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;
207 if (opts.version != 2 && opts.version != 3)
209 msg (ME, _("Unknown system file version %d. Treating as version %d."),
214 /* Create and initialize writer. */
215 struct sfm_writer *w = XZALLOC (struct sfm_writer);
221 /* Use the requested compression, except that no EBCDIC-based ZLIB compressed
222 files have been observed, so drop back to simple compression for those
224 w->compression = opts.compression;
225 if (w->compression == ANY_COMP_ZLIB
226 && is_encoding_ebcdic_compatible (dict_get_encoding (d)))
227 w->compression = ANY_COMP_SIMPLE;
231 w->n_opcodes = w->n_elements = 0;
232 memset (w->cbuf[0], 0, 8);
234 /* Figure out how to map in-memory case data to on-disk case
235 data. Also count the number of segments. Very long strings
236 occupy multiple segments, otherwise each variable only takes
238 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
241 /* Open file handle as an exclusive writer. */
242 /* TRANSLATORS: this fragment will be interpolated into
243 messages in fh_lock() that identify types of files. */
244 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
248 /* Create the file on disk. */
250 if (opts.create_writeable)
252 w->rf = replace_file_start (fh, "wb", mode, &w->file);
255 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
256 fh_get_file_name (fh), strerror (errno));
260 get_encoding_info (&encoding_info, dict_get_encoding (d));
261 w->space = encoding_info.space[0];
263 /* Write the file header. */
266 /* Write basic variable info. */
267 short_names_assign (d);
268 for (i = 0; i < dict_get_var_cnt (d); i++)
269 write_variable (w, dict_get_var (d, i));
271 write_value_labels (w, d);
273 if (dict_get_document_line_cnt (d) > 0)
274 write_documents (w, d);
276 write_integer_info_record (w, d);
277 write_float_info_record (w);
279 write_mrsets (w, d, true);
281 write_variable_display_parameters (w, d);
283 if (opts.version >= 3)
284 write_longvar_table (w, d);
286 write_vls_length_table (w, d);
288 write_long_string_value_labels (w, d);
289 write_long_string_missing_values (w, d);
291 if (opts.version >= 3)
293 if (attrset_count (dict_get_attributes (d)))
294 write_data_file_attributes (w, d);
295 write_variable_attributes (w, d);
298 write_mrsets (w, d, false);
300 write_encoding_record (w, d);
302 /* Write end-of-headers record. */
306 if (w->compression == ANY_COMP_ZLIB)
308 w->zstream.zalloc = Z_NULL;
309 w->zstream.zfree = Z_NULL;
310 w->zstream.opaque = Z_NULL;
311 w->zstart = ftello (w->file);
313 write_int64 (w, w->zstart);
323 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
330 /* Returns value of X truncated to two least-significant digits. */
341 /* Calculates the offset of data for TARGET_VAR from the
342 beginning of each case's data for dictionary D. The return
343 value is in "octs" (8-byte units). */
345 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
351 for (i = 0; i < dict_get_var_cnt (d); i++)
353 struct variable *var = dict_get_var (d, i);
354 if (var == target_var)
356 oct_idx += sfm_width_to_octs (var_get_width (var));
361 /* Write the sysfile_header header to system file W. */
363 write_header (struct sfm_writer *w, const struct dictionary *d)
365 const char *dict_encoding = dict_get_encoding (d);
367 const char *file_label;
368 struct variable *weight;
372 /* Record-type code. */
373 if (is_encoding_ebcdic_compatible (dict_encoding))
374 write_string (w, EBCDIC_MAGIC, 4);
375 else if (w->compression == ANY_COMP_ZLIB)
376 write_string (w, ASCII_ZMAGIC, 4);
378 write_string (w, ASCII_MAGIC, 4);
380 /* Product identification. */
381 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
382 version, host_system);
383 write_utf8_string (w, dict_encoding, prod_name, 60);
388 /* Number of `union value's per case. */
389 write_int (w, calc_oct_idx (d, NULL));
392 write_int (w, (w->compression == ANY_COMP_NONE ? 0
393 : w->compression == ANY_COMP_SIMPLE ? 1
396 /* Weight variable. */
397 weight = dict_get_weight (d);
398 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
400 /* Number of cases. We don't know this in advance, so we write
401 -1 to indicate an unknown number of cases. Later we can
402 come back and overwrite it with the true value. */
405 /* Compression bias. */
406 write_float (w, COMPRESSION_BIAS);
408 /* Creation date and time. */
409 char *creation_date, *creation_time;
410 if (time (&t) == (time_t) -1)
412 creation_date = xstrdup ("01 Jan 70");
413 creation_time = xstrdup ("00:00:00");
417 static const char *const month_name[12] =
419 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
420 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
422 struct tm *tmp = localtime (&t);
423 int day = rerange (tmp->tm_mday);
424 int mon = rerange (tmp->tm_mon + 1);
425 int year = rerange (tmp->tm_year);
426 int hour = rerange (tmp->tm_hour + 1);
427 int min = rerange (tmp->tm_min + 1);
428 int sec = rerange (tmp->tm_sec + 1);
430 creation_date = xasprintf ("%02d %s %02d",
431 day, month_name[mon - 1], year);
432 creation_time = xasprintf ("%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
434 write_utf8_string (w, dict_encoding, creation_date, 9);
435 write_utf8_string (w, dict_encoding, creation_time, 8);
436 free (creation_time);
437 free (creation_date);
440 file_label = dict_get_label (d);
441 if (file_label == NULL)
443 write_utf8_string (w, dict_encoding, file_label, 64);
449 /* Write format spec FMT to W, after adjusting it to be
450 compatible with the given WIDTH. */
452 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
454 assert (fmt_check_output (&fmt));
455 assert (sfm_width_to_segments (width) == 1);
458 fmt_resize (&fmt, width);
459 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
462 /* Write a string continuation variable record for each 8-byte
463 section beyond the initial 8 bytes, for a variable of the
466 write_variable_continuation_records (struct sfm_writer *w, int width)
470 assert (sfm_width_to_segments (width) == 1);
471 for (position = 8; position < width; position += 8)
473 write_int (w, 2); /* Record type. */
474 write_int (w, -1); /* Width. */
475 write_int (w, 0); /* No variable label. */
476 write_int (w, 0); /* No missing values. */
477 write_int (w, 0); /* Print format. */
478 write_int (w, 0); /* Write format. */
479 write_zeros (w, 8); /* Name. */
483 /* Write the variable record(s) for variable V to system file
486 write_variable (struct sfm_writer *w, const struct variable *v)
488 int width = var_get_width (v);
489 int segment_cnt = sfm_width_to_segments (width);
490 int seg0_width = sfm_segment_alloc_width (width, 0);
491 const char *encoding = var_get_encoding (v);
498 write_int (w, seg0_width);
500 /* Variable has a variable label? */
501 write_int (w, var_has_label (v));
503 /* Number of missing values. If there is a range, then the
504 range counts as 2 missing values and causes the number to be
507 Missing values for long string variables are written in a separate
509 enum { MAX_SHORT_STRING = 8 };
510 if (width <= MAX_SHORT_STRING)
512 const struct missing_values *mv = var_get_missing_values (v);
513 if (mv_has_range (mv))
514 write_int (w, -2 - mv_n_values (mv));
516 write_int (w, mv_n_values (mv));
521 /* Print and write formats. */
522 write_format (w, *var_get_print_format (v), seg0_width);
523 write_format (w, *var_get_write_format (v), seg0_width);
526 The full name is in a translation table written
528 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
531 if (var_has_label (v))
533 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
534 size_t label_len = MIN (strlen (label), 255);
535 size_t padded_len = ROUND_UP (label_len, 4);
536 write_int (w, label_len);
537 write_string (w, label, padded_len);
541 /* Write the missing values, if any, range first. */
542 if (width <= MAX_SHORT_STRING)
544 const struct missing_values *mv = var_get_missing_values (v);
545 if (mv_has_range (mv))
548 mv_get_range (mv, &x, &y);
552 for (i = 0; i < mv_n_values (mv); i++)
553 write_value (w, mv_get_value (mv, i), width);
556 write_variable_continuation_records (w, seg0_width);
558 /* Write additional segments for very long string variables. */
559 for (i = 1; i < segment_cnt; i++)
561 int seg_width = sfm_segment_alloc_width (width, i);
562 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
564 write_int (w, 2); /* Variable record. */
565 write_int (w, seg_width); /* Width. */
566 write_int (w, 0); /* No variable label. */
567 write_int (w, 0); /* No missing values. */
568 write_format (w, fmt, seg_width); /* Print format. */
569 write_format (w, fmt, seg_width); /* Write format. */
570 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
572 write_variable_continuation_records (w, seg_width);
576 /* Writes the value labels to system file W.
578 Value labels for long string variables are written separately,
579 by write_long_string_value_labels. */
581 write_value_labels (struct sfm_writer *w, const struct dictionary *d)
585 struct hmap_node hmap_node;
586 const struct val_labs *val_labs;
588 size_t n_indexes, allocated_indexes;
591 size_t n_sets, allocated_sets;
592 struct label_set **sets;
593 struct hmap same_sets;
597 n_sets = allocated_sets = 0;
599 hmap_init (&same_sets);
602 for (i = 0; i < dict_get_var_cnt (d); i++)
604 struct variable *v = dict_get_var (d, i);
606 if (var_has_value_labels (v) && var_get_width (v) <= 8)
608 const struct val_labs *val_labs = var_get_value_labels (v);
609 unsigned int hash = val_labs_hash (val_labs, 0);
610 struct label_set *set;
612 HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node,
615 if (val_labs_equal (set->val_labs, val_labs))
617 if (set->n_indexes >= set->allocated_indexes)
618 set->indexes = x2nrealloc (set->indexes,
619 &set->allocated_indexes,
620 sizeof *set->indexes);
621 set->indexes[set->n_indexes++] = idx;
626 set = xmalloc (sizeof *set);
627 set->val_labs = val_labs;
628 set->indexes = xmalloc (sizeof *set->indexes);
629 set->indexes[0] = idx;
631 set->allocated_indexes = 1;
632 hmap_insert (&same_sets, &set->hmap_node, hash);
634 if (n_sets >= allocated_sets)
635 sets = x2nrealloc (sets, &allocated_sets, sizeof *sets);
636 sets[n_sets++] = set;
640 idx += sfm_width_to_octs (var_get_width (v));
643 for (i = 0; i < n_sets; i++)
645 const struct label_set *set = sets[i];
646 const struct val_labs *val_labs = set->val_labs;
647 size_t n_labels = val_labs_count (val_labs);
648 int width = val_labs_get_width (val_labs);
649 const struct val_lab **labels;
652 /* Value label record. */
653 write_int (w, 3); /* Record type. */
654 write_int (w, n_labels);
655 labels = val_labs_sorted (val_labs);
656 for (j = 0; j < n_labels; j++)
658 const struct val_lab *vl = labels[j];
659 char *label = recode_string (dict_get_encoding (d), UTF8,
660 val_lab_get_escaped_label (vl), -1);
661 uint8_t len = MIN (strlen (label), 255);
663 write_value (w, val_lab_get_value (vl), width);
664 write_bytes (w, &len, 1);
665 write_bytes (w, label, len);
666 write_zeros (w, REM_RND_UP (len + 1, 8));
671 /* Value label variable record. */
672 write_int (w, 4); /* Record type. */
673 write_int (w, set->n_indexes);
674 for (j = 0; j < set->n_indexes; j++)
675 write_int (w, set->indexes[j] + 1);
678 for (i = 0; i < n_sets; i++)
680 struct label_set *set = sets[i];
686 hmap_destroy (&same_sets);
689 /* Writes record type 6, document record. */
691 write_documents (struct sfm_writer *w, const struct dictionary *d)
693 const struct string_array *docs = dict_get_documents (d);
694 const char *enc = dict_get_encoding (d);
697 write_int (w, 6); /* Record type. */
698 write_int (w, docs->n);
699 for (i = 0; i < docs->n; i++)
701 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
702 size_t s_len = strlen (s);
703 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
705 write_bytes (w, s, write_len);
706 write_spaces (w, DOC_LINE_LENGTH - write_len);
712 put_attrset (struct string *string, const struct attrset *attrs)
714 const struct attribute *attr;
715 struct attrset_iterator i;
717 for (attr = attrset_first (attrs, &i); attr != NULL;
718 attr = attrset_next (attrs, &i))
720 size_t n_values = attribute_get_n_values (attr);
723 ds_put_cstr (string, attribute_get_name (attr));
724 ds_put_byte (string, '(');
725 for (j = 0; j < n_values; j++)
726 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
727 ds_put_byte (string, ')');
732 write_data_file_attributes (struct sfm_writer *w,
733 const struct dictionary *d)
735 struct string s = DS_EMPTY_INITIALIZER;
736 put_attrset (&s, dict_get_attributes (d));
737 write_utf8_record (w, dict_get_encoding (d), &s, 17);
742 add_role_attribute (enum var_role role, struct attrset *attrs)
744 struct attribute *attr;
774 attrset_delete (attrs, "$@Role");
776 attr = attribute_create ("$@Role");
777 attribute_add_value (attr, s);
778 attrset_add (attrs, attr);
782 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
784 struct string s = DS_EMPTY_INITIALIZER;
785 size_t n_vars = dict_get_var_cnt (d);
786 size_t n_attrsets = 0;
789 for (i = 0; i < n_vars; i++)
791 struct variable *v = dict_get_var (d, i);
792 struct attrset attrs;
794 attrset_clone (&attrs, var_get_attributes (v));
796 add_role_attribute (var_get_role (v), &attrs);
798 ds_put_byte (&s, '/');
799 ds_put_format (&s, "%s:", var_get_name (v));
800 put_attrset (&s, &attrs);
801 attrset_destroy (&attrs);
804 write_utf8_record (w, dict_get_encoding (d), &s, 18);
808 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
809 SPSS before release 14, otherwise writes sets supported only by later
812 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
815 const char *encoding = dict_get_encoding (dict);
816 struct string s = DS_EMPTY_INITIALIZER;
820 if (is_encoding_ebcdic_compatible (encoding))
826 n_mrsets = dict_get_n_mrsets (dict);
830 for (i = 0; i < n_mrsets; i++)
832 const struct mrset *mrset = dict_get_mrset (dict, i);
836 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
840 name = recode_string (encoding, "UTF-8", mrset->name, -1);
841 ds_put_format (&s, "%s=", name);
844 if (mrset->type == MRSET_MD)
848 if (mrset->cat_source == MRSET_COUNTEDVALUES)
849 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
851 ds_put_byte (&s, 'D');
853 if (mrset->width == 0)
854 counted = xasprintf ("%.0f", mrset->counted.f);
856 counted = xmemdup0 (mrset->counted.s, mrset->width);
857 ds_put_format (&s, "%zu %s", strlen (counted), counted);
861 ds_put_byte (&s, 'C');
862 ds_put_byte (&s, ' ');
864 if (mrset->label && !mrset->label_from_var_label)
866 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
867 ds_put_format (&s, "%zu %s", strlen (label), label);
871 ds_put_cstr (&s, "0 ");
873 for (j = 0; j < mrset->n_vars; j++)
875 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
876 char *lower_name_utf8 = utf8_to_lower (short_name_utf8);
877 char *short_name = recode_string (encoding, "UTF-8",
878 lower_name_utf8, -1);
879 ds_put_format (&s, " %s", short_name);
881 free (lower_name_utf8);
883 ds_put_byte (&s, '\n');
886 if (!ds_is_empty (&s))
887 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
891 /* Write the alignment, width and scale values. */
893 write_variable_display_parameters (struct sfm_writer *w,
894 const struct dictionary *dict)
898 write_int (w, 7); /* Record type. */
899 write_int (w, 11); /* Record subtype. */
900 write_int (w, 4); /* Data item (int32) size. */
901 write_int (w, w->segment_cnt * 3); /* Number of data items. */
903 for (i = 0; i < dict_get_var_cnt (dict); ++i)
905 struct variable *v = dict_get_var (dict, i);
906 int width = var_get_width (v);
907 int segment_cnt = sfm_width_to_segments (width);
908 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
909 : var_get_measure (v) == MEASURE_ORDINAL ? 2
911 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
912 : var_get_alignment (v) == ALIGN_RIGHT ? 1
916 for (i = 0; i < segment_cnt; i++)
918 int width_left = width - sfm_segment_effective_offset (width, i);
919 write_int (w, measure);
920 write_int (w, (i == 0 ? var_get_display_width (v)
921 : var_default_display_width (width_left)));
922 write_int (w, alignment);
927 /* Writes the table of lengths for very long string variables. */
929 write_vls_length_table (struct sfm_writer *w,
930 const struct dictionary *dict)
935 ds_init_empty (&map);
936 for (i = 0; i < dict_get_var_cnt (dict); ++i)
938 const struct variable *v = dict_get_var (dict, i);
939 if (sfm_width_to_segments (var_get_width (v)) > 1)
940 ds_put_format (&map, "%s=%05d%c\t",
941 var_get_short_name (v, 0), var_get_width (v), 0);
943 if (!ds_is_empty (&map))
944 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
949 write_long_string_value_labels (struct sfm_writer *w,
950 const struct dictionary *dict)
952 const char *encoding = dict_get_encoding (dict);
953 size_t n_vars = dict_get_var_cnt (dict);
956 /* Figure out the size in advance. */
958 for (i = 0; i < n_vars; i++)
960 struct variable *var = dict_get_var (dict, i);
961 const struct val_labs *val_labs = var_get_value_labels (var);
962 int width = var_get_width (var);
963 const struct val_lab *val_lab;
965 if (val_labs_count (val_labs) == 0 || width < 9)
969 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
970 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
971 val_lab = val_labs_next (val_labs, val_lab))
974 size += recode_string_len (encoding, "UTF-8",
975 val_lab_get_escaped_label (val_lab), -1);
981 write_int (w, 7); /* Record type. */
982 write_int (w, 21); /* Record subtype */
983 write_int (w, 1); /* Data item (byte) size. */
984 write_int (w, size); /* Number of data items. */
986 for (i = 0; i < n_vars; i++)
988 struct variable *var = dict_get_var (dict, i);
989 const struct val_labs *val_labs = var_get_value_labels (var);
990 int width = var_get_width (var);
991 const struct val_lab *val_lab;
994 if (val_labs_count (val_labs) == 0 || width < 9)
997 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
998 write_int (w, strlen (var_name));
999 write_bytes (w, var_name, strlen (var_name));
1002 write_int (w, width);
1003 write_int (w, val_labs_count (val_labs));
1004 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
1005 val_lab = val_labs_next (val_labs, val_lab))
1010 write_int (w, width);
1011 write_bytes (w, val_lab_get_value (val_lab)->s, width);
1013 label = recode_string (var_get_encoding (var), "UTF-8",
1014 val_lab_get_escaped_label (val_lab), -1);
1015 len = strlen (label);
1017 write_bytes (w, label, len);
1024 write_long_string_missing_values (struct sfm_writer *w,
1025 const struct dictionary *dict)
1027 const char *encoding = dict_get_encoding (dict);
1028 size_t n_vars = dict_get_var_cnt (dict);
1031 /* Figure out the size in advance. */
1033 for (i = 0; i < n_vars; i++)
1035 struct variable *var = dict_get_var (dict, i);
1036 const struct missing_values *mv = var_get_missing_values (var);
1037 int width = var_get_width (var);
1039 if (mv_is_empty (mv) || width < 9)
1043 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
1045 size += mv_n_values (mv) * (4 + 8);
1050 write_int (w, 7); /* Record type. */
1051 write_int (w, 22); /* Record subtype */
1052 write_int (w, 1); /* Data item (byte) size. */
1053 write_int (w, size); /* Number of data items. */
1055 for (i = 0; i < n_vars; i++)
1057 struct variable *var = dict_get_var (dict, i);
1058 const struct missing_values *mv = var_get_missing_values (var);
1059 int width = var_get_width (var);
1060 uint8_t n_missing_values;
1064 if (mv_is_empty (mv) || width < 9)
1067 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
1068 write_int (w, strlen (var_name));
1069 write_bytes (w, var_name, strlen (var_name));
1072 n_missing_values = mv_n_values (mv);
1073 write_bytes (w, &n_missing_values, 1);
1075 for (j = 0; j < n_missing_values; j++)
1077 const union value *value = mv_get_value (mv, j);
1080 write_bytes (w, value->s, 8);
1086 write_encoding_record (struct sfm_writer *w,
1087 const struct dictionary *d)
1089 /* IANA says "...character set names may be up to 40 characters taken
1090 from the printable characters of US-ASCII," so character set names
1091 don't need to be recoded to be in UTF-8.
1093 We convert encoding names to uppercase because SPSS writes encoding
1094 names in uppercase. */
1095 char *encoding = xstrdup (dict_get_encoding (d));
1096 str_uppercase (encoding);
1097 write_string_record (w, ss_cstr (encoding), 20);
1101 /* Writes the long variable name table. */
1103 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
1108 ds_init_empty (&map);
1109 for (i = 0; i < dict_get_var_cnt (dict); i++)
1111 struct variable *v = dict_get_var (dict, i);
1113 ds_put_byte (&map, '\t');
1114 ds_put_format (&map, "%s=%s",
1115 var_get_short_name (v, 0), var_get_name (v));
1117 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
1121 /* Write integer information record. */
1123 write_integer_info_record (struct sfm_writer *w,
1124 const struct dictionary *d)
1126 const char *dict_encoding = dict_get_encoding (d);
1127 int version_component[3];
1131 /* Parse the version string. */
1132 memset (version_component, 0, sizeof version_component);
1133 sscanf (bare_version, "%d.%d.%d",
1134 &version_component[0], &version_component[1], &version_component[2]);
1136 /* Figure out the floating-point format. */
1137 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
1138 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
1140 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
1142 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
1147 /* Choose codepage. */
1148 codepage = sys_get_codepage_from_encoding (dict_encoding);
1151 /* The codepage is unknown. Choose a default.
1153 For an EBCDIC-compatible encoding, use the value for EBCDIC.
1155 For an ASCII-compatible encoding, default to "7-bit ASCII", because
1156 many files use this codepage number regardless of their actual
1159 if (is_encoding_ascii_compatible (dict_encoding))
1161 else if (is_encoding_ebcdic_compatible (dict_encoding))
1166 write_int (w, 7); /* Record type. */
1167 write_int (w, 3); /* Record subtype. */
1168 write_int (w, 4); /* Data item (int32) size. */
1169 write_int (w, 8); /* Number of data items. */
1170 write_int (w, version_component[0]);
1171 write_int (w, version_component[1]);
1172 write_int (w, version_component[2]);
1173 write_int (w, -1); /* Machine code. */
1174 write_int (w, float_format);
1175 write_int (w, 1); /* Compression code. */
1176 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
1177 write_int (w, codepage);
1180 /* Write floating-point information record. */
1182 write_float_info_record (struct sfm_writer *w)
1184 write_int (w, 7); /* Record type. */
1185 write_int (w, 4); /* Record subtype. */
1186 write_int (w, 8); /* Data item (flt64) size. */
1187 write_int (w, 3); /* Number of data items. */
1188 write_float (w, SYSMIS); /* System-missing value. */
1189 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
1190 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
1193 /* Writes case C to system file W. */
1195 sys_file_casewriter_write (struct casewriter *writer, void *w_,
1198 struct sfm_writer *w = w_;
1200 if (ferror (w->file))
1202 casewriter_force_error (writer);
1209 if (w->compression == ANY_COMP_NONE)
1210 write_case_uncompressed (w, c);
1212 write_case_compressed (w, c);
1217 /* Destroys system file writer W. */
1219 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
1221 struct sfm_writer *w = w_;
1222 if (!close_writer (w))
1223 casewriter_force_error (writer);
1226 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
1228 write_error (const struct sfm_writer *writer)
1230 return ferror (writer->file);
1233 /* Closes a system file after we're done with it.
1234 Returns true if successful, false if an I/O error occurred. */
1236 close_writer (struct sfm_writer *w)
1244 if (w->file != NULL)
1247 flush_compressed (w);
1248 if (w->compression == ANY_COMP_ZLIB)
1255 ok = !write_error (w);
1257 /* Seek back to the beginning and update the number of cases.
1258 This is just a courtesy to later readers, so there's no need
1259 to check return values or report errors. */
1260 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1262 write_int (w, w->case_cnt);
1266 if (fclose (w->file) == EOF)
1270 msg (ME, _("An I/O error occurred writing system file `%s'."),
1271 fh_get_file_name (w->fh));
1273 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1279 fh_unlock (w->lock);
1288 /* System file writer casewriter class. */
1289 static const struct casewriter_class sys_file_casewriter_class =
1291 sys_file_casewriter_write,
1292 sys_file_casewriter_destroy,
1296 /* Writes case C to system file W, without compressing it. */
1298 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1302 for (i = 0; i < w->sfm_var_cnt; i++)
1304 struct sfm_var *v = &w->sfm_vars[i];
1306 if (v->var_width == 0)
1307 write_float (w, case_num_idx (c, v->case_index));
1310 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1312 write_spaces (w, v->padding);
1317 /* Writes case C to system file W, with compression. */
1319 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1323 for (i = 0; i < w->sfm_var_cnt; i++)
1325 struct sfm_var *v = &w->sfm_vars[i];
1327 if (v->var_width == 0)
1329 double d = case_num_idx (c, v->case_index);
1331 put_cmp_opcode (w, 255);
1332 else if (d >= 1 - COMPRESSION_BIAS
1333 && d <= 251 - COMPRESSION_BIAS
1335 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1337 put_cmp_number (w, d);
1341 int offset = v->offset;
1344 /* This code properly deals with a width that is not a
1345 multiple of 8, by ensuring that the final partial
1346 oct (8 byte unit) is treated as padded with spaces
1348 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1350 const void *data = case_str_idx (c, v->case_index) + offset;
1351 int chunk_size = MIN (width, 8);
1352 if (!memcmp (data, " ", chunk_size))
1353 put_cmp_opcode (w, 254);
1355 put_cmp_string (w, data, chunk_size);
1358 /* This code deals properly with padding that is not a
1359 multiple of 8 bytes, by discarding the remainder,
1360 which was already effectively padded with spaces in
1361 the previous loop. (Note that v->width + v->padding
1362 is always a multiple of 8.) */
1363 for (padding = v->padding / 8; padding > 0; padding--)
1364 put_cmp_opcode (w, 254);
1370 start_zstream (struct sfm_writer *w)
1374 error = deflateInit (&w->zstream, 1);
1377 msg (ME, _("Failed to initialize ZLIB for compression (%s)."),
1385 finish_zstream (struct sfm_writer *w)
1387 struct zblock *block;
1390 assert (w->zstream.total_in <= ZBLOCK_SIZE);
1392 w->zstream.next_in = NULL;
1393 w->zstream.avail_in = 0;
1398 w->zstream.next_out = buf;
1399 w->zstream.avail_out = sizeof buf;
1400 error = deflate (&w->zstream, Z_FINISH);
1401 write_bytes (w, buf, w->zstream.next_out - buf);
1403 while (error == Z_OK);
1405 if (error != Z_STREAM_END)
1406 msg (ME, _("Failed to complete ZLIB stream compression (%s)."),
1409 if (w->n_blocks >= w->allocated_blocks)
1410 w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks,
1412 block = &w->blocks[w->n_blocks++];
1413 block->uncompressed_size = w->zstream.total_in;
1414 block->compressed_size = w->zstream.total_out;
1415 deflateEnd (&w->zstream);
1419 write_zlib (struct sfm_writer *w, const void *data_, unsigned int n)
1421 const uint8_t *data = data_;
1427 if (w->zstream.total_in >= ZBLOCK_SIZE)
1433 chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in);
1435 w->zstream.next_in = CONST_CAST (uint8_t *, data);
1436 w->zstream.avail_in = chunk;
1442 w->zstream.next_out = buf;
1443 w->zstream.avail_out = sizeof buf;
1444 error = deflate (&w->zstream, Z_NO_FLUSH);
1445 write_bytes (w, buf, w->zstream.next_out - buf);
1448 msg (ME, _("ZLIB stream compression failed (%s)."),
1453 while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0);
1460 write_ztrailer (struct sfm_writer *w)
1462 long long int uncompressed_ofs;
1463 long long int compressed_ofs;
1464 const struct zblock *block;
1466 write_int64 (w, -COMPRESSION_BIAS);
1468 write_int (w, ZBLOCK_SIZE);
1469 write_int (w, w->n_blocks);
1471 uncompressed_ofs = w->zstart;
1472 compressed_ofs = w->zstart + 24;
1473 for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++)
1475 write_int64 (w, uncompressed_ofs);
1476 write_int64 (w, compressed_ofs);
1477 write_int (w, block->uncompressed_size);
1478 write_int (w, block->compressed_size);
1480 uncompressed_ofs += block->uncompressed_size;
1481 compressed_ofs += block->compressed_size;
1484 if (!fseeko (w->file, w->zstart + 8, SEEK_SET))
1486 write_int64 (w, compressed_ofs);
1487 write_int64 (w, 24 + (w->n_blocks * 24));
1490 msg (ME, _("%s: Seek failed (%s)."),
1491 fh_get_file_name (w->fh), strerror (errno));
1494 /* Flushes buffered compressed opcodes and data to W. */
1496 flush_compressed (struct sfm_writer *w)
1500 unsigned int n = 8 * (1 + w->n_elements);
1501 if (w->compression == ANY_COMP_SIMPLE)
1502 write_bytes (w, w->cbuf, n);
1504 write_zlib (w, w->cbuf, n);
1506 w->n_opcodes = w->n_elements = 0;
1507 memset (w->cbuf[0], 0, 8);
1511 /* Appends OPCODE to the buffered set of compression opcodes in
1512 W. Flushes the compression buffer beforehand if necessary. */
1514 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1516 if (w->n_opcodes >= 8)
1517 flush_compressed (w);
1519 w->cbuf[0][w->n_opcodes++] = opcode;
1522 /* Appends NUMBER to the buffered compression data in W. */
1524 put_cmp_number (struct sfm_writer *w, double number)
1526 put_cmp_opcode (w, 253);
1527 convert_double_to_output_format (number, w->cbuf[++w->n_elements]);
1530 /* Appends SIZE bytes of DATA to the buffered compression data in
1531 W, followed by enough spaces to pad the output data to exactly
1532 8 bytes (thus, SIZE must be no greater than 8). */
1534 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1538 put_cmp_opcode (w, 253);
1540 memset (w->cbuf[w->n_elements], w->space, 8);
1541 memcpy (w->cbuf[w->n_elements], data, size);
1544 /* Writes 32-bit integer X to the output file for writer W. */
1546 write_int (struct sfm_writer *w, int32_t x)
1548 write_bytes (w, &x, sizeof x);
1551 /* Writes 64-bit integer X to the output file for writer W. */
1553 write_int64 (struct sfm_writer *w, int64_t x)
1555 write_bytes (w, &x, sizeof x);
1558 /* Converts NATIVE to the 64-bit format used in output files in
1561 convert_double_to_output_format (double native, uint8_t output[8])
1563 /* If "double" is not a 64-bit type, then convert it to a
1564 64-bit type. Otherwise just copy it. */
1565 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1566 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1568 memcpy (output, &native, sizeof native);
1571 /* Writes floating-point number X to the output file for writer
1574 write_float (struct sfm_writer *w, double x)
1577 convert_double_to_output_format (x, output);
1578 write_bytes (w, output, sizeof output);
1581 /* Writes contents of VALUE with the given WIDTH to W, padding
1582 with zeros to a multiple of 8 bytes.
1583 To avoid a branch, and because we don't actually need to
1584 support it, WIDTH must be no bigger than 8. */
1586 write_value (struct sfm_writer *w, const union value *value, int width)
1588 assert (width <= 8);
1590 write_float (w, value->f);
1593 write_bytes (w, value->s, width);
1594 write_zeros (w, 8 - width);
1598 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1599 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1600 is padded on the right with spaces. */
1602 write_string (struct sfm_writer *w, const char *string, size_t width)
1604 size_t data_bytes = MIN (strlen (string), width);
1605 size_t pad_bytes = width - data_bytes;
1606 write_bytes (w, string, data_bytes);
1607 while (pad_bytes-- > 0)
1608 putc (w->space, w->file);
1611 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1612 recoded version in a field of the given WIDTH to W. The string is truncated
1613 or padded on the right with spaces to exactly WIDTH bytes. */
1615 write_utf8_string (struct sfm_writer *w, const char *encoding,
1616 const char *string, size_t width)
1618 char *s = recode_string (encoding, "UTF-8", string, -1);
1619 write_string (w, s, width);
1623 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1624 from UTF-8 encoded into ENCODING. */
1626 write_utf8_record (struct sfm_writer *w, const char *encoding,
1627 const struct string *content, int subtype)
1631 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1632 write_string_record (w, s, subtype);
1636 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1639 write_string_record (struct sfm_writer *w,
1640 const struct substring content, int subtype)
1643 write_int (w, subtype);
1645 write_int (w, ss_length (content));
1646 write_bytes (w, ss_data (content), ss_length (content));
1649 /* Writes SIZE bytes of DATA to W's output file. */
1651 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1653 fwrite (data, 1, size, w->file);
1656 /* Writes N zeros to W's output file. */
1658 write_zeros (struct sfm_writer *w, size_t n)
1664 /* Writes N spaces to W's output file. */
1666 write_spaces (struct sfm_writer *w, size_t n)
1669 putc (w->space, w->file);