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/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 enum any_compression compression;
76 casenumber case_cnt; /* Number of cases written so far. */
77 uint8_t space; /* ' ' in the file's character encoding. */
79 /* Simple compression buffering.
81 Compressed data is output as a series of 8-byte elements, with 1 to 9
82 such elements clustered together. The first element in a cluster is 8
83 1-byte opcodes. Some opcodes call for an additional element in the
84 cluster (hence, if there are eight such opcodes, then the cluster
85 contains a full 9 elements).
87 cbuf[] holds a cluster at a time. */
89 int n_opcodes; /* Number of opcodes in cbuf[0] so far. */
90 int n_elements; /* Number of elements in cbuf[] so far. */
92 /* ZLIB compression. */
93 z_stream zstream; /* ZLIB deflater. */
95 struct zblock *blocks;
96 size_t n_blocks, allocated_blocks;
99 struct sfm_var *sfm_vars; /* Variables. */
100 size_t sfm_var_cnt; /* Number of variables. */
101 size_t segment_cnt; /* Number of variables including extra segments
102 for long string variables. */
107 unsigned int uncompressed_size;
108 unsigned int compressed_size;
111 static const struct casewriter_class sys_file_casewriter_class;
113 static void write_header (struct sfm_writer *, const struct dictionary *);
114 static void write_variable (struct sfm_writer *, const struct variable *);
115 static void write_value_labels (struct sfm_writer *,
116 const struct dictionary *);
117 static void write_integer_info_record (struct sfm_writer *,
118 const struct dictionary *);
119 static void write_float_info_record (struct sfm_writer *);
121 static void write_longvar_table (struct sfm_writer *w,
122 const struct dictionary *dict);
124 static void write_encoding_record (struct sfm_writer *w,
125 const struct dictionary *);
127 static void write_vls_length_table (struct sfm_writer *w,
128 const struct dictionary *dict);
130 static void write_long_string_value_labels (struct sfm_writer *,
131 const struct dictionary *);
132 static void write_long_string_missing_values (struct sfm_writer *,
133 const struct dictionary *);
135 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
138 static void write_variable_display_parameters (struct sfm_writer *w,
139 const struct dictionary *dict);
141 static void write_documents (struct sfm_writer *, const struct dictionary *);
143 static void write_data_file_attributes (struct sfm_writer *,
144 const struct dictionary *);
145 static void write_variable_attributes (struct sfm_writer *,
146 const struct dictionary *);
148 static void write_int (struct sfm_writer *, int32_t);
149 static void write_int64 (struct sfm_writer *, int64_t);
150 static inline void convert_double_to_output_format (double, uint8_t[8]);
151 static void write_float (struct sfm_writer *, double);
152 static void write_string (struct sfm_writer *, const char *, size_t);
153 static void write_utf8_string (struct sfm_writer *, const char *encoding,
154 const char *string, size_t width);
155 static void write_utf8_record (struct sfm_writer *, const char *encoding,
156 const struct string *content, int subtype);
157 static void write_string_record (struct sfm_writer *,
158 const struct substring content, int subtype);
159 static void write_bytes (struct sfm_writer *, const void *, size_t);
160 static void write_zeros (struct sfm_writer *, size_t);
161 static void write_spaces (struct sfm_writer *, size_t);
162 static void write_value (struct sfm_writer *, const union value *, int width);
164 static void write_case_uncompressed (struct sfm_writer *,
165 const struct ccase *);
166 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
167 static void flush_compressed (struct sfm_writer *);
168 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
169 static void put_cmp_number (struct sfm_writer *, double);
170 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
172 static bool start_zstream (struct sfm_writer *);
173 static void finish_zstream (struct sfm_writer *);
174 static void write_ztrailer (struct sfm_writer *);
176 static bool write_error (const struct sfm_writer *);
177 static bool close_writer (struct sfm_writer *);
179 /* Returns default options for writing a system file. */
180 struct sfm_write_options
181 sfm_writer_default_options (void)
183 struct sfm_write_options opts;
184 opts.compression = (settings_get_scompression ()
187 opts.create_writeable = true;
192 /* Opens the system file designated by file handle FH for writing
193 cases from dictionary D according to the given OPTS.
195 No reference to D is retained, so it may be modified or
196 destroyed at will after this function returns. D is not
197 modified by this function, except to assign short names. */
199 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
200 struct sfm_write_options opts)
202 struct encoding_info encoding_info;
203 struct sfm_writer *w;
208 if (opts.version != 2 && opts.version != 3)
210 msg (ME, _("Unknown system file version %d. Treating as version %d."),
215 /* Create and initialize writer. */
216 w = xzalloc (sizeof *w);
222 /* Use the requested compression, except that no EBCDIC-based ZLIB compressed
223 files have been observed, so drop back to simple compression for those
225 w->compression = opts.compression;
226 if (w->compression == ANY_COMP_ZLIB
227 && is_encoding_ebcdic_compatible (dict_get_encoding (d)))
228 w->compression = ANY_COMP_SIMPLE;
232 w->n_opcodes = w->n_elements = 0;
233 memset (w->cbuf[0], 0, 8);
235 /* Figure out how to map in-memory case data to on-disk case
236 data. Also count the number of segments. Very long strings
237 occupy multiple segments, otherwise each variable only takes
239 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
242 /* Open file handle as an exclusive writer. */
243 /* TRANSLATORS: this fragment will be interpolated into
244 messages in fh_lock() that identify types of files. */
245 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
249 /* Create the file on disk. */
251 if (opts.create_writeable)
253 w->rf = replace_file_start (fh, "wb", mode, &w->file);
256 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
257 fh_get_file_name (fh), strerror (errno));
261 get_encoding_info (&encoding_info, dict_get_encoding (d));
262 w->space = encoding_info.space[0];
264 /* Write the file header. */
267 /* Write basic variable info. */
268 short_names_assign (d);
269 for (i = 0; i < dict_get_var_cnt (d); i++)
270 write_variable (w, dict_get_var (d, i));
272 write_value_labels (w, d);
274 if (dict_get_document_line_cnt (d) > 0)
275 write_documents (w, d);
277 write_integer_info_record (w, d);
278 write_float_info_record (w);
280 write_mrsets (w, d, true);
282 write_variable_display_parameters (w, d);
284 if (opts.version >= 3)
285 write_longvar_table (w, d);
287 write_vls_length_table (w, d);
289 write_long_string_value_labels (w, d);
290 write_long_string_missing_values (w, d);
292 if (opts.version >= 3)
294 if (attrset_count (dict_get_attributes (d)))
295 write_data_file_attributes (w, d);
296 write_variable_attributes (w, d);
299 write_mrsets (w, d, false);
301 write_encoding_record (w, d);
303 /* Write end-of-headers record. */
307 if (w->compression == ANY_COMP_ZLIB)
309 w->zstream.zalloc = Z_NULL;
310 w->zstream.zfree = Z_NULL;
311 w->zstream.opaque = Z_NULL;
312 w->zstart = ftello (w->file);
314 write_int64 (w, w->zstart);
324 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
331 /* Returns value of X truncated to two least-significant digits. */
342 /* Calculates the offset of data for TARGET_VAR from the
343 beginning of each case's data for dictionary D. The return
344 value is in "octs" (8-byte units). */
346 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
352 for (i = 0; i < dict_get_var_cnt (d); i++)
354 struct variable *var = dict_get_var (d, i);
355 if (var == target_var)
357 oct_idx += sfm_width_to_octs (var_get_width (var));
362 /* Write the sysfile_header header to system file W. */
364 write_header (struct sfm_writer *w, const struct dictionary *d)
366 const char *dict_encoding = dict_get_encoding (d);
368 const char *file_label;
369 struct variable *weight;
373 /* Record-type code. */
374 if (is_encoding_ebcdic_compatible (dict_encoding))
375 write_string (w, EBCDIC_MAGIC, 4);
376 else if (w->compression == ANY_COMP_ZLIB)
377 write_string (w, ASCII_ZMAGIC, 4);
379 write_string (w, ASCII_MAGIC, 4);
381 /* Product identification. */
382 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
383 version, host_system);
384 write_utf8_string (w, dict_encoding, prod_name, 60);
389 /* Number of `union value's per case. */
390 write_int (w, calc_oct_idx (d, NULL));
393 write_int (w, (w->compression == ANY_COMP_NONE ? 0
394 : w->compression == ANY_COMP_SIMPLE ? 1
397 /* Weight variable. */
398 weight = dict_get_weight (d);
399 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
401 /* Number of cases. We don't know this in advance, so we write
402 -1 to indicate an unknown number of cases. Later we can
403 come back and overwrite it with the true value. */
406 /* Compression bias. */
407 write_float (w, COMPRESSION_BIAS);
409 /* Creation date and time. */
410 char *creation_date, *creation_time;
411 if (time (&t) == (time_t) -1)
413 creation_date = xstrdup ("01 Jan 70");
414 creation_time = xstrdup ( "00:00:00");
418 static const char *const month_name[12] =
420 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
421 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
423 struct tm *tmp = localtime (&t);
424 int day = rerange (tmp->tm_mday);
425 int mon = rerange (tmp->tm_mon + 1);
426 int year = rerange (tmp->tm_year);
427 int hour = rerange (tmp->tm_hour + 1);
428 int min = rerange (tmp->tm_min + 1);
429 int sec = rerange (tmp->tm_sec + 1);
431 creation_date = xasprintf ("%02d %s %02d",
432 day, month_name[mon - 1], year);
433 creation_time = xasprintf ("%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
435 write_utf8_string (w, dict_encoding, creation_date, 9);
436 write_utf8_string (w, dict_encoding, creation_time, 8);
437 free (creation_time);
438 free (creation_date);
441 file_label = dict_get_label (d);
442 if (file_label == NULL)
444 write_utf8_string (w, dict_encoding, file_label, 64);
450 /* Write format spec FMT to W, after adjusting it to be
451 compatible with the given WIDTH. */
453 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
455 assert (fmt_check_output (&fmt));
456 assert (sfm_width_to_segments (width) == 1);
459 fmt_resize (&fmt, width);
460 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
463 /* Write a string continuation variable record for each 8-byte
464 section beyond the initial 8 bytes, for a variable of the
467 write_variable_continuation_records (struct sfm_writer *w, int width)
471 assert (sfm_width_to_segments (width) == 1);
472 for (position = 8; position < width; position += 8)
474 write_int (w, 2); /* Record type. */
475 write_int (w, -1); /* Width. */
476 write_int (w, 0); /* No variable label. */
477 write_int (w, 0); /* No missing values. */
478 write_int (w, 0); /* Print format. */
479 write_int (w, 0); /* Write format. */
480 write_zeros (w, 8); /* Name. */
484 /* Write the variable record(s) for variable V to system file
487 write_variable (struct sfm_writer *w, const struct variable *v)
489 int width = var_get_width (v);
490 int segment_cnt = sfm_width_to_segments (width);
491 int seg0_width = sfm_segment_alloc_width (width, 0);
492 const char *encoding = var_get_encoding (v);
499 write_int (w, seg0_width);
501 /* Variable has a variable label? */
502 write_int (w, var_has_label (v));
504 /* Number of missing values. If there is a range, then the
505 range counts as 2 missing values and causes the number to be
508 Missing values for long string variables are written in a separate
510 enum { MAX_SHORT_STRING = 8 };
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 (mrset->counted.s, mrset->width);
858 ds_put_format (&s, "%zu %s", strlen (counted), counted);
862 ds_put_byte (&s, 'C');
863 ds_put_byte (&s, ' ');
865 if (mrset->label && !mrset->label_from_var_label)
867 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
868 ds_put_format (&s, "%zu %s", strlen (label), label);
872 ds_put_cstr (&s, "0 ");
874 for (j = 0; j < mrset->n_vars; j++)
876 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
877 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
878 char *short_name = recode_string (encoding, "UTF-8",
879 lower_name_utf8, -1);
880 ds_put_format (&s, " %s", short_name);
882 free (lower_name_utf8);
884 ds_put_byte (&s, '\n');
887 if (!ds_is_empty (&s))
888 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
892 /* Write the alignment, width and scale values. */
894 write_variable_display_parameters (struct sfm_writer *w,
895 const struct dictionary *dict)
899 write_int (w, 7); /* Record type. */
900 write_int (w, 11); /* Record subtype. */
901 write_int (w, 4); /* Data item (int32) size. */
902 write_int (w, w->segment_cnt * 3); /* Number of data items. */
904 for (i = 0; i < dict_get_var_cnt (dict); ++i)
906 struct variable *v = dict_get_var (dict, i);
907 int width = var_get_width (v);
908 int segment_cnt = sfm_width_to_segments (width);
909 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
910 : var_get_measure (v) == MEASURE_ORDINAL ? 2
912 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
913 : var_get_alignment (v) == ALIGN_RIGHT ? 1
917 for (i = 0; i < segment_cnt; i++)
919 int width_left = width - sfm_segment_effective_offset (width, i);
920 write_int (w, measure);
921 write_int (w, (i == 0 ? var_get_display_width (v)
922 : var_default_display_width (width_left)));
923 write_int (w, alignment);
928 /* Writes the table of lengths for very long string variables. */
930 write_vls_length_table (struct sfm_writer *w,
931 const struct dictionary *dict)
936 ds_init_empty (&map);
937 for (i = 0; i < dict_get_var_cnt (dict); ++i)
939 const struct variable *v = dict_get_var (dict, i);
940 if (sfm_width_to_segments (var_get_width (v)) > 1)
941 ds_put_format (&map, "%s=%05d%c\t",
942 var_get_short_name (v, 0), var_get_width (v), 0);
944 if (!ds_is_empty (&map))
945 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
950 write_long_string_value_labels (struct sfm_writer *w,
951 const struct dictionary *dict)
953 const char *encoding = dict_get_encoding (dict);
954 size_t n_vars = dict_get_var_cnt (dict);
957 /* Figure out the size in advance. */
959 for (i = 0; i < n_vars; i++)
961 struct variable *var = dict_get_var (dict, i);
962 const struct val_labs *val_labs = var_get_value_labels (var);
963 int width = var_get_width (var);
964 const struct val_lab *val_lab;
966 if (val_labs_count (val_labs) == 0 || width < 9)
970 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
971 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
972 val_lab = val_labs_next (val_labs, val_lab))
975 size += recode_string_len (encoding, "UTF-8",
976 val_lab_get_escaped_label (val_lab), -1);
982 write_int (w, 7); /* Record type. */
983 write_int (w, 21); /* Record subtype */
984 write_int (w, 1); /* Data item (byte) size. */
985 write_int (w, size); /* Number of data items. */
987 for (i = 0; i < n_vars; i++)
989 struct variable *var = dict_get_var (dict, i);
990 const struct val_labs *val_labs = var_get_value_labels (var);
991 int width = var_get_width (var);
992 const struct val_lab *val_lab;
995 if (val_labs_count (val_labs) == 0 || width < 9)
998 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
999 write_int (w, strlen (var_name));
1000 write_bytes (w, var_name, strlen (var_name));
1003 write_int (w, width);
1004 write_int (w, val_labs_count (val_labs));
1005 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
1006 val_lab = val_labs_next (val_labs, val_lab))
1011 write_int (w, width);
1012 write_bytes (w, val_lab_get_value (val_lab)->s, width);
1014 label = recode_string (var_get_encoding (var), "UTF-8",
1015 val_lab_get_escaped_label (val_lab), -1);
1016 len = strlen (label);
1018 write_bytes (w, label, len);
1025 write_long_string_missing_values (struct sfm_writer *w,
1026 const struct dictionary *dict)
1028 const char *encoding = dict_get_encoding (dict);
1029 size_t n_vars = dict_get_var_cnt (dict);
1032 /* Figure out the size in advance. */
1034 for (i = 0; i < n_vars; i++)
1036 struct variable *var = dict_get_var (dict, i);
1037 const struct missing_values *mv = var_get_missing_values (var);
1038 int width = var_get_width (var);
1040 if (mv_is_empty (mv) || width < 9)
1044 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1046 size += mv_n_values (mv) * (4 + 8);
1051 write_int (w, 7); /* Record type. */
1052 write_int (w, 22); /* Record subtype */
1053 write_int (w, 1); /* Data item (byte) size. */
1054 write_int (w, size); /* Number of data items. */
1056 for (i = 0; i < n_vars; i++)
1058 struct variable *var = dict_get_var (dict, i);
1059 const struct missing_values *mv = var_get_missing_values (var);
1060 int width = var_get_width (var);
1061 uint8_t n_missing_values;
1065 if (mv_is_empty (mv) || width < 9)
1068 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1069 write_int (w, strlen (var_name));
1070 write_bytes (w, var_name, strlen (var_name));
1073 n_missing_values = mv_n_values (mv);
1074 write_bytes (w, &n_missing_values, 1);
1076 for (j = 0; j < n_missing_values; j++)
1078 const union value *value = mv_get_value (mv, j);
1081 write_bytes (w, value->s, 8);
1087 write_encoding_record (struct sfm_writer *w,
1088 const struct dictionary *d)
1090 /* IANA says "...character set names may be up to 40 characters taken
1091 from the printable characters of US-ASCII," so character set names
1092 don't need to be recoded to be in UTF-8.
1094 We convert encoding names to uppercase because SPSS writes encoding
1095 names in uppercase. */
1096 char *encoding = xstrdup (dict_get_encoding (d));
1097 str_uppercase (encoding);
1098 write_string_record (w, ss_cstr (encoding), 20);
1102 /* Writes the long variable name table. */
1104 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1109 ds_init_empty (&map);
1110 for (i = 0; i < dict_get_var_cnt (dict); i++)
1112 struct variable *v = dict_get_var (dict, i);
1114 ds_put_byte (&map, '\t');
1115 ds_put_format (&map, "%s=%s",
1116 var_get_short_name (v, 0), var_get_name (v));
1118 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1122 /* Write integer information record. */
1124 write_integer_info_record (struct sfm_writer *w,
1125 const struct dictionary *d)
1127 const char *dict_encoding = dict_get_encoding (d);
1128 int version_component[3];
1132 /* Parse the version string. */
1133 memset (version_component, 0, sizeof version_component);
1134 sscanf (bare_version, "%d.%d.%d",
1135 &version_component[0], &version_component[1], &version_component[2]);
1137 /* Figure out the floating-point format. */
1138 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1139 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1141 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1143 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1148 /* Choose codepage. */
1149 codepage = sys_get_codepage_from_encoding (dict_encoding);
1152 /* The codepage is unknown. Choose a default.
1154 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1156 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1157 many files use this codepage number regardless of their actual
1160 if (is_encoding_ascii_compatible (dict_encoding))
1162 else if (is_encoding_ebcdic_compatible (dict_encoding))
1167 write_int (w, 7); /* Record type. */
1168 write_int (w, 3); /* Record subtype. */
1169 write_int (w, 4); /* Data item (int32) size. */
1170 write_int (w, 8); /* Number of data items. */
1171 write_int (w, version_component[0]);
1172 write_int (w, version_component[1]);
1173 write_int (w, version_component[2]);
1174 write_int (w, -1); /* Machine code. */
1175 write_int (w, float_format);
1176 write_int (w, 1); /* Compression code. */
1177 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1178 write_int (w, codepage);
1181 /* Write floating-point information record. */
1183 write_float_info_record (struct sfm_writer *w)
1185 write_int (w, 7); /* Record type. */
1186 write_int (w, 4); /* Record subtype. */
1187 write_int (w, 8); /* Data item (flt64) size. */
1188 write_int (w, 3); /* Number of data items. */
1189 write_float (w, SYSMIS); /* System-missing value. */
1190 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1191 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1194 /* Writes case C to system file W. */
1196 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1199 struct sfm_writer *w = w_;
1201 if (ferror (w->file))
1203 casewriter_force_error (writer);
1210 if (w->compression == ANY_COMP_NONE)
1211 write_case_uncompressed (w, c);
1213 write_case_compressed (w, c);
1218 /* Destroys system file writer W. */
1220 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1222 struct sfm_writer *w = w_;
1223 if (!close_writer (w))
1224 casewriter_force_error (writer);
1227 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1229 write_error (const struct sfm_writer *writer)
1231 return ferror (writer->file);
1234 /* Closes a system file after we're done with it.
1235 Returns true if successful, false if an I/O error occurred. */
1237 close_writer (struct sfm_writer *w)
1245 if (w->file != NULL)
1248 flush_compressed (w);
1249 if (w->compression == ANY_COMP_ZLIB)
1256 ok = !write_error (w);
1258 /* Seek back to the beginning and update the number of cases.
1259 This is just a courtesy to later readers, so there's no need
1260 to check return values or report errors. */
1261 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1263 write_int (w, w->case_cnt);
1267 if (fclose (w->file) == EOF)
1271 msg (ME, _("An I/O error occurred writing system file `%s'."),
1272 fh_get_file_name (w->fh));
1274 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1280 fh_unlock (w->lock);
1289 /* System file writer casewriter class. */
1290 static const struct casewriter_class sys_file_casewriter_class =
1292 sys_file_casewriter_write,
1293 sys_file_casewriter_destroy,
1297 /* Writes case C to system file W, without compressing it. */
1299 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1303 for (i = 0; i < w->sfm_var_cnt; i++)
1305 struct sfm_var *v = &w->sfm_vars[i];
1307 if (v->var_width == 0)
1308 write_float (w, case_num_idx (c, v->case_index));
1311 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1313 write_spaces (w, v->padding);
1318 /* Writes case C to system file W, with compression. */
1320 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1324 for (i = 0; i < w->sfm_var_cnt; i++)
1326 struct sfm_var *v = &w->sfm_vars[i];
1328 if (v->var_width == 0)
1330 double d = case_num_idx (c, v->case_index);
1332 put_cmp_opcode (w, 255);
1333 else if (d >= 1 - COMPRESSION_BIAS
1334 && d <= 251 - COMPRESSION_BIAS
1336 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1338 put_cmp_number (w, d);
1342 int offset = v->offset;
1345 /* This code properly deals with a width that is not a
1346 multiple of 8, by ensuring that the final partial
1347 oct (8 byte unit) is treated as padded with spaces
1349 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1351 const void *data = case_str_idx (c, v->case_index) + offset;
1352 int chunk_size = MIN (width, 8);
1353 if (!memcmp (data, " ", chunk_size))
1354 put_cmp_opcode (w, 254);
1356 put_cmp_string (w, data, chunk_size);
1359 /* This code deals properly with padding that is not a
1360 multiple of 8 bytes, by discarding the remainder,
1361 which was already effectively padded with spaces in
1362 the previous loop. (Note that v->width + v->padding
1363 is always a multiple of 8.) */
1364 for (padding = v->padding / 8; padding > 0; padding--)
1365 put_cmp_opcode (w, 254);
1371 start_zstream (struct sfm_writer *w)
1375 error = deflateInit (&w->zstream, 1);
1378 msg (ME, _("Failed to initialize ZLIB for compression (%s)."),
1386 finish_zstream (struct sfm_writer *w)
1388 struct zblock *block;
1391 assert (w->zstream.total_in <= ZBLOCK_SIZE);
1393 w->zstream.next_in = NULL;
1394 w->zstream.avail_in = 0;
1399 w->zstream.next_out = buf;
1400 w->zstream.avail_out = sizeof buf;
1401 error = deflate (&w->zstream, Z_FINISH);
1402 write_bytes (w, buf, w->zstream.next_out - buf);
1404 while (error == Z_OK);
1406 if (error != Z_STREAM_END)
1407 msg (ME, _("Failed to complete ZLIB stream compression (%s)."),
1410 if (w->n_blocks >= w->allocated_blocks)
1411 w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks,
1413 block = &w->blocks[w->n_blocks++];
1414 block->uncompressed_size = w->zstream.total_in;
1415 block->compressed_size = w->zstream.total_out;
1416 deflateEnd (&w->zstream);
1420 write_zlib (struct sfm_writer *w, const void *data_, unsigned int n)
1422 const uint8_t *data = data_;
1428 if (w->zstream.total_in >= ZBLOCK_SIZE)
1434 chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in);
1436 w->zstream.next_in = CONST_CAST (uint8_t *, data);
1437 w->zstream.avail_in = chunk;
1443 w->zstream.next_out = buf;
1444 w->zstream.avail_out = sizeof buf;
1445 error = deflate (&w->zstream, Z_NO_FLUSH);
1446 write_bytes (w, buf, w->zstream.next_out - buf);
1449 msg (ME, _("ZLIB stream compression failed (%s)."),
1454 while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0);
1461 write_ztrailer (struct sfm_writer *w)
1463 long long int uncompressed_ofs;
1464 long long int compressed_ofs;
1465 const struct zblock *block;
1467 write_int64 (w, -COMPRESSION_BIAS);
1469 write_int (w, ZBLOCK_SIZE);
1470 write_int (w, w->n_blocks);
1472 uncompressed_ofs = w->zstart;
1473 compressed_ofs = w->zstart + 24;
1474 for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++)
1476 write_int64 (w, uncompressed_ofs);
1477 write_int64 (w, compressed_ofs);
1478 write_int (w, block->uncompressed_size);
1479 write_int (w, block->compressed_size);
1481 uncompressed_ofs += block->uncompressed_size;
1482 compressed_ofs += block->compressed_size;
1485 if (!fseeko (w->file, w->zstart + 8, SEEK_SET))
1487 write_int64 (w, compressed_ofs);
1488 write_int64 (w, 24 + (w->n_blocks * 24));
1491 msg (ME, _("%s: Seek failed (%s)."),
1492 fh_get_file_name (w->fh), strerror (errno));
1495 /* Flushes buffered compressed opcodes and data to W. */
1497 flush_compressed (struct sfm_writer *w)
1501 unsigned int n = 8 * (1 + w->n_elements);
1502 if (w->compression == ANY_COMP_SIMPLE)
1503 write_bytes (w, w->cbuf, n);
1505 write_zlib (w, w->cbuf, n);
1507 w->n_opcodes = w->n_elements = 0;
1508 memset (w->cbuf[0], 0, 8);
1512 /* Appends OPCODE to the buffered set of compression opcodes in
1513 W. Flushes the compression buffer beforehand if necessary. */
1515 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1517 if (w->n_opcodes >= 8)
1518 flush_compressed (w);
1520 w->cbuf[0][w->n_opcodes++] = opcode;
1523 /* Appends NUMBER to the buffered compression data in W. */
1525 put_cmp_number (struct sfm_writer *w, double number)
1527 put_cmp_opcode (w, 253);
1528 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1531 /* Appends SIZE bytes of DATA to the buffered compression data in
1532 W, followed by enough spaces to pad the output data to exactly
1533 8 bytes (thus, SIZE must be no greater than 8). */
1535 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1539 put_cmp_opcode (w, 253);
1541 memset (w->cbuf[w->n_elements], w->space, 8);
1542 memcpy (w->cbuf[w->n_elements], data, size);
1545 /* Writes 32-bit integer X to the output file for writer W. */
1547 write_int (struct sfm_writer *w, int32_t x)
1549 write_bytes (w, &x, sizeof x);
1552 /* Writes 64-bit integer X to the output file for writer W. */
1554 write_int64 (struct sfm_writer *w, int64_t x)
1556 write_bytes (w, &x, sizeof x);
1559 /* Converts NATIVE to the 64-bit format used in output files in
1562 convert_double_to_output_format (double native, uint8_t output[8])
1564 /* If "double" is not a 64-bit type, then convert it to a
1565 64-bit type. Otherwise just copy it. */
1566 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1567 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1569 memcpy (output, &native, sizeof native);
1572 /* Writes floating-point number X to the output file for writer
1575 write_float (struct sfm_writer *w, double x)
1578 convert_double_to_output_format (x, output);
1579 write_bytes (w, output, sizeof output);
1582 /* Writes contents of VALUE with the given WIDTH to W, padding
1583 with zeros to a multiple of 8 bytes.
1584 To avoid a branch, and because we don't actually need to
1585 support it, WIDTH must be no bigger than 8. */
1587 write_value (struct sfm_writer *w, const union value *value, int width)
1589 assert (width <= 8);
1591 write_float (w, value->f);
1594 write_bytes (w, value->s, width);
1595 write_zeros (w, 8 - width);
1599 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1600 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1601 is padded on the right with spaces. */
1603 write_string (struct sfm_writer *w, const char *string, size_t width)
1605 size_t data_bytes = MIN (strlen (string), width);
1606 size_t pad_bytes = width - data_bytes;
1607 write_bytes (w, string, data_bytes);
1608 while (pad_bytes-- > 0)
1609 putc (w->space, w->file);
1612 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1613 recoded version in a field of the given WIDTH to W. The string is truncated
1614 or padded on the right with spaces to exactly WIDTH bytes. */
1616 write_utf8_string (struct sfm_writer *w, const char *encoding,
1617 const char *string, size_t width)
1619 char *s = recode_string (encoding, "UTF-8", string, -1);
1620 write_string (w, s, width);
1624 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1625 from UTF-8 encoded into ENCODING. */
1627 write_utf8_record (struct sfm_writer *w, const char *encoding,
1628 const struct string *content, int subtype)
1632 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1633 write_string_record (w, s, subtype);
1637 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1640 write_string_record (struct sfm_writer *w,
1641 const struct substring content, int subtype)
1644 write_int (w, subtype);
1646 write_int (w, ss_length (content));
1647 write_bytes (w, ss_data (content), ss_length (content));
1650 /* Writes SIZE bytes of DATA to W's output file. */
1652 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1654 fwrite (data, 1, size, w->file);
1657 /* Writes N zeros to W's output file. */
1659 write_zeros (struct sfm_writer *w, size_t n)
1665 /* Writes N spaces to W's output file. */
1667 write_spaces (struct sfm_writer *w, size_t n)
1670 putc (w->space, w->file);