, because
918 many files use this codepage number regardless of their actual
924 write_int (w, 7); /* Record type. */
925 write_int (w, 3); /* Record subtype. */
926 write_int (w, 4); /* Data item (int32) size. */
927 write_int (w, 8); /* Number of data items. */
928 write_int (w, version_component[0]);
929 write_int (w, version_component[1]);
930 write_int (w, version_component[2]);
931 write_int (w, -1); /* Machine code. */
932 write_int (w, float_format);
933 write_int (w, 1); /* Compression code. */
934 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
935 write_int (w, codepage);
938 /* Write floating-point information record. */
940 write_float_info_record (struct sfm_writer *w)
942 write_int (w, 7); /* Record type. */
943 write_int (w, 4); /* Record subtype. */
944 write_int (w, 8); /* Data item (flt64) size. */
945 write_int (w, 3); /* Number of data items. */
946 write_float (w, SYSMIS); /* System-missing value. */
947 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
948 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
951 /* Writes case C to system file W. */
953 sys_file_casewriter_write (struct casewriter *writer, void *w_,
956 struct sfm_writer *w = w_;
958 if (ferror (w->file))
960 casewriter_force_error (writer);
968 write_case_uncompressed (w, c);
970 write_case_compressed (w, c);
975 /* Destroys system file writer W. */
977 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
979 struct sfm_writer *w = w_;
980 if (!close_writer (w))
981 casewriter_force_error (writer);
984 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
986 write_error (const struct sfm_writer *writer)
988 return ferror (writer->file);
991 /* Closes a system file after we're done with it.
992 Returns true if successful, false if an I/O error occurred. */
994 close_writer (struct sfm_writer *w)
1002 if (w->file != NULL)
1005 if (w->opcode_cnt > 0)
1006 flush_compressed (w);
1009 ok = !write_error (w);
1011 /* Seek back to the beginning and update the number of cases.
1012 This is just a courtesy to later readers, so there's no need
1013 to check return values or report errors. */
1014 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1016 write_int (w, w->case_cnt);
1020 if (fclose (w->file) == EOF)
1024 msg (ME, _("An I/O error occurred writing system file `%s'."),
1025 fh_get_file_name (w->fh));
1027 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1031 fh_unlock (w->lock);
1040 /* System file writer casewriter class. */
1041 static const struct casewriter_class sys_file_casewriter_class =
1043 sys_file_casewriter_write,
1044 sys_file_casewriter_destroy,
1048 /* Writes case C to system file W, without compressing it. */
1050 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1054 for (i = 0; i < w->sfm_var_cnt; i++)
1056 struct sfm_var *v = &w->sfm_vars[i];
1058 if (v->var_width == 0)
1059 write_float (w, case_num_idx (c, v->case_index));
1062 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1064 write_spaces (w, v->padding);
1069 /* Writes case C to system file W, with compression. */
1071 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1075 for (i = 0; i < w->sfm_var_cnt; i++)
1077 struct sfm_var *v = &w->sfm_vars[i];
1079 if (v->var_width == 0)
1081 double d = case_num_idx (c, v->case_index);
1083 put_cmp_opcode (w, 255);
1084 else if (d >= 1 - COMPRESSION_BIAS
1085 && d <= 251 - COMPRESSION_BIAS
1087 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1090 put_cmp_opcode (w, 253);
1091 put_cmp_number (w, d);
1096 int offset = v->offset;
1099 /* This code properly deals with a width that is not a
1100 multiple of 8, by ensuring that the final partial
1101 oct (8 byte unit) is treated as padded with spaces
1103 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1105 const void *data = case_str_idx (c, v->case_index) + offset;
1106 int chunk_size = MIN (width, 8);
1107 if (!memcmp (data, " ", chunk_size))
1108 put_cmp_opcode (w, 254);
1111 put_cmp_opcode (w, 253);
1112 put_cmp_string (w, data, chunk_size);
1116 /* This code deals properly with padding that is not a
1117 multiple of 8 bytes, by discarding the remainder,
1118 which was already effectively padded with spaces in
1119 the previous loop. (Note that v->width + v->padding
1120 is always a multiple of 8.) */
1121 for (padding = v->padding / 8; padding > 0; padding--)
1122 put_cmp_opcode (w, 254);
1127 /* Flushes buffered compressed opcodes and data to W.
1128 The compression buffer must not be empty. */
1130 flush_compressed (struct sfm_writer *w)
1132 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1134 write_bytes (w, w->opcodes, w->opcode_cnt);
1135 write_zeros (w, 8 - w->opcode_cnt);
1137 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1139 w->opcode_cnt = w->data_cnt = 0;
1142 /* Appends OPCODE to the buffered set of compression opcodes in
1143 W. Flushes the compression buffer beforehand if necessary. */
1145 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1147 if (w->opcode_cnt >= 8)
1148 flush_compressed (w);
1150 w->opcodes[w->opcode_cnt++] = opcode;
1153 /* Appends NUMBER to the buffered compression data in W. The
1154 buffer must not be full; the way to assure that is to call
1155 this function only just after a call to put_cmp_opcode, which
1156 will flush the buffer as necessary. */
1158 put_cmp_number (struct sfm_writer *w, double number)
1160 assert (w->opcode_cnt > 0);
1161 assert (w->data_cnt < 8);
1163 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1166 /* Appends SIZE bytes of DATA to the buffered compression data in
1167 W, followed by enough spaces to pad the output data to exactly
1168 8 bytes (thus, SIZE must be no greater than 8). The buffer
1169 must not be full; the way to assure that is to call this
1170 function only just after a call to put_cmp_opcode, which will
1171 flush the buffer as necessary. */
1173 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1175 assert (w->opcode_cnt > 0);
1176 assert (w->data_cnt < 8);
1179 memset (w->data[w->data_cnt], ' ', 8);
1180 memcpy (w->data[w->data_cnt], data, size);
1184 /* Writes 32-bit integer X to the output file for writer W. */
1186 write_int (struct sfm_writer *w, int32_t x)
1188 write_bytes (w, &x, sizeof x);
1191 /* Converts NATIVE to the 64-bit format used in output files in
1194 convert_double_to_output_format (double native, uint8_t output[8])
1196 /* If "double" is not a 64-bit type, then convert it to a
1197 64-bit type. Otherwise just copy it. */
1198 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1199 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1201 memcpy (output, &native, sizeof native);
1204 /* Writes floating-point number X to the output file for writer
1207 write_float (struct sfm_writer *w, double x)
1210 convert_double_to_output_format (x, output);
1211 write_bytes (w, output, sizeof output);
1214 /* Writes contents of VALUE with the given WIDTH to W, padding
1215 with zeros to a multiple of 8 bytes.
1216 To avoid a branch, and because we don't actually need to
1217 support it, WIDTH must be no bigger than 8. */
1219 write_value (struct sfm_writer *w, const union value *value, int width)
1221 assert (width <= 8);
1223 write_float (w, value->f);
1226 write_bytes (w, value_str (value, width), width);
1227 write_zeros (w, 8 - width);
1231 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1232 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1233 is padded on the right with spaces. */
1235 write_string (struct sfm_writer *w, const char *string, size_t width)
1237 size_t data_bytes = MIN (strlen (string), width);
1238 size_t pad_bytes = width - data_bytes;
1239 write_bytes (w, string, data_bytes);
1240 while (pad_bytes-- > 0)
1241 putc (' ', w->file);
1244 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1245 recoded version in a field of the given WIDTH to W. The string is truncated
1246 or padded on the right with spaces to exactly WIDTH bytes. */
1248 write_utf8_string (struct sfm_writer *w, const char *encoding,
1249 const char *string, size_t width)
1251 char *s = recode_string (encoding, "UTF-8", string, -1);
1252 write_string (w, s, width);
1256 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1257 from UTF-8 encoded into ENCODING. */
1259 write_utf8_record (struct sfm_writer *w, const char *encoding,
1260 const struct string *content, int subtype)
1264 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1265 write_string_record (w, s, subtype);
1269 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1272 write_string_record (struct sfm_writer *w,
1273 const struct substring content, int subtype)
1276 write_int (w, subtype);
1278 write_int (w, ss_length (content));
1279 write_bytes (w, ss_data (content), ss_length (content));
1282 /* Writes SIZE bytes of DATA to W's output file. */
1284 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1286 fwrite (data, 1, size, w->file);
1289 /* Writes N zeros to W's output file. */
1291 write_zeros (struct sfm_writer *w, size_t n)
1297 /* Writes N spaces to W's output file. */
1299 write_spaces (struct sfm_writer *w, size_t n)
1302 putc (' ', w->file);