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/fbuf.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 struct fbuf *fbuf; /* 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)
256 w->rf = replace_file_start_fd (fh, true, mode, &fd);
259 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
260 fh_get_file_name (fh), strerror (errno));
263 w->fbuf = fbuf_open_fd (fd);
265 get_encoding_info (&encoding_info, dict_get_encoding (d));
266 w->space = encoding_info.space[0];
268 /* Write the file header. */
271 /* Write basic variable info. */
272 short_names_assign (d);
273 for (i = 0; i < dict_get_var_cnt (d); i++)
274 write_variable (w, dict_get_var (d, i));
276 write_value_labels (w, d);
278 if (dict_get_document_line_cnt (d) > 0)
279 write_documents (w, d);
281 write_integer_info_record (w, d);
282 write_float_info_record (w);
284 write_mrsets (w, d, true);
286 write_variable_display_parameters (w, d);
288 if (opts.version >= 3)
289 write_longvar_table (w, d);
291 write_vls_length_table (w, d);
293 write_long_string_value_labels (w, d);
294 write_long_string_missing_values (w, d);
296 if (opts.version >= 3)
298 if (attrset_count (dict_get_attributes (d)))
299 write_data_file_attributes (w, d);
300 write_variable_attributes (w, d);
303 write_mrsets (w, d, false);
305 write_encoding_record (w, d);
307 /* Write end-of-headers record. */
311 if (w->compression == ANY_COMP_ZLIB)
313 w->zstream.zalloc = Z_NULL;
314 w->zstream.zfree = Z_NULL;
315 w->zstream.opaque = Z_NULL;
316 w->zstart = fbuf_tell (w->fbuf);
318 write_int64 (w, w->zstart);
328 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
335 /* Returns value of X truncated to two least-significant digits. */
346 /* Calculates the offset of data for TARGET_VAR from the
347 beginning of each case's data for dictionary D. The return
348 value is in "octs" (8-byte units). */
350 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
356 for (i = 0; i < dict_get_var_cnt (d); i++)
358 struct variable *var = dict_get_var (d, i);
359 if (var == target_var)
361 oct_idx += sfm_width_to_octs (var_get_width (var));
366 /* Write the sysfile_header header to system file W. */
368 write_header (struct sfm_writer *w, const struct dictionary *d)
370 const char *dict_encoding = dict_get_encoding (d);
372 const char *file_label;
373 struct variable *weight;
377 /* Record-type code. */
378 if (is_encoding_ebcdic_compatible (dict_encoding))
379 write_string (w, EBCDIC_MAGIC, 4);
380 else if (w->compression == ANY_COMP_ZLIB)
381 write_string (w, ASCII_ZMAGIC, 4);
383 write_string (w, ASCII_MAGIC, 4);
385 /* Product identification. */
386 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
387 version, host_system);
388 write_utf8_string (w, dict_encoding, prod_name, 60);
393 /* Number of `union value's per case. */
394 write_int (w, calc_oct_idx (d, NULL));
397 write_int (w, (w->compression == ANY_COMP_NONE ? 0
398 : w->compression == ANY_COMP_SIMPLE ? 1
401 /* Weight variable. */
402 weight = dict_get_weight (d);
403 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
405 /* Number of cases. We don't know this in advance, so we write
406 -1 to indicate an unknown number of cases. Later we can
407 come back and overwrite it with the true value. */
410 /* Compression bias. */
411 write_float (w, COMPRESSION_BIAS);
413 /* Creation date and time. */
414 char *creation_date, *creation_time;
415 if (time (&t) == (time_t) -1)
417 creation_date = xstrdup ("01 Jan 70");
418 creation_time = xstrdup ( "00:00:00");
422 static const char *const month_name[12] =
424 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
425 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
427 struct tm *tmp = localtime (&t);
428 int day = rerange (tmp->tm_mday);
429 int mon = rerange (tmp->tm_mon + 1);
430 int year = rerange (tmp->tm_year);
431 int hour = rerange (tmp->tm_hour + 1);
432 int min = rerange (tmp->tm_min + 1);
433 int sec = rerange (tmp->tm_sec + 1);
435 creation_date = xasprintf ("%02d %s %02d",
436 day, month_name[mon - 1], year);
437 creation_time = xasprintf ("%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
439 write_utf8_string (w, dict_encoding, creation_date, 9);
440 write_utf8_string (w, dict_encoding, creation_time, 8);
441 free (creation_time);
442 free (creation_date);
445 file_label = dict_get_label (d);
446 if (file_label == NULL)
448 write_utf8_string (w, dict_encoding, file_label, 64);
454 /* Write format spec FMT to W, after adjusting it to be
455 compatible with the given WIDTH. */
457 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
459 assert (fmt_check_output (&fmt));
460 assert (sfm_width_to_segments (width) == 1);
463 fmt_resize (&fmt, width);
464 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
467 /* Write a string continuation variable record for each 8-byte
468 section beyond the initial 8 bytes, for a variable of the
471 write_variable_continuation_records (struct sfm_writer *w, int width)
475 assert (sfm_width_to_segments (width) == 1);
476 for (position = 8; position < width; position += 8)
478 write_int (w, 2); /* Record type. */
479 write_int (w, -1); /* Width. */
480 write_int (w, 0); /* No variable label. */
481 write_int (w, 0); /* No missing values. */
482 write_int (w, 0); /* Print format. */
483 write_int (w, 0); /* Write format. */
484 write_zeros (w, 8); /* Name. */
488 /* Write the variable record(s) for variable V to system file
491 write_variable (struct sfm_writer *w, const struct variable *v)
493 int width = var_get_width (v);
494 int segment_cnt = sfm_width_to_segments (width);
495 int seg0_width = sfm_segment_alloc_width (width, 0);
496 const char *encoding = var_get_encoding (v);
503 write_int (w, seg0_width);
505 /* Variable has a variable label? */
506 write_int (w, var_has_label (v));
508 /* Number of missing values. If there is a range, then the
509 range counts as 2 missing values and causes the number to be
512 Missing values for long string variables are written in a separate
514 if (width <= MAX_SHORT_STRING)
516 const struct missing_values *mv = var_get_missing_values (v);
517 if (mv_has_range (mv))
518 write_int (w, -2 - mv_n_values (mv));
520 write_int (w, mv_n_values (mv));
525 /* Print and write formats. */
526 write_format (w, *var_get_print_format (v), seg0_width);
527 write_format (w, *var_get_write_format (v), seg0_width);
530 The full name is in a translation table written
532 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
535 if (var_has_label (v))
537 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
538 size_t label_len = MIN (strlen (label), 255);
539 size_t padded_len = ROUND_UP (label_len, 4);
540 write_int (w, label_len);
541 write_string (w, label, padded_len);
545 /* Write the missing values, if any, range first. */
546 if (width <= MAX_SHORT_STRING)
548 const struct missing_values *mv = var_get_missing_values (v);
549 if (mv_has_range (mv))
552 mv_get_range (mv, &x, &y);
556 for (i = 0; i < mv_n_values (mv); i++)
557 write_value (w, mv_get_value (mv, i), width);
560 write_variable_continuation_records (w, seg0_width);
562 /* Write additional segments for very long string variables. */
563 for (i = 1; i < segment_cnt; i++)
565 int seg_width = sfm_segment_alloc_width (width, i);
566 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
568 write_int (w, 2); /* Variable record. */
569 write_int (w, seg_width); /* Width. */
570 write_int (w, 0); /* No variable label. */
571 write_int (w, 0); /* No missing values. */
572 write_format (w, fmt, seg_width); /* Print format. */
573 write_format (w, fmt, seg_width); /* Write format. */
574 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
576 write_variable_continuation_records (w, seg_width);
580 /* Writes the value labels to system file W.
582 Value labels for long string variables are written separately,
583 by write_long_string_value_labels. */
585 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
589 struct hmap_node hmap_node;
590 const struct val_labs *val_labs;
592 size_t n_indexes, allocated_indexes;
595 size_t n_sets, allocated_sets;
596 struct label_set **sets;
597 struct hmap same_sets;
601 n_sets = allocated_sets = 0;
603 hmap_init (&same_sets);
606 for (i = 0; i < dict_get_var_cnt (d); i++)
608 struct variable *v = dict_get_var (d, i);
610 if (var_has_value_labels (v) && var_get_width (v) <= 8)
612 const struct val_labs *val_labs = var_get_value_labels (v);
613 unsigned int hash = val_labs_hash (val_labs, 0);
614 struct label_set *set;
616 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
619 if (val_labs_equal (set->val_labs, val_labs))
621 if (set->n_indexes >= set->allocated_indexes)
622 set->indexes = x2nrealloc (set->indexes,
623 &set->allocated_indexes,
624 sizeof *set->indexes);
625 set->indexes[set->n_indexes++] = idx;
630 set = xmalloc (sizeof *set);
631 set->val_labs = val_labs;
632 set->indexes = xmalloc (sizeof *set->indexes);
633 set->indexes[0] = idx;
635 set->allocated_indexes = 1;
636 hmap_insert (&same_sets, &set->hmap_node, hash);
638 if (n_sets >= allocated_sets)
639 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
640 sets[n_sets++] = set;
644 idx += sfm_width_to_octs (var_get_width (v));
647 for (i = 0; i < n_sets; i++)
649 const struct label_set *set = sets[i];
650 const struct val_labs *val_labs = set->val_labs;
651 size_t n_labels = val_labs_count (val_labs);
652 int width = val_labs_get_width (val_labs);
653 const struct val_lab **labels;
656 /* Value label record. */
657 write_int (w, 3); /* Record type. */
658 write_int (w, n_labels);
659 labels = val_labs_sorted (val_labs);
660 for (j = 0; j < n_labels; j++)
662 const struct val_lab *vl = labels[j];
663 char *label = recode_string (dict_get_encoding (d), UTF8,
664 val_lab_get_escaped_label (vl), -1);
665 uint8_t len = MIN (strlen (label), 255);
667 write_value (w, val_lab_get_value (vl), width);
668 write_bytes (w, &len, 1);
669 write_bytes (w, label, len);
670 write_zeros (w, REM_RND_UP (len + 1, 8));
675 /* Value label variable record. */
676 write_int (w, 4); /* Record type. */
677 write_int (w, set->n_indexes);
678 for (j = 0; j < set->n_indexes; j++)
679 write_int (w, set->indexes[j] + 1);
682 for (i = 0; i < n_sets; i++)
684 struct label_set *set = sets[i];
690 hmap_destroy (&same_sets);
693 /* Writes record type 6, document record. */
695 write_documents (struct sfm_writer *w, const struct dictionary *d)
697 const struct string_array *docs = dict_get_documents (d);
698 const char *enc = dict_get_encoding (d);
701 write_int (w, 6); /* Record type. */
702 write_int (w, docs->n);
703 for (i = 0; i < docs->n; i++)
705 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
706 size_t s_len = strlen (s);
707 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
709 write_bytes (w, s, write_len);
710 write_spaces (w, DOC_LINE_LENGTH - write_len);
716 put_attrset (struct string *string, const struct attrset *attrs)
718 const struct attribute *attr;
719 struct attrset_iterator i;
721 for (attr = attrset_first (attrs, &i); attr != NULL;
722 attr = attrset_next (attrs, &i))
724 size_t n_values = attribute_get_n_values (attr);
727 ds_put_cstr (string, attribute_get_name (attr));
728 ds_put_byte (string, '(');
729 for (j = 0; j < n_values; j++)
730 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
731 ds_put_byte (string, ')');
736 write_data_file_attributes (struct sfm_writer *w,
737 const struct dictionary *d)
739 struct string s = DS_EMPTY_INITIALIZER;
740 put_attrset (&s, dict_get_attributes (d));
741 write_utf8_record (w, dict_get_encoding (d), &s, 17);
746 add_role_attribute (enum var_role role, struct attrset *attrs)
748 struct attribute *attr;
778 attrset_delete (attrs, "$@Role");
780 attr = attribute_create ("$@Role");
781 attribute_add_value (attr, s);
782 attrset_add (attrs, attr);
786 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
788 struct string s = DS_EMPTY_INITIALIZER;
789 size_t n_vars = dict_get_var_cnt (d);
790 size_t n_attrsets = 0;
793 for (i = 0; i < n_vars; i++)
795 struct variable *v = dict_get_var (d, i);
796 struct attrset attrs;
798 attrset_clone (&attrs, var_get_attributes (v));
800 add_role_attribute (var_get_role (v), &attrs);
802 ds_put_byte (&s, '/');
803 ds_put_format (&s, "%s:", var_get_name (v));
804 put_attrset (&s, &attrs);
805 attrset_destroy (&attrs);
808 write_utf8_record (w, dict_get_encoding (d), &s, 18);
812 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
813 SPSS before release 14, otherwise writes sets supported only by later
816 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
819 const char *encoding = dict_get_encoding (dict);
820 struct string s = DS_EMPTY_INITIALIZER;
824 if (is_encoding_ebcdic_compatible (encoding))
830 n_mrsets = dict_get_n_mrsets (dict);
834 for (i = 0; i < n_mrsets; i++)
836 const struct mrset *mrset = dict_get_mrset (dict, i);
840 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
844 name = recode_string (encoding, "UTF-8", mrset->name, -1);
845 ds_put_format (&s, "%s=", name);
848 if (mrset->type == MRSET_MD)
852 if (mrset->cat_source == MRSET_COUNTEDVALUES)
853 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
855 ds_put_byte (&s, 'D');
857 if (mrset->width == 0)
858 counted = xasprintf ("%.0f", mrset->counted.f);
860 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
862 ds_put_format (&s, "%zu %s", strlen (counted), counted);
866 ds_put_byte (&s, 'C');
867 ds_put_byte (&s, ' ');
869 if (mrset->label && !mrset->label_from_var_label)
871 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
872 ds_put_format (&s, "%zu %s", strlen (label), label);
876 ds_put_cstr (&s, "0 ");
878 for (j = 0; j < mrset->n_vars; j++)
880 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
881 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
882 char *short_name = recode_string (encoding, "UTF-8",
883 lower_name_utf8, -1);
884 ds_put_format (&s, " %s", short_name);
886 free (lower_name_utf8);
888 ds_put_byte (&s, '\n');
891 if (!ds_is_empty (&s))
892 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
896 /* Write the alignment, width and scale values. */
898 write_variable_display_parameters (struct sfm_writer *w,
899 const struct dictionary *dict)
903 write_int (w, 7); /* Record type. */
904 write_int (w, 11); /* Record subtype. */
905 write_int (w, 4); /* Data item (int32) size. */
906 write_int (w, w->segment_cnt * 3); /* Number of data items. */
908 for (i = 0; i < dict_get_var_cnt (dict); ++i)
910 struct variable *v = dict_get_var (dict, i);
911 int width = var_get_width (v);
912 int segment_cnt = sfm_width_to_segments (width);
913 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
914 : var_get_measure (v) == MEASURE_ORDINAL ? 2
916 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
917 : var_get_alignment (v) == ALIGN_RIGHT ? 1
921 for (i = 0; i < segment_cnt; i++)
923 int width_left = width - sfm_segment_effective_offset (width, i);
924 write_int (w, measure);
925 write_int (w, (i == 0 ? var_get_display_width (v)
926 : var_default_display_width (width_left)));
927 write_int (w, alignment);
932 /* Writes the table of lengths for very long string variables. */
934 write_vls_length_table (struct sfm_writer *w,
935 const struct dictionary *dict)
940 ds_init_empty (&map);
941 for (i = 0; i < dict_get_var_cnt (dict); ++i)
943 const struct variable *v = dict_get_var (dict, i);
944 if (sfm_width_to_segments (var_get_width (v)) > 1)
945 ds_put_format (&map, "%s=%05d%c\t",
946 var_get_short_name (v, 0), var_get_width (v), 0);
948 if (!ds_is_empty (&map))
949 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
954 write_long_string_value_labels (struct sfm_writer *w,
955 const struct dictionary *dict)
957 const char *encoding = dict_get_encoding (dict);
958 size_t n_vars = dict_get_var_cnt (dict);
962 /* Figure out the size in advance. */
964 for (i = 0; i < n_vars; i++)
966 struct variable *var = dict_get_var (dict, i);
967 const struct val_labs *val_labs = var_get_value_labels (var);
968 int width = var_get_width (var);
969 const struct val_lab *val_lab;
971 if (val_labs_count (val_labs) == 0 || width < 9)
975 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
976 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
977 val_lab = val_labs_next (val_labs, val_lab))
980 size += recode_string_len (encoding, "UTF-8",
981 val_lab_get_escaped_label (val_lab), -1);
987 write_int (w, 7); /* Record type. */
988 write_int (w, 21); /* Record subtype */
989 write_int (w, 1); /* Data item (byte) size. */
990 write_int (w, size); /* Number of data items. */
992 start = fbuf_tell (w->fbuf);
993 for (i = 0; i < n_vars; i++)
995 struct variable *var = dict_get_var (dict, i);
996 const struct val_labs *val_labs = var_get_value_labels (var);
997 int width = var_get_width (var);
998 const struct val_lab *val_lab;
1001 if (val_labs_count (val_labs) == 0 || width < 9)
1004 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1005 write_int (w, strlen (var_name));
1006 write_bytes (w, var_name, strlen (var_name));
1009 write_int (w, width);
1010 write_int (w, val_labs_count (val_labs));
1011 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
1012 val_lab = val_labs_next (val_labs, val_lab))
1017 write_int (w, width);
1018 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
1021 label = recode_string (var_get_encoding (var), "UTF-8",
1022 val_lab_get_escaped_label (val_lab), -1);
1023 len = strlen (label);
1025 write_bytes (w, label, len);
1029 assert (fbuf_tell (w->fbuf) == start + size);
1033 write_long_string_missing_values (struct sfm_writer *w,
1034 const struct dictionary *dict)
1036 const char *encoding = dict_get_encoding (dict);
1037 size_t n_vars = dict_get_var_cnt (dict);
1041 /* Figure out the size in advance. */
1043 for (i = 0; i < n_vars; i++)
1045 struct variable *var = dict_get_var (dict, i);
1046 const struct missing_values *mv = var_get_missing_values (var);
1047 int width = var_get_width (var);
1049 if (mv_is_empty (mv) || width < 9)
1053 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1055 size += mv_n_values (mv) * (4 + 8);
1060 write_int (w, 7); /* Record type. */
1061 write_int (w, 22); /* Record subtype */
1062 write_int (w, 1); /* Data item (byte) size. */
1063 write_int (w, size); /* Number of data items. */
1065 start = fbuf_tell (w->fbuf);
1066 for (i = 0; i < n_vars; i++)
1068 struct variable *var = dict_get_var (dict, i);
1069 const struct missing_values *mv = var_get_missing_values (var);
1070 int width = var_get_width (var);
1071 uint8_t n_missing_values;
1075 if (mv_is_empty (mv) || width < 9)
1078 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1079 write_int (w, strlen (var_name));
1080 write_bytes (w, var_name, strlen (var_name));
1083 n_missing_values = mv_n_values (mv);
1084 write_bytes (w, &n_missing_values, 1);
1086 for (j = 0; j < n_missing_values; j++)
1088 const union value *value = mv_get_value (mv, j);
1091 write_bytes (w, value_str (value, width), 8);
1094 assert (fbuf_tell (w->fbuf) == start + size);
1098 write_encoding_record (struct sfm_writer *w,
1099 const struct dictionary *d)
1101 /* IANA says "...character set names may be up to 40 characters taken
1102 from the printable characters of US-ASCII," so character set names
1103 don't need to be recoded to be in UTF-8.
1105 We convert encoding names to uppercase because SPSS writes encoding
1106 names in uppercase. */
1107 char *encoding = xstrdup (dict_get_encoding (d));
1108 str_uppercase (encoding);
1109 write_string_record (w, ss_cstr (encoding), 20);
1113 /* Writes the long variable name table. */
1115 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1120 ds_init_empty (&map);
1121 for (i = 0; i < dict_get_var_cnt (dict); i++)
1123 struct variable *v = dict_get_var (dict, i);
1125 ds_put_byte (&map, '\t');
1126 ds_put_format (&map, "%s=%s",
1127 var_get_short_name (v, 0), var_get_name (v));
1129 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1133 /* Write integer information record. */
1135 write_integer_info_record (struct sfm_writer *w,
1136 const struct dictionary *d)
1138 const char *dict_encoding = dict_get_encoding (d);
1139 int version_component[3];
1143 /* Parse the version string. */
1144 memset (version_component, 0, sizeof version_component);
1145 sscanf (bare_version, "%d.%d.%d",
1146 &version_component[0], &version_component[1], &version_component[2]);
1148 /* Figure out the floating-point format. */
1149 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1150 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1152 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1154 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1159 /* Choose codepage. */
1160 codepage = sys_get_codepage_from_encoding (dict_encoding);
1163 /* The codepage is unknown. Choose a default.
1165 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1167 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1168 many files use this codepage number regardless of their actual
1171 if (is_encoding_ascii_compatible (dict_encoding))
1173 else if (is_encoding_ebcdic_compatible (dict_encoding))
1178 write_int (w, 7); /* Record type. */
1179 write_int (w, 3); /* Record subtype. */
1180 write_int (w, 4); /* Data item (int32) size. */
1181 write_int (w, 8); /* Number of data items. */
1182 write_int (w, version_component[0]);
1183 write_int (w, version_component[1]);
1184 write_int (w, version_component[2]);
1185 write_int (w, -1); /* Machine code. */
1186 write_int (w, float_format);
1187 write_int (w, 1); /* Compression code. */
1188 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1189 write_int (w, codepage);
1192 /* Write floating-point information record. */
1194 write_float_info_record (struct sfm_writer *w)
1196 write_int (w, 7); /* Record type. */
1197 write_int (w, 4); /* Record subtype. */
1198 write_int (w, 8); /* Data item (flt64) size. */
1199 write_int (w, 3); /* Number of data items. */
1200 write_float (w, SYSMIS); /* System-missing value. */
1201 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1202 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1205 /* Writes case C to system file W. */
1207 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1210 struct sfm_writer *w = w_;
1212 if (fbuf_get_status (w->fbuf) > 0)
1214 casewriter_force_error (writer);
1221 if (w->compression == ANY_COMP_NONE)
1222 write_case_uncompressed (w, c);
1224 write_case_compressed (w, c);
1229 /* Destroys system file writer W. */
1231 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1233 struct sfm_writer *w = w_;
1234 if (!close_writer (w))
1235 casewriter_force_error (writer);
1238 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1240 write_error (const struct sfm_writer *writer)
1242 return fbuf_get_status (writer->fbuf) > 0;
1245 /* Closes a system file after we're done with it.
1246 Returns true if successful, false if an I/O error occurred. */
1248 close_writer (struct sfm_writer *w)
1256 if (w->fbuf != NULL)
1259 flush_compressed (w);
1260 if (w->compression == ANY_COMP_ZLIB)
1265 fbuf_flush (w->fbuf);
1267 ok = !write_error (w);
1269 /* Seek back to the beginning and update the number of cases.
1270 This is just a courtesy to later readers, so there's no need
1271 to check return values or report errors. */
1272 if (ok && w->case_cnt <= INT32_MAX && !fbuf_seek (w->fbuf, 80))
1274 write_int (w, w->case_cnt);
1275 fbuf_clear_status (w->fbuf);
1278 if (fbuf_close (w->fbuf) != 0)
1282 msg (ME, _("An I/O error occurred writing system file `%s'."),
1283 fh_get_file_name (w->fh));
1285 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1291 fh_unlock (w->lock);
1300 /* System file writer casewriter class. */
1301 static const struct casewriter_class sys_file_casewriter_class =
1303 sys_file_casewriter_write,
1304 sys_file_casewriter_destroy,
1308 /* Writes case C to system file W, without compressing it. */
1310 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1314 for (i = 0; i < w->sfm_var_cnt; i++)
1316 struct sfm_var *v = &w->sfm_vars[i];
1318 if (v->var_width == 0)
1319 write_float (w, case_num_idx (c, v->case_index));
1322 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1324 write_spaces (w, v->padding);
1329 /* Writes case C to system file W, with compression. */
1331 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1335 for (i = 0; i < w->sfm_var_cnt; i++)
1337 struct sfm_var *v = &w->sfm_vars[i];
1339 if (v->var_width == 0)
1341 double d = case_num_idx (c, v->case_index);
1343 put_cmp_opcode (w, 255);
1344 else if (d >= 1 - COMPRESSION_BIAS
1345 && d <= 251 - COMPRESSION_BIAS
1347 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1349 put_cmp_number (w, d);
1353 int offset = v->offset;
1356 /* This code properly deals with a width that is not a
1357 multiple of 8, by ensuring that the final partial
1358 oct (8 byte unit) is treated as padded with spaces
1360 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1362 const void *data = case_str_idx (c, v->case_index) + offset;
1363 int chunk_size = MIN (width, 8);
1364 if (!memcmp (data, " ", chunk_size))
1365 put_cmp_opcode (w, 254);
1367 put_cmp_string (w, data, chunk_size);
1370 /* This code deals properly with padding that is not a
1371 multiple of 8 bytes, by discarding the remainder,
1372 which was already effectively padded with spaces in
1373 the previous loop. (Note that v->width + v->padding
1374 is always a multiple of 8.) */
1375 for (padding = v->padding / 8; padding > 0; padding--)
1376 put_cmp_opcode (w, 254);
1382 start_zstream (struct sfm_writer *w)
1386 error = deflateInit (&w->zstream, 1);
1389 msg (ME, _("Failed to initialize ZLIB for compression (%s)."),
1397 finish_zstream (struct sfm_writer *w)
1399 struct zblock *block;
1402 assert (w->zstream.total_in <= ZBLOCK_SIZE);
1404 w->zstream.next_in = NULL;
1405 w->zstream.avail_in = 0;
1410 w->zstream.next_out = buf;
1411 w->zstream.avail_out = sizeof buf;
1412 error = deflate (&w->zstream, Z_FINISH);
1413 write_bytes (w, buf, w->zstream.next_out - buf);
1415 while (error == Z_OK);
1417 if (error != Z_STREAM_END)
1418 msg (ME, _("Failed to complete ZLIB stream compression (%s)."),
1421 if (w->n_blocks >= w->allocated_blocks)
1422 w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks,
1424 block = &w->blocks[w->n_blocks++];
1425 block->uncompressed_size = w->zstream.total_in;
1426 block->compressed_size = w->zstream.total_out;
1427 deflateEnd (&w->zstream);
1431 write_zlib (struct sfm_writer *w, const void *data_, unsigned int n)
1433 const uint8_t *data = data_;
1439 if (w->zstream.total_in >= ZBLOCK_SIZE)
1445 chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in);
1447 w->zstream.next_in = CONST_CAST (uint8_t *, data);
1448 w->zstream.avail_in = chunk;
1454 w->zstream.next_out = buf;
1455 w->zstream.avail_out = sizeof buf;
1456 error = deflate (&w->zstream, Z_NO_FLUSH);
1457 write_bytes (w, buf, w->zstream.next_out - buf);
1460 msg (ME, _("ZLIB stream compression failed (%s)."),
1465 while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0);
1472 write_ztrailer (struct sfm_writer *w)
1474 long long int uncompressed_ofs;
1475 long long int compressed_ofs;
1476 const struct zblock *block;
1478 write_int64 (w, -COMPRESSION_BIAS);
1480 write_int (w, ZBLOCK_SIZE);
1481 write_int (w, w->n_blocks);
1483 uncompressed_ofs = w->zstart;
1484 compressed_ofs = w->zstart + 24;
1485 for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++)
1487 write_int64 (w, uncompressed_ofs);
1488 write_int64 (w, compressed_ofs);
1489 write_int (w, block->uncompressed_size);
1490 write_int (w, block->compressed_size);
1492 uncompressed_ofs += block->uncompressed_size;
1493 compressed_ofs += block->compressed_size;
1496 if (!fbuf_seek (w->fbuf, w->zstart + 8))
1498 write_int64 (w, compressed_ofs);
1499 write_int64 (w, 24 + (w->n_blocks * 24));
1502 msg (ME, _("%s: Seek failed (%s)."),
1503 fh_get_file_name (w->fh), strerror (errno));
1506 /* Flushes buffered compressed opcodes and data to W. */
1508 flush_compressed (struct sfm_writer *w)
1512 unsigned int n = 8 * (1 + w->n_elements);
1513 if (w->compression == ANY_COMP_SIMPLE)
1514 write_bytes (w, w->cbuf, n);
1516 write_zlib (w, w->cbuf, n);
1518 w->n_opcodes = w->n_elements = 0;
1519 memset (w->cbuf[0], 0, 8);
1523 /* Appends OPCODE to the buffered set of compression opcodes in
1524 W. Flushes the compression buffer beforehand if necessary. */
1526 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1528 if (w->n_opcodes >= 8)
1529 flush_compressed (w);
1531 w->cbuf[0][w->n_opcodes++] = opcode;
1534 /* Appends NUMBER to the buffered compression data in W. */
1536 put_cmp_number (struct sfm_writer *w, double number)
1538 put_cmp_opcode (w, 253);
1539 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1542 /* Appends SIZE bytes of DATA to the buffered compression data in
1543 W, followed by enough spaces to pad the output data to exactly
1544 8 bytes (thus, SIZE must be no greater than 8). */
1546 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1550 put_cmp_opcode (w, 253);
1552 memset (w->cbuf[w->n_elements], w->space, 8);
1553 memcpy (w->cbuf[w->n_elements], data, size);
1556 /* Writes 32-bit integer X to the output file for writer W. */
1558 write_int (struct sfm_writer *w, int32_t x)
1560 write_bytes (w, &x, sizeof x);
1563 /* Writes 64-bit integer X to the output file for writer W. */
1565 write_int64 (struct sfm_writer *w, int64_t x)
1567 write_bytes (w, &x, sizeof x);
1570 /* Converts NATIVE to the 64-bit format used in output files in
1573 convert_double_to_output_format (double native, uint8_t output[8])
1575 /* If "double" is not a 64-bit type, then convert it to a
1576 64-bit type. Otherwise just copy it. */
1577 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1578 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1580 memcpy (output, &native, sizeof native);
1583 /* Writes floating-point number X to the output file for writer
1586 write_float (struct sfm_writer *w, double x)
1589 convert_double_to_output_format (x, output);
1590 write_bytes (w, output, sizeof output);
1593 /* Writes contents of VALUE with the given WIDTH to W, padding
1594 with zeros to a multiple of 8 bytes.
1595 To avoid a branch, and because we don't actually need to
1596 support it, WIDTH must be no bigger than 8. */
1598 write_value (struct sfm_writer *w, const union value *value, int width)
1600 assert (width <= 8);
1602 write_float (w, value->f);
1605 write_bytes (w, value_str (value, width), width);
1606 write_zeros (w, 8 - width);
1610 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1611 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1612 is padded on the right with spaces. */
1614 write_string (struct sfm_writer *w, const char *string, size_t width)
1616 size_t data_bytes = MIN (strlen (string), width);
1617 size_t pad_bytes = width - data_bytes;
1618 write_bytes (w, string, data_bytes);
1619 while (pad_bytes-- > 0)
1620 fbuf_putc (w->fbuf, w->space);
1623 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1624 recoded version in a field of the given WIDTH to W. The string is truncated
1625 or padded on the right with spaces to exactly WIDTH bytes. */
1627 write_utf8_string (struct sfm_writer *w, const char *encoding,
1628 const char *string, size_t width)
1630 char *s = recode_string (encoding, "UTF-8", string, -1);
1631 write_string (w, s, width);
1635 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1636 from UTF-8 encoded into ENCODING. */
1638 write_utf8_record (struct sfm_writer *w, const char *encoding,
1639 const struct string *content, int subtype)
1643 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1644 write_string_record (w, s, subtype);
1648 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1651 write_string_record (struct sfm_writer *w,
1652 const struct substring content, int subtype)
1655 write_int (w, subtype);
1657 write_int (w, ss_length (content));
1658 write_bytes (w, ss_data (content), ss_length (content));
1661 /* Writes SIZE bytes of DATA to W's output file. */
1663 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1665 fbuf_write (w->fbuf, data, size);
1668 /* Writes N zeros to W's output file. */
1670 write_zeros (struct sfm_writer *w, size_t n)
1673 fbuf_putc (w->fbuf, 0);
1676 /* Writes N spaces to W's output file. */
1678 write_spaces (struct sfm_writer *w, size_t n)
1681 fbuf_putc (w->fbuf, w->space);