1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-2000, 2006-2014 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"
30 #include "data/attributes.h"
31 #include "data/case.h"
32 #include "data/casewriter-provider.h"
33 #include "data/casewriter.h"
34 #include "data/dictionary.h"
35 #include "data/file-handle-def.h"
36 #include "data/file-name.h"
37 #include "data/format.h"
38 #include "data/make-file.h"
39 #include "data/missing-values.h"
40 #include "data/mrset.h"
41 #include "data/settings.h"
42 #include "data/short-names.h"
43 #include "data/value-labels.h"
44 #include "data/variable.h"
45 #include "libpspp/float-format.h"
46 #include "libpspp/i18n.h"
47 #include "libpspp/integer-format.h"
48 #include "libpspp/message.h"
49 #include "libpspp/misc.h"
50 #include "libpspp/str.h"
51 #include "libpspp/string-array.h"
52 #include "libpspp/version.h"
54 #include "gl/xmemdup0.h"
55 #include "gl/minmax.h"
56 #include "gl/unlocked-io.h"
57 #include "gl/xalloc.h"
60 #define _(msgid) gettext (msgid)
61 #define N_(msgid) (msgid)
63 /* Compression bias used by PSPP. Values between (1 -
64 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
66 #define COMPRESSION_BIAS 100
68 /* System file writer. */
71 struct file_handle *fh; /* File handle. */
72 struct fh_lock *lock; /* Mutual exclusion for file. */
73 FILE *file; /* File stream. */
74 struct replace_file *rf; /* Ticket for replacing output file. */
76 enum any_compression compression;
77 casenumber case_cnt; /* Number of cases written so far. */
78 uint8_t space; /* ' ' in the file's character encoding. */
80 /* Simple compression buffering.
82 Compressed data is output as a series of 8-byte elements, with 1 to 9
83 such elements clustered together. The first element in a cluster is 8
84 1-byte opcodes. Some opcodes call for an additional element in the
85 cluster (hence, if there are eight such opcodes, then the cluster
86 contains a full 9 elements).
88 cbuf[] holds a cluster at a time. */
90 int n_opcodes; /* Number of opcodes in cbuf[0] so far. */
91 int n_elements; /* Number of elements in cbuf[] so far. */
93 /* ZLIB compression. */
94 z_stream zstream; /* ZLIB deflater. */
96 struct zblock *blocks;
97 size_t n_blocks, allocated_blocks;
100 struct sfm_var *sfm_vars; /* Variables. */
101 size_t sfm_var_cnt; /* Number of variables. */
102 size_t segment_cnt; /* Number of variables including extra segments
103 for long string variables. */
108 unsigned int uncompressed_size;
109 unsigned int compressed_size;
112 static const struct casewriter_class sys_file_casewriter_class;
114 static void write_header (struct sfm_writer *, const struct dictionary *);
115 static void write_variable (struct sfm_writer *, const struct variable *);
116 static void write_value_labels (struct sfm_writer *,
117 const struct dictionary *);
118 static void write_integer_info_record (struct sfm_writer *,
119 const struct dictionary *);
120 static void write_float_info_record (struct sfm_writer *);
122 static void write_longvar_table (struct sfm_writer *w,
123 const struct dictionary *dict);
125 static void write_encoding_record (struct sfm_writer *w,
126 const struct dictionary *);
128 static void write_vls_length_table (struct sfm_writer *w,
129 const struct dictionary *dict);
131 static void write_long_string_value_labels (struct sfm_writer *,
132 const struct dictionary *);
133 static void write_long_string_missing_values (struct sfm_writer *,
134 const struct dictionary *);
136 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
139 static void write_variable_display_parameters (struct sfm_writer *w,
140 const struct dictionary *dict);
142 static void write_documents (struct sfm_writer *, const struct dictionary *);
144 static void write_data_file_attributes (struct sfm_writer *,
145 const struct dictionary *);
146 static void write_variable_attributes (struct sfm_writer *,
147 const struct dictionary *);
149 static void write_int (struct sfm_writer *, int32_t);
150 static void write_int64 (struct sfm_writer *, int64_t);
151 static inline void convert_double_to_output_format (double, uint8_t[8]);
152 static void write_float (struct sfm_writer *, double);
153 static void write_string (struct sfm_writer *, const char *, size_t);
154 static void write_utf8_string (struct sfm_writer *, const char *encoding,
155 const char *string, size_t width);
156 static void write_utf8_record (struct sfm_writer *, const char *encoding,
157 const struct string *content, int subtype);
158 static void write_string_record (struct sfm_writer *,
159 const struct substring content, int subtype);
160 static void write_bytes (struct sfm_writer *, const void *, size_t);
161 static void write_zeros (struct sfm_writer *, size_t);
162 static void write_spaces (struct sfm_writer *, size_t);
163 static void write_value (struct sfm_writer *, const union value *, int width);
165 static void write_case_uncompressed (struct sfm_writer *,
166 const struct ccase *);
167 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
168 static void flush_compressed (struct sfm_writer *);
169 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
170 static void put_cmp_number (struct sfm_writer *, double);
171 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
173 static bool start_zstream (struct sfm_writer *);
174 static void finish_zstream (struct sfm_writer *);
175 static void write_ztrailer (struct sfm_writer *);
177 static bool write_error (const struct sfm_writer *);
178 static bool close_writer (struct sfm_writer *);
180 /* Returns default options for writing a system file. */
181 struct sfm_write_options
182 sfm_writer_default_options (void)
184 struct sfm_write_options opts;
185 opts.compression = (settings_get_scompression ()
188 opts.create_writeable = true;
193 /* Opens the system file designated by file handle FH for writing
194 cases from dictionary D according to the given OPTS.
196 No reference to D is retained, so it may be modified or
197 destroyed at will after this function returns. D is not
198 modified by this function, except to assign short names. */
200 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
201 struct sfm_write_options opts)
203 struct encoding_info encoding_info;
204 struct sfm_writer *w;
209 if (opts.version != 2 && opts.version != 3)
211 msg (ME, _("Unknown system file version %d. Treating as version %d."),
216 /* Create and initialize writer. */
217 w = xzalloc (sizeof *w);
223 /* Use the requested compression, except that no EBCDIC-based ZLIB compressed
224 files have been observed, so drop back to simple compression for those
226 w->compression = opts.compression;
227 if (w->compression == ANY_COMP_ZLIB
228 && is_encoding_ebcdic_compatible (dict_get_encoding (d)))
229 w->compression = ANY_COMP_SIMPLE;
233 w->n_opcodes = w->n_elements = 0;
234 memset (w->cbuf[0], 0, 8);
236 /* Figure out how to map in-memory case data to on-disk case
237 data. Also count the number of segments. Very long strings
238 occupy multiple segments, otherwise each variable only takes
240 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
243 /* Open file handle as an exclusive writer. */
244 /* TRANSLATORS: this fragment will be interpolated into
245 messages in fh_lock() that identify types of files. */
246 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
250 /* Create the file on disk. */
252 if (opts.create_writeable)
254 w->rf = replace_file_start (fh, "wb", mode, &w->file);
257 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
258 fh_get_file_name (fh), strerror (errno));
262 get_encoding_info (&encoding_info, dict_get_encoding (d));
263 w->space = encoding_info.space[0];
265 /* Write the file header. */
268 /* Write basic variable info. */
269 short_names_assign (d);
270 for (i = 0; i < dict_get_var_cnt (d); i++)
271 write_variable (w, dict_get_var (d, i));
273 write_value_labels (w, d);
275 if (dict_get_document_line_cnt (d) > 0)
276 write_documents (w, d);
278 write_integer_info_record (w, d);
279 write_float_info_record (w);
281 write_mrsets (w, d, true);
283 write_variable_display_parameters (w, d);
285 if (opts.version >= 3)
286 write_longvar_table (w, d);
288 write_vls_length_table (w, d);
290 write_long_string_value_labels (w, d);
291 write_long_string_missing_values (w, d);
293 if (opts.version >= 3)
295 if (attrset_count (dict_get_attributes (d)))
296 write_data_file_attributes (w, d);
297 write_variable_attributes (w, d);
300 write_mrsets (w, d, false);
302 write_encoding_record (w, d);
304 /* Write end-of-headers record. */
308 if (w->compression == ANY_COMP_ZLIB)
310 w->zstream.zalloc = Z_NULL;
311 w->zstream.zfree = Z_NULL;
312 w->zstream.opaque = Z_NULL;
313 w->zstart = ftello (w->file);
315 write_int64 (w, w->zstart);
325 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
332 /* Returns value of X truncated to two least-significant digits. */
343 /* Calculates the offset of data for TARGET_VAR from the
344 beginning of each case's data for dictionary D. The return
345 value is in "octs" (8-byte units). */
347 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
353 for (i = 0; i < dict_get_var_cnt (d); i++)
355 struct variable *var = dict_get_var (d, i);
356 if (var == target_var)
358 oct_idx += sfm_width_to_octs (var_get_width (var));
363 /* Write the sysfile_header header to system file W. */
365 write_header (struct sfm_writer *w, const struct dictionary *d)
367 const char *dict_encoding = dict_get_encoding (d);
369 char creation_date[10];
370 char creation_time[9];
371 const char *file_label;
372 struct variable *weight;
376 /* Record-type code. */
377 if (is_encoding_ebcdic_compatible (dict_encoding))
378 write_string (w, EBCDIC_MAGIC, 4);
379 else if (w->compression == ANY_COMP_ZLIB)
380 write_string (w, ASCII_ZMAGIC, 4);
382 write_string (w, ASCII_MAGIC, 4);
384 /* Product identification. */
385 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
386 version, host_system);
387 write_utf8_string (w, dict_encoding, prod_name, 60);
392 /* Number of `union value's per case. */
393 write_int (w, calc_oct_idx (d, NULL));
396 write_int (w, (w->compression == ANY_COMP_NONE ? 0
397 : w->compression == ANY_COMP_SIMPLE ? 1
400 /* Weight variable. */
401 weight = dict_get_weight (d);
402 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
404 /* Number of cases. We don't know this in advance, so we write
405 -1 to indicate an unknown number of cases. Later we can
406 come back and overwrite it with the true value. */
409 /* Compression bias. */
410 write_float (w, COMPRESSION_BIAS);
412 /* Creation date and time. */
413 if (time (&t) == (time_t) -1)
415 strcpy (creation_date, "01 Jan 70");
416 strcpy (creation_time, "00:00:00");
420 static const char *const month_name[12] =
422 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
423 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
425 struct tm *tmp = localtime (&t);
426 int day = rerange (tmp->tm_mday);
427 int mon = rerange (tmp->tm_mon + 1);
428 int year = rerange (tmp->tm_year);
429 int hour = rerange (tmp->tm_hour + 1);
430 int min = rerange (tmp->tm_min + 1);
431 int sec = rerange (tmp->tm_sec + 1);
433 snprintf (creation_date, sizeof creation_date,
434 "%02d %s %02d", day, month_name[mon - 1], year);
435 snprintf (creation_time, sizeof creation_time,
436 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
438 write_utf8_string (w, dict_encoding, creation_date, 9);
439 write_utf8_string (w, dict_encoding, creation_time, 8);
442 file_label = dict_get_label (d);
443 if (file_label == NULL)
445 write_utf8_string (w, dict_encoding, file_label, 64);
451 /* Write format spec FMT to W, after adjusting it to be
452 compatible with the given WIDTH. */
454 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
456 assert (fmt_check_output (&fmt));
457 assert (sfm_width_to_segments (width) == 1);
460 fmt_resize (&fmt, width);
461 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
464 /* Write a string continuation variable record for each 8-byte
465 section beyond the initial 8 bytes, for a variable of the
468 write_variable_continuation_records (struct sfm_writer *w, int width)
472 assert (sfm_width_to_segments (width) == 1);
473 for (position = 8; position < width; position += 8)
475 write_int (w, 2); /* Record type. */
476 write_int (w, -1); /* Width. */
477 write_int (w, 0); /* No variable label. */
478 write_int (w, 0); /* No missing values. */
479 write_int (w, 0); /* Print format. */
480 write_int (w, 0); /* Write format. */
481 write_zeros (w, 8); /* Name. */
485 /* Write the variable record(s) for variable V to system file
488 write_variable (struct sfm_writer *w, const struct variable *v)
490 int width = var_get_width (v);
491 int segment_cnt = sfm_width_to_segments (width);
492 int seg0_width = sfm_segment_alloc_width (width, 0);
493 const char *encoding = var_get_encoding (v);
500 write_int (w, seg0_width);
502 /* Variable has a variable label? */
503 write_int (w, var_has_label (v));
505 /* Number of missing values. If there is a range, then the
506 range counts as 2 missing values and causes the number to be
509 Missing values for long string variables are written in a separate
511 if (width <= MAX_SHORT_STRING)
513 const struct missing_values *mv = var_get_missing_values (v);
514 if (mv_has_range (mv))
515 write_int (w, -2 - mv_n_values (mv));
517 write_int (w, mv_n_values (mv));
522 /* Print and write formats. */
523 write_format (w, *var_get_print_format (v), seg0_width);
524 write_format (w, *var_get_write_format (v), seg0_width);
527 The full name is in a translation table written
529 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
532 if (var_has_label (v))
534 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
535 size_t label_len = MIN (strlen (label), 255);
536 size_t padded_len = ROUND_UP (label_len, 4);
537 write_int (w, label_len);
538 write_string (w, label, padded_len);
542 /* Write the missing values, if any, range first. */
543 if (width <= MAX_SHORT_STRING)
545 const struct missing_values *mv = var_get_missing_values (v);
546 if (mv_has_range (mv))
549 mv_get_range (mv, &x, &y);
553 for (i = 0; i < mv_n_values (mv); i++)
554 write_value (w, mv_get_value (mv, i), width);
557 write_variable_continuation_records (w, seg0_width);
559 /* Write additional segments for very long string variables. */
560 for (i = 1; i < segment_cnt; i++)
562 int seg_width = sfm_segment_alloc_width (width, i);
563 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
565 write_int (w, 2); /* Variable record. */
566 write_int (w, seg_width); /* Width. */
567 write_int (w, 0); /* No variable label. */
568 write_int (w, 0); /* No missing values. */
569 write_format (w, fmt, seg_width); /* Print format. */
570 write_format (w, fmt, seg_width); /* Write format. */
571 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
573 write_variable_continuation_records (w, seg_width);
577 /* Writes the value labels to system file W.
579 Value labels for long string variables are written separately,
580 by write_long_string_value_labels. */
582 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
586 struct hmap_node hmap_node;
587 const struct val_labs *val_labs;
589 size_t n_indexes, allocated_indexes;
592 size_t n_sets, allocated_sets;
593 struct label_set **sets;
594 struct hmap same_sets;
598 n_sets = allocated_sets = 0;
600 hmap_init (&same_sets);
603 for (i = 0; i < dict_get_var_cnt (d); i++)
605 struct variable *v = dict_get_var (d, i);
607 if (var_has_value_labels (v) && var_get_width (v) <= 8)
609 const struct val_labs *val_labs = var_get_value_labels (v);
610 unsigned int hash = val_labs_hash (val_labs, 0);
611 struct label_set *set;
613 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
616 if (val_labs_equal (set->val_labs, val_labs))
618 if (set->n_indexes >= set->allocated_indexes)
619 set->indexes = x2nrealloc (set->indexes,
620 &set->allocated_indexes,
621 sizeof *set->indexes);
622 set->indexes[set->n_indexes++] = idx;
627 set = xmalloc (sizeof *set);
628 set->val_labs = val_labs;
629 set->indexes = xmalloc (sizeof *set->indexes);
630 set->indexes[0] = idx;
632 set->allocated_indexes = 1;
633 hmap_insert (&same_sets, &set->hmap_node, hash);
635 if (n_sets >= allocated_sets)
636 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
637 sets[n_sets++] = set;
641 idx += sfm_width_to_octs (var_get_width (v));
644 for (i = 0; i < n_sets; i++)
646 const struct label_set *set = sets[i];
647 const struct val_labs *val_labs = set->val_labs;
648 size_t n_labels = val_labs_count (val_labs);
649 int width = val_labs_get_width (val_labs);
650 const struct val_lab **labels;
653 /* Value label record. */
654 write_int (w, 3); /* Record type. */
655 write_int (w, n_labels);
656 labels = val_labs_sorted (val_labs);
657 for (j = 0; j < n_labels; j++)
659 const struct val_lab *vl = labels[j];
660 char *label = recode_string (dict_get_encoding (d), UTF8,
661 val_lab_get_escaped_label (vl), -1);
662 uint8_t len = MIN (strlen (label), 255);
664 write_value (w, val_lab_get_value (vl), width);
665 write_bytes (w, &len, 1);
666 write_bytes (w, label, len);
667 write_zeros (w, REM_RND_UP (len + 1, 8));
672 /* Value label variable record. */
673 write_int (w, 4); /* Record type. */
674 write_int (w, set->n_indexes);
675 for (j = 0; j < set->n_indexes; j++)
676 write_int (w, set->indexes[j] + 1);
679 for (i = 0; i < n_sets; i++)
681 struct label_set *set = sets[i];
687 hmap_destroy (&same_sets);
690 /* Writes record type 6, document record. */
692 write_documents (struct sfm_writer *w, const struct dictionary *d)
694 const struct string_array *docs = dict_get_documents (d);
695 const char *enc = dict_get_encoding (d);
698 write_int (w, 6); /* Record type. */
699 write_int (w, docs->n);
700 for (i = 0; i < docs->n; i++)
702 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
703 size_t s_len = strlen (s);
704 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
706 write_bytes (w, s, write_len);
707 write_spaces (w, DOC_LINE_LENGTH - write_len);
713 put_attrset (struct string *string, const struct attrset *attrs)
715 const struct attribute *attr;
716 struct attrset_iterator i;
718 for (attr = attrset_first (attrs, &i); attr != NULL;
719 attr = attrset_next (attrs, &i))
721 size_t n_values = attribute_get_n_values (attr);
724 ds_put_cstr (string, attribute_get_name (attr));
725 ds_put_byte (string, '(');
726 for (j = 0; j < n_values; j++)
727 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
728 ds_put_byte (string, ')');
733 write_data_file_attributes (struct sfm_writer *w,
734 const struct dictionary *d)
736 struct string s = DS_EMPTY_INITIALIZER;
737 put_attrset (&s, dict_get_attributes (d));
738 write_utf8_record (w, dict_get_encoding (d), &s, 17);
743 add_role_attribute (enum var_role role, struct attrset *attrs)
745 struct attribute *attr;
775 attrset_delete (attrs, "$@Role");
777 attr = attribute_create ("$@Role");
778 attribute_add_value (attr, s);
779 attrset_add (attrs, attr);
783 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
785 struct string s = DS_EMPTY_INITIALIZER;
786 size_t n_vars = dict_get_var_cnt (d);
787 size_t n_attrsets = 0;
790 for (i = 0; i < n_vars; i++)
792 struct variable *v = dict_get_var (d, i);
793 struct attrset attrs;
795 attrset_clone (&attrs, var_get_attributes (v));
797 add_role_attribute (var_get_role (v), &attrs);
799 ds_put_byte (&s, '/');
800 ds_put_format (&s, "%s:", var_get_name (v));
801 put_attrset (&s, &attrs);
802 attrset_destroy (&attrs);
805 write_utf8_record (w, dict_get_encoding (d), &s, 18);
809 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
810 SPSS before release 14, otherwise writes sets supported only by later
813 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
816 const char *encoding = dict_get_encoding (dict);
817 struct string s = DS_EMPTY_INITIALIZER;
821 if (is_encoding_ebcdic_compatible (encoding))
827 n_mrsets = dict_get_n_mrsets (dict);
831 for (i = 0; i < n_mrsets; i++)
833 const struct mrset *mrset = dict_get_mrset (dict, i);
837 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
841 name = recode_string (encoding, "UTF-8", mrset->name, -1);
842 ds_put_format (&s, "%s=", name);
845 if (mrset->type == MRSET_MD)
849 if (mrset->cat_source == MRSET_COUNTEDVALUES)
850 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
852 ds_put_byte (&s, 'D');
854 if (mrset->width == 0)
855 counted = xasprintf ("%.0f", mrset->counted.f);
857 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
859 ds_put_format (&s, "%zu %s", strlen (counted), counted);
863 ds_put_byte (&s, 'C');
864 ds_put_byte (&s, ' ');
866 if (mrset->label && !mrset->label_from_var_label)
868 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
869 ds_put_format (&s, "%zu %s", strlen (label), label);
873 ds_put_cstr (&s, "0 ");
875 for (j = 0; j < mrset->n_vars; j++)
877 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
878 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
879 char *short_name = recode_string (encoding, "UTF-8",
880 lower_name_utf8, -1);
881 ds_put_format (&s, " %s", short_name);
883 free (lower_name_utf8);
885 ds_put_byte (&s, '\n');
888 if (!ds_is_empty (&s))
889 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
893 /* Write the alignment, width and scale values. */
895 write_variable_display_parameters (struct sfm_writer *w,
896 const struct dictionary *dict)
900 write_int (w, 7); /* Record type. */
901 write_int (w, 11); /* Record subtype. */
902 write_int (w, 4); /* Data item (int32) size. */
903 write_int (w, w->segment_cnt * 3); /* Number of data items. */
905 for (i = 0; i < dict_get_var_cnt (dict); ++i)
907 struct variable *v = dict_get_var (dict, i);
908 int width = var_get_width (v);
909 int segment_cnt = sfm_width_to_segments (width);
910 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
911 : var_get_measure (v) == MEASURE_ORDINAL ? 2
913 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
914 : var_get_alignment (v) == ALIGN_RIGHT ? 1
918 for (i = 0; i < segment_cnt; i++)
920 int width_left = width - sfm_segment_effective_offset (width, i);
921 write_int (w, measure);
922 write_int (w, (i == 0 ? var_get_display_width (v)
923 : var_default_display_width (width_left)));
924 write_int (w, alignment);
929 /* Writes the table of lengths for very long string variables. */
931 write_vls_length_table (struct sfm_writer *w,
932 const struct dictionary *dict)
937 ds_init_empty (&map);
938 for (i = 0; i < dict_get_var_cnt (dict); ++i)
940 const struct variable *v = dict_get_var (dict, i);
941 if (sfm_width_to_segments (var_get_width (v)) > 1)
942 ds_put_format (&map, "%s=%05d%c\t",
943 var_get_short_name (v, 0), var_get_width (v), 0);
945 if (!ds_is_empty (&map))
946 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
951 write_long_string_value_labels (struct sfm_writer *w,
952 const struct dictionary *dict)
954 const char *encoding = dict_get_encoding (dict);
955 size_t n_vars = dict_get_var_cnt (dict);
959 /* Figure out the size in advance. */
961 for (i = 0; i < n_vars; i++)
963 struct variable *var = dict_get_var (dict, i);
964 const struct val_labs *val_labs = var_get_value_labels (var);
965 int width = var_get_width (var);
966 const struct val_lab *val_lab;
968 if (val_labs_count (val_labs) == 0 || width < 9)
972 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
973 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
974 val_lab = val_labs_next (val_labs, val_lab))
977 size += recode_string_len (encoding, "UTF-8",
978 val_lab_get_escaped_label (val_lab), -1);
984 write_int (w, 7); /* Record type. */
985 write_int (w, 21); /* Record subtype */
986 write_int (w, 1); /* Data item (byte) size. */
987 write_int (w, size); /* Number of data items. */
989 start = ftello (w->file);
990 for (i = 0; i < n_vars; i++)
992 struct variable *var = dict_get_var (dict, i);
993 const struct val_labs *val_labs = var_get_value_labels (var);
994 int width = var_get_width (var);
995 const struct val_lab *val_lab;
998 if (val_labs_count (val_labs) == 0 || width < 9)
1001 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1002 write_int (w, strlen (var_name));
1003 write_bytes (w, var_name, strlen (var_name));
1006 write_int (w, width);
1007 write_int (w, val_labs_count (val_labs));
1008 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
1009 val_lab = val_labs_next (val_labs, val_lab))
1014 write_int (w, width);
1015 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
1018 label = recode_string (var_get_encoding (var), "UTF-8",
1019 val_lab_get_escaped_label (val_lab), -1);
1020 len = strlen (label);
1022 write_bytes (w, label, len);
1026 assert (ftello (w->file) == start + size);
1030 write_long_string_missing_values (struct sfm_writer *w,
1031 const struct dictionary *dict)
1033 const char *encoding = dict_get_encoding (dict);
1034 size_t n_vars = dict_get_var_cnt (dict);
1038 /* Figure out the size in advance. */
1040 for (i = 0; i < n_vars; i++)
1042 struct variable *var = dict_get_var (dict, i);
1043 const struct missing_values *mv = var_get_missing_values (var);
1044 int width = var_get_width (var);
1046 if (mv_is_empty (mv) || width < 9)
1050 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1052 size += mv_n_values (mv) * (4 + 8);
1057 write_int (w, 7); /* Record type. */
1058 write_int (w, 22); /* Record subtype */
1059 write_int (w, 1); /* Data item (byte) size. */
1060 write_int (w, size); /* Number of data items. */
1062 start = ftello (w->file);
1063 for (i = 0; i < n_vars; i++)
1065 struct variable *var = dict_get_var (dict, i);
1066 const struct missing_values *mv = var_get_missing_values (var);
1067 int width = var_get_width (var);
1068 uint8_t n_missing_values;
1072 if (mv_is_empty (mv) || width < 9)
1075 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1076 write_int (w, strlen (var_name));
1077 write_bytes (w, var_name, strlen (var_name));
1080 n_missing_values = mv_n_values (mv);
1081 write_bytes (w, &n_missing_values, 1);
1083 for (j = 0; j < n_missing_values; j++)
1085 const union value *value = mv_get_value (mv, j);
1088 write_bytes (w, value_str (value, width), 8);
1091 assert (ftello (w->file) == start + size);
1095 write_encoding_record (struct sfm_writer *w,
1096 const struct dictionary *d)
1098 /* IANA says "...character set names may be up to 40 characters taken
1099 from the printable characters of US-ASCII," so character set names
1100 don't need to be recoded to be in UTF-8.
1102 We convert encoding names to uppercase because SPSS writes encoding
1103 names in uppercase. */
1104 char *encoding = xstrdup (dict_get_encoding (d));
1105 str_uppercase (encoding);
1106 write_string_record (w, ss_cstr (encoding), 20);
1110 /* Writes the long variable name table. */
1112 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1117 ds_init_empty (&map);
1118 for (i = 0; i < dict_get_var_cnt (dict); i++)
1120 struct variable *v = dict_get_var (dict, i);
1122 ds_put_byte (&map, '\t');
1123 ds_put_format (&map, "%s=%s",
1124 var_get_short_name (v, 0), var_get_name (v));
1126 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1130 /* Write integer information record. */
1132 write_integer_info_record (struct sfm_writer *w,
1133 const struct dictionary *d)
1135 const char *dict_encoding = dict_get_encoding (d);
1136 int version_component[3];
1140 /* Parse the version string. */
1141 memset (version_component, 0, sizeof version_component);
1142 sscanf (bare_version, "%d.%d.%d",
1143 &version_component[0], &version_component[1], &version_component[2]);
1145 /* Figure out the floating-point format. */
1146 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1147 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1149 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1151 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1156 /* Choose codepage. */
1157 codepage = sys_get_codepage_from_encoding (dict_encoding);
1160 /* The codepage is unknown. Choose a default.
1162 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1164 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1165 many files use this codepage number regardless of their actual
1168 if (is_encoding_ascii_compatible (dict_encoding))
1170 else if (is_encoding_ebcdic_compatible (dict_encoding))
1175 write_int (w, 7); /* Record type. */
1176 write_int (w, 3); /* Record subtype. */
1177 write_int (w, 4); /* Data item (int32) size. */
1178 write_int (w, 8); /* Number of data items. */
1179 write_int (w, version_component[0]);
1180 write_int (w, version_component[1]);
1181 write_int (w, version_component[2]);
1182 write_int (w, -1); /* Machine code. */
1183 write_int (w, float_format);
1184 write_int (w, 1); /* Compression code. */
1185 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1186 write_int (w, codepage);
1189 /* Write floating-point information record. */
1191 write_float_info_record (struct sfm_writer *w)
1193 write_int (w, 7); /* Record type. */
1194 write_int (w, 4); /* Record subtype. */
1195 write_int (w, 8); /* Data item (flt64) size. */
1196 write_int (w, 3); /* Number of data items. */
1197 write_float (w, SYSMIS); /* System-missing value. */
1198 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1199 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1202 /* Writes case C to system file W. */
1204 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1207 struct sfm_writer *w = w_;
1209 if (ferror (w->file))
1211 casewriter_force_error (writer);
1218 if (w->compression == ANY_COMP_NONE)
1219 write_case_uncompressed (w, c);
1221 write_case_compressed (w, c);
1226 /* Destroys system file writer W. */
1228 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1230 struct sfm_writer *w = w_;
1231 if (!close_writer (w))
1232 casewriter_force_error (writer);
1235 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1237 write_error (const struct sfm_writer *writer)
1239 return ferror (writer->file);
1242 /* Closes a system file after we're done with it.
1243 Returns true if successful, false if an I/O error occurred. */
1245 close_writer (struct sfm_writer *w)
1253 if (w->file != NULL)
1256 flush_compressed (w);
1257 if (w->compression == ANY_COMP_ZLIB)
1264 ok = !write_error (w);
1266 /* Seek back to the beginning and update the number of cases.
1267 This is just a courtesy to later readers, so there's no need
1268 to check return values or report errors. */
1269 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1271 write_int (w, w->case_cnt);
1275 if (fclose (w->file) == EOF)
1279 msg (ME, _("An I/O error occurred writing system file `%s'."),
1280 fh_get_file_name (w->fh));
1282 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1288 fh_unlock (w->lock);
1297 /* System file writer casewriter class. */
1298 static const struct casewriter_class sys_file_casewriter_class =
1300 sys_file_casewriter_write,
1301 sys_file_casewriter_destroy,
1305 /* Writes case C to system file W, without compressing it. */
1307 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1311 for (i = 0; i < w->sfm_var_cnt; i++)
1313 struct sfm_var *v = &w->sfm_vars[i];
1315 if (v->var_width == 0)
1316 write_float (w, case_num_idx (c, v->case_index));
1319 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1321 write_spaces (w, v->padding);
1326 /* Writes case C to system file W, with compression. */
1328 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1332 for (i = 0; i < w->sfm_var_cnt; i++)
1334 struct sfm_var *v = &w->sfm_vars[i];
1336 if (v->var_width == 0)
1338 double d = case_num_idx (c, v->case_index);
1340 put_cmp_opcode (w, 255);
1341 else if (d >= 1 - COMPRESSION_BIAS
1342 && d <= 251 - COMPRESSION_BIAS
1344 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1346 put_cmp_number (w, d);
1350 int offset = v->offset;
1353 /* This code properly deals with a width that is not a
1354 multiple of 8, by ensuring that the final partial
1355 oct (8 byte unit) is treated as padded with spaces
1357 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1359 const void *data = case_str_idx (c, v->case_index) + offset;
1360 int chunk_size = MIN (width, 8);
1361 if (!memcmp (data, " ", chunk_size))
1362 put_cmp_opcode (w, 254);
1364 put_cmp_string (w, data, chunk_size);
1367 /* This code deals properly with padding that is not a
1368 multiple of 8 bytes, by discarding the remainder,
1369 which was already effectively padded with spaces in
1370 the previous loop. (Note that v->width + v->padding
1371 is always a multiple of 8.) */
1372 for (padding = v->padding / 8; padding > 0; padding--)
1373 put_cmp_opcode (w, 254);
1379 start_zstream (struct sfm_writer *w)
1383 error = deflateInit (&w->zstream, 1);
1386 msg (ME, _("Failed to initialize ZLIB for compression (%s)."),
1394 finish_zstream (struct sfm_writer *w)
1396 struct zblock *block;
1399 assert (w->zstream.total_in <= ZBLOCK_SIZE);
1401 w->zstream.next_in = NULL;
1402 w->zstream.avail_in = 0;
1407 w->zstream.next_out = buf;
1408 w->zstream.avail_out = sizeof buf;
1409 error = deflate (&w->zstream, Z_FINISH);
1410 write_bytes (w, buf, w->zstream.next_out - buf);
1412 while (error == Z_OK);
1414 if (error != Z_STREAM_END)
1415 msg (ME, _("Failed to complete ZLIB stream compression (%s)."),
1418 if (w->n_blocks >= w->allocated_blocks)
1419 w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks,
1421 block = &w->blocks[w->n_blocks++];
1422 block->uncompressed_size = w->zstream.total_in;
1423 block->compressed_size = w->zstream.total_out;
1427 write_zlib (struct sfm_writer *w, const void *data_, unsigned int n)
1429 const uint8_t *data = data_;
1435 if (w->zstream.total_in >= ZBLOCK_SIZE)
1441 chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in);
1443 w->zstream.next_in = CONST_CAST (uint8_t *, data);
1444 w->zstream.avail_in = chunk;
1450 w->zstream.next_out = buf;
1451 w->zstream.avail_out = sizeof buf;
1452 error = deflate (&w->zstream, Z_NO_FLUSH);
1453 write_bytes (w, buf, w->zstream.next_out - buf);
1456 msg (ME, _("ZLIB stream compression failed (%s)."),
1461 while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0);
1468 write_ztrailer (struct sfm_writer *w)
1470 long long int uncompressed_ofs;
1471 long long int compressed_ofs;
1472 const struct zblock *block;
1474 write_int64 (w, -COMPRESSION_BIAS);
1476 write_int (w, ZBLOCK_SIZE);
1477 write_int (w, w->n_blocks);
1479 uncompressed_ofs = w->zstart;
1480 compressed_ofs = w->zstart + 24;
1481 for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++)
1483 write_int64 (w, uncompressed_ofs);
1484 write_int64 (w, compressed_ofs);
1485 write_int (w, block->uncompressed_size);
1486 write_int (w, block->compressed_size);
1488 uncompressed_ofs += block->uncompressed_size;
1489 compressed_ofs += block->compressed_size;
1492 if (!fseeko (w->file, w->zstart + 8, SEEK_SET))
1494 write_int64 (w, compressed_ofs);
1495 write_int64 (w, 24 + (w->n_blocks * 24));
1498 msg (ME, _("%s: Seek failed (%s)."),
1499 fh_get_file_name (w->fh), strerror (errno));
1502 /* Flushes buffered compressed opcodes and data to W. */
1504 flush_compressed (struct sfm_writer *w)
1508 unsigned int n = 8 * (1 + w->n_elements);
1509 if (w->compression == ANY_COMP_SIMPLE)
1510 write_bytes (w, w->cbuf, n);
1512 write_zlib (w, w->cbuf, n);
1514 w->n_opcodes = w->n_elements = 0;
1515 memset (w->cbuf[0], 0, 8);
1519 /* Appends OPCODE to the buffered set of compression opcodes in
1520 W. Flushes the compression buffer beforehand if necessary. */
1522 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1524 if (w->n_opcodes >= 8)
1525 flush_compressed (w);
1527 w->cbuf[0][w->n_opcodes++] = opcode;
1530 /* Appends NUMBER to the buffered compression data in W. */
1532 put_cmp_number (struct sfm_writer *w, double number)
1534 put_cmp_opcode (w, 253);
1535 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1538 /* Appends SIZE bytes of DATA to the buffered compression data in
1539 W, followed by enough spaces to pad the output data to exactly
1540 8 bytes (thus, SIZE must be no greater than 8). */
1542 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1546 put_cmp_opcode (w, 253);
1548 memset (w->cbuf[w->n_elements], w->space, 8);
1549 memcpy (w->cbuf[w->n_elements], data, size);
1552 /* Writes 32-bit integer X to the output file for writer W. */
1554 write_int (struct sfm_writer *w, int32_t x)
1556 write_bytes (w, &x, sizeof x);
1559 /* Writes 64-bit integer X to the output file for writer W. */
1561 write_int64 (struct sfm_writer *w, int64_t x)
1563 write_bytes (w, &x, sizeof x);
1566 /* Converts NATIVE to the 64-bit format used in output files in
1569 convert_double_to_output_format (double native, uint8_t output[8])
1571 /* If "double" is not a 64-bit type, then convert it to a
1572 64-bit type. Otherwise just copy it. */
1573 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1574 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1576 memcpy (output, &native, sizeof native);
1579 /* Writes floating-point number X to the output file for writer
1582 write_float (struct sfm_writer *w, double x)
1585 convert_double_to_output_format (x, output);
1586 write_bytes (w, output, sizeof output);
1589 /* Writes contents of VALUE with the given WIDTH to W, padding
1590 with zeros to a multiple of 8 bytes.
1591 To avoid a branch, and because we don't actually need to
1592 support it, WIDTH must be no bigger than 8. */
1594 write_value (struct sfm_writer *w, const union value *value, int width)
1596 assert (width <= 8);
1598 write_float (w, value->f);
1601 write_bytes (w, value_str (value, width), width);
1602 write_zeros (w, 8 - width);
1606 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1607 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1608 is padded on the right with spaces. */
1610 write_string (struct sfm_writer *w, const char *string, size_t width)
1612 size_t data_bytes = MIN (strlen (string), width);
1613 size_t pad_bytes = width - data_bytes;
1614 write_bytes (w, string, data_bytes);
1615 while (pad_bytes-- > 0)
1616 putc (w->space, w->file);
1619 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1620 recoded version in a field of the given WIDTH to W. The string is truncated
1621 or padded on the right with spaces to exactly WIDTH bytes. */
1623 write_utf8_string (struct sfm_writer *w, const char *encoding,
1624 const char *string, size_t width)
1626 char *s = recode_string (encoding, "UTF-8", string, -1);
1627 write_string (w, s, width);
1631 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1632 from UTF-8 encoded into ENCODING. */
1634 write_utf8_record (struct sfm_writer *w, const char *encoding,
1635 const struct string *content, int subtype)
1639 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1640 write_string_record (w, s, subtype);
1644 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1647 write_string_record (struct sfm_writer *w,
1648 const struct substring content, int subtype)
1651 write_int (w, subtype);
1653 write_int (w, ss_length (content));
1654 write_bytes (w, ss_data (content), ss_length (content));
1657 /* Writes SIZE bytes of DATA to W's output file. */
1659 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1661 fwrite (data, 1, size, w->file);
1664 /* Writes N zeros to W's output file. */
1666 write_zeros (struct sfm_writer *w, size_t n)
1672 /* Writes N spaces to W's output file. */
1674 write_spaces (struct sfm_writer *w, size_t n)
1677 putc (w->space, w->file);