1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006 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/>. */
18 #include "por-file-writer.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/missing-values.h>
38 #include <data/short-names.h>
39 #include <data/value-labels.h>
40 #include <data/variable.h>
42 #include <libpspp/alloc.h>
43 #include <libpspp/hash.h>
44 #include <libpspp/magic.h>
45 #include <libpspp/message.h>
46 #include <libpspp/misc.h>
47 #include <libpspp/str.h>
48 #include <libpspp/version.h>
51 #define _(msgid) gettext (msgid)
53 /* Maximum width of a variable in a portable file. */
54 #define MAX_POR_WIDTH 255
56 /* Portable file writer. */
59 struct file_handle *fh; /* File handle. */
60 FILE *file; /* File stream. */
62 int lc; /* Number of characters on this line so far. */
64 size_t var_cnt; /* Number of variables. */
65 struct pfm_var *vars; /* Variables. */
67 int digits; /* Digits of precision. */
70 /* A variable to write to the portable file. */
73 int width; /* 0=numeric, otherwise string var width. */
74 int fv; /* Starting case index. */
77 static struct casewriter_class por_file_casewriter_class;
79 static bool close_writer (struct pfm_writer *);
80 static void buf_write (struct pfm_writer *, const void *, size_t);
81 static void write_header (struct pfm_writer *);
82 static void write_version_data (struct pfm_writer *);
83 static void write_variables (struct pfm_writer *, struct dictionary *);
84 static void write_value_labels (struct pfm_writer *,
85 const struct dictionary *);
86 static void write_documents (struct pfm_writer *,
87 const struct dictionary *);
89 static void format_trig_double (long double, int base_10_precision, char[]);
90 static char *format_trig_int (int, bool force_sign, char[]);
92 /* Returns default options for writing a portable file. */
93 struct pfm_write_options
94 pfm_writer_default_options (void)
96 struct pfm_write_options opts;
97 opts.create_writeable = true;
99 opts.digits = DBL_DIG;
103 /* Writes the dictionary DICT to portable file HANDLE according
104 to the given OPTS. Returns nonzero only if successful. DICT
105 will not be modified, except to assign short names. */
107 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
108 struct pfm_write_options opts)
110 struct pfm_writer *w = NULL;
116 mode = S_IRUSR | S_IRGRP | S_IROTH;
117 if (opts.create_writeable)
118 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
119 fd = open (fh_get_file_name (fh), O_WRONLY | O_CREAT | O_TRUNC, mode);
123 /* Open file handle. */
124 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
127 /* Initialize data structures. */
128 w = xmalloc (sizeof *w);
130 w->file = fdopen (fd, "w");
141 w->var_cnt = dict_get_var_cnt (dict);
142 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
143 for (i = 0; i < w->var_cnt; i++)
145 const struct variable *dv = dict_get_var (dict, i);
146 struct pfm_var *pv = &w->vars[i];
147 pv->width = MIN (var_get_width (dv), MAX_POR_WIDTH);
148 pv->fv = var_get_case_index (dv);
151 w->digits = opts.digits;
154 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
159 /* Write file header. */
161 write_version_data (w);
162 write_variables (w, dict);
163 write_value_labels (w, dict);
164 if (dict_get_document_line_cnt (dict) > 0)
165 write_documents (w, dict);
166 buf_write (w, "F", 1);
167 if (ferror (w->file))
169 return casewriter_create (&por_file_casewriter_class, w);
176 msg (ME, _("An error occurred while opening \"%s\" for writing "
177 "as a portable file: %s."),
178 fh_get_file_name (fh), strerror (errno));
182 /* Write NBYTES starting at BUF to the portable file represented by
183 H. Break lines properly every 80 characters. */
185 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
187 const char *buf = buf_;
189 if (ferror (w->file))
192 assert (buf != NULL);
193 while (nbytes + w->lc >= 80)
195 size_t n = 80 - w->lc;
198 fwrite (buf, n, 1, w->file);
199 fwrite ("\r\n", 2, 1, w->file);
205 fwrite (buf, nbytes, 1, w->file);
210 /* Write D to the portable file as a floating-point field. */
212 write_float (struct pfm_writer *w, double d)
215 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
216 buf_write (w, buffer, strlen (buffer));
217 buf_write (w, "/", 1);
220 /* Write N to the portable file as an integer field. */
222 write_int (struct pfm_writer *w, int n)
225 format_trig_int (n, false, buffer);
226 buf_write (w, buffer, strlen (buffer));
227 buf_write (w, "/", 1);
230 /* Write S to the portable file as a string field. */
232 write_string (struct pfm_writer *w, const char *s)
234 size_t n = strlen (s);
235 write_int (w, (int) n);
239 /* Write file header. */
241 write_header (struct pfm_writer *w)
243 static const char spss2ascii[256] =
245 "0000000000000000000000000000000000000000000000000000000000000000"
246 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
247 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
248 "0000000000000000000000000000000000000000000000000000000000000000"
252 for (i = 0; i < 5; i++)
253 buf_write (w, "ASCII SPSS PORT FILE ", 40);
255 buf_write (w, spss2ascii, 256);
256 buf_write (w, "SPSSPORT", 8);
259 /* Writes version, date, and identification records. */
261 write_version_data (struct pfm_writer *w)
269 if ((time_t) -1 == time (&t))
271 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
276 tmp = localtime (&t);
278 sprintf (date_str, "%04d%02d%02d",
279 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
280 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
281 buf_write (w, "A", 1);
282 write_string (w, date_str);
283 write_string (w, time_str);
285 /* Product identification. */
286 buf_write (w, "1", 1);
287 write_string (w, version);
289 /* Subproduct identification. */
290 buf_write (w, "3", 1);
291 write_string (w, host_system);
294 /* Write format F to file H. The format is first resized to fit
295 a value of the given WIDTH, which is handy in case F
296 represents a string longer than 255 bytes and thus WIDTH is
297 truncated to 255 bytes. */
299 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
301 fmt_resize (&f, width);
302 write_int (w, fmt_to_io (f.type));
307 /* Write value V for variable VV to file H. */
309 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
311 if (var_is_numeric (vv))
312 write_float (w, v->f);
315 int width = MIN (var_get_width (vv), MAX_POR_WIDTH);
316 write_int (w, width);
317 buf_write (w, v->s, width);
321 /* Write variable records. */
323 write_variables (struct pfm_writer *w, struct dictionary *dict)
327 short_names_assign (dict);
329 if (dict_get_weight (dict) != NULL)
331 buf_write (w, "6", 1);
332 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
335 buf_write (w, "4", 1);
336 write_int (w, dict_get_var_cnt (dict));
339 for (i = 0; i < dict_get_var_cnt (dict); i++)
341 struct variable *v = dict_get_var (dict, i);
342 struct missing_values mv;
343 int width = MIN (var_get_width (v), MAX_POR_WIDTH);
345 buf_write (w, "7", 1);
346 write_int (w, width);
347 write_string (w, var_get_short_name (v, 0));
348 write_format (w, *var_get_print_format (v), width);
349 write_format (w, *var_get_write_format (v), width);
351 /* Write missing values. */
352 mv_copy (&mv, var_get_missing_values (v));
353 while (mv_has_range (&mv))
356 mv_pop_range (&mv, &x, &y);
359 buf_write (w, "9", 1);
362 else if (y == HIGHEST)
364 buf_write (w, "A", 1);
369 buf_write (w, "B", 1);
374 while (mv_has_value (&mv))
377 mv_pop_value (&mv, &value);
378 buf_write (w, "8", 1);
379 write_value (w, &value, v);
382 /* Write variable label. */
383 if (var_get_label (v) != NULL)
385 buf_write (w, "C", 1);
386 write_string (w, var_get_label (v));
391 /* Write value labels to disk. FIXME: Inefficient. */
393 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
397 for (i = 0; i < dict_get_var_cnt (dict); i++)
399 struct val_labs_iterator *j;
400 struct variable *v = dict_get_var (dict, i);
401 const struct val_labs *val_labs = var_get_value_labels (v);
404 if (val_labs == NULL)
407 buf_write (w, "D", 1);
409 write_string (w, var_get_short_name (v, 0));
410 write_int (w, val_labs_count (val_labs));
412 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
413 vl = val_labs_next (val_labs, &j))
415 write_value (w, &vl->value, v);
416 write_string (w, vl->label);
421 /* Write documents in DICT to portable file W. */
423 write_documents (struct pfm_writer *w, const struct dictionary *dict)
425 size_t line_cnt = dict_get_document_line_cnt (dict);
426 struct string line = DS_EMPTY_INITIALIZER;
429 buf_write (w, "E", 1);
430 write_int (w, line_cnt);
431 for (i = 0; i < line_cnt; i++)
433 dict_get_document_line (dict, i, &line);
434 write_string (w, ds_cstr (&line));
439 /* Writes case C to the portable file represented by WRITER. */
441 por_file_casewriter_write (struct casewriter *writer, void *w_,
444 struct pfm_writer *w = w_;
447 if (!ferror (w->file))
449 for (i = 0; i < w->var_cnt; i++)
451 struct pfm_var *v = &w->vars[i];
454 write_float (w, case_num_idx (c, v->fv));
457 write_int (w, v->width);
458 buf_write (w, case_str_idx (c, v->fv), v->width);
463 casewriter_force_error (writer);
469 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
471 struct pfm_writer *w = w_;
472 if (!close_writer (w))
473 casewriter_force_error (writer);
476 /* Closes a portable file after we're done with it.
477 Returns true if successful, false if an I/O error occurred. */
479 close_writer (struct pfm_writer *w)
490 memset (buf, 'Z', sizeof buf);
491 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
493 ok = !ferror (w->file);
494 if (fclose (w->file) == EOF)
498 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
499 fh_get_file_name (w->fh));
502 fh_close (w->fh, "portable file", "we");
510 /* Base-30 conversion.
512 Portable files represent numbers in base-30 format, so we need
513 to be able to convert real and integer number to that base.
514 Older versions of PSPP used libgmp to do so, but this added a
515 big library dependency to do just one thing. Now we do it
516 ourselves internally.
518 Important fact: base 30 is called "trigesimal". */
520 /* Conversion base. */
521 #define BASE 30 /* As an integer. */
522 #define LDBASE ((long double) BASE) /* As a long double. */
524 /* This is floor(log30(2**31)), the minimum number of trigesimal
525 digits that a `long int' can hold. */
528 /* pow_tab[i] = pow (30, pow (2, i)) */
529 static long double pow_tab[16];
531 /* Initializes pow_tab[]. */
535 static bool did_init = false;
539 /* Only initialize once. */
544 /* Set each element of pow_tab[] until we run out of numerical
547 for (power = 30.0L; power < DBL_MAX; power *= power)
549 assert (i < sizeof pow_tab / sizeof *pow_tab);
550 pow_tab[i++] = power;
554 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
556 pow30_nonnegative (int exponent)
561 assert (exponent >= 0);
562 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
565 for (i = 0; exponent > 0; exponent >>= 1, i++)
572 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
578 return pow30_nonnegative (exponent);
580 return 1.L / pow30_nonnegative (-exponent);
583 /* Returns the character corresponding to TRIG. */
585 trig_to_char (int trig)
587 assert (trig >= 0 && trig < 30);
588 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
591 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
592 null-terminated STRING. The trigesimal point is inserted
593 after TRIG_PLACES characters have been printed, if necessary
594 adding extra zeros at either end for correctness. Returns the
595 character after the formatted number. */
597 format_trig_digits (char *string,
598 const char trigs[], int trig_cnt, int trig_places)
603 while (trig_places++ < 0)
607 while (trig_cnt-- > 0)
609 if (trig_places-- == 0)
611 *string++ = trig_to_char (*trigs++);
613 while (trig_places-- > 0)
619 /* Helper function for format_trig_int() that formats VALUE as a
620 trigesimal integer at CP. VALUE must be nonnegative.
621 Returns the character following the formatted integer. */
623 recurse_format_trig_int (char *cp, int value)
625 int trig = value % BASE;
628 cp = recurse_format_trig_int (cp, value);
629 *cp++ = trig_to_char (trig);
633 /* Formats VALUE as a trigesimal integer in null-terminated
634 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
635 FORCE_SIGN is true, a sign is always inserted; otherwise, a
636 sign is only inserted if VALUE is negative. */
638 format_trig_int (int value, bool force_sign, char string[])
649 /* Format integer. */
650 string = recurse_format_trig_int (string, value);
655 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
656 rounding up or down. Returns true if TRIGS[] represents a
657 value greater than half, false if less than half. If TRIGS[]
658 is exactly half, examines TRIGS[-1] and returns true if odd,
659 false if even ("round to even"). */
661 should_round_up (const char trigs[], int trig_cnt)
663 assert (trig_cnt > 0);
665 if (*trigs < BASE / 2)
667 /* Less than half: round down. */
670 else if (*trigs > BASE / 2)
672 /* Greater than half: round up. */
677 /* Approximately half: look more closely. */
679 for (i = 1; i < trig_cnt; i++)
682 /* Slightly greater than half: round up. */
686 /* Exactly half: round to even. */
687 return trigs[-1] % 2;
691 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
692 carrying to the left as necessary. Returns true if
693 successful, false on failure (due to a carry out of the
694 leftmost position). */
696 try_round_up (char *trigs, int trig_cnt)
700 char *round_trig = trigs + --trig_cnt;
701 if (*round_trig != BASE - 1)
703 /* Round this trig up to the next value. */
708 /* Carry over to the next trig to the left. */
712 /* Ran out of trigs to carry. */
716 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
717 equivalent of at least BASE_10_PRECISION decimal digits of
718 precision. The output format may use conventional or
719 scientific notation. Missing, infinite, and extreme values
720 are represented with "*.". */
722 format_trig_double (long double value, int base_10_precision, char output[])
724 /* Original VALUE was negative? */
727 /* Number of significant trigesimals. */
728 int base_30_precision;
730 /* Base-2 significand and exponent for original VALUE. */
734 /* VALUE as a set of trigesimals. */
735 char buffer[DBL_DIG + 16];
739 /* Number of trigesimal places for trigs.
740 trigs[0] has coefficient 30**(trig_places - 1),
741 trigs[1] has coefficient 30**(trig_places - 2),
743 In other words, the trigesimal point is just before trigs[0].
747 /* Number of trigesimal places left to write into BUFFER. */
752 /* Handle special cases. */
758 /* Make VALUE positive. */
767 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
768 point left or right as necessary. We approximate the
769 base-30 exponent by obtaining the base-2 exponent, then
770 multiplying by log30(2). This approximation is sufficient
771 to ensure that the adjusted VALUE is always in the range
772 0...30**6, an invariant of the loop below. */
774 base_2_sig = frexp (value, &base_2_exp);
775 if (errno != 0 || !finite (base_2_sig))
777 if (base_2_exp == 0 && base_2_sig == 0.)
779 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
781 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
782 value *= pow30 (CHUNK_SIZE - trig_places);
784 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
787 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
789 trigs_to_output -= CHUNK_SIZE)
794 /* The current chunk is just the integer part of VALUE,
795 truncated to the nearest integer. The chunk fits in a
798 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
799 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
803 /* Append the chunk, in base 30, to trigs[]. */
804 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
806 trigs[trig_cnt + --trigs_left] = chunk % 30;
809 while (trigs_left > 0)
810 trigs[trig_cnt + --trigs_left] = 0;
811 trig_cnt += CHUNK_SIZE;
813 /* Proceed to the next chunk. */
816 value *= pow (LDBASE, CHUNK_SIZE);
819 /* Strip leading zeros. */
820 while (trig_cnt > 1 && *trigs == 0)
827 /* Round to requested precision, conservatively estimating the
828 required base-30 precision as 2/3 of the base-10 precision
829 (log30(10) = .68). */
830 assert (base_10_precision > 0);
831 if (base_10_precision > LDBL_DIG)
832 base_10_precision = LDBL_DIG;
833 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
834 if (trig_cnt > base_30_precision)
836 if (should_round_up (trigs + base_30_precision,
837 trig_cnt - base_30_precision))
839 /* Try to round up. */
840 if (try_round_up (trigs, base_30_precision))
842 /* Rounding up worked. */
843 trig_cnt = base_30_precision;
847 /* Couldn't round up because we ran out of trigs to
848 carry into. Do the carry here instead. */
857 trig_cnt = base_30_precision;
862 /* No rounding required: fewer digits available than
866 /* Strip trailing zeros. */
867 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
873 if (trig_places >= -1 && trig_places < trig_cnt + 3)
875 /* Use conventional notation. */
876 format_trig_digits (output, trigs, trig_cnt, trig_places);
880 /* Use scientific notation. */
882 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
883 op = format_trig_int (trig_places - trig_cnt, true, op);
888 strcpy (output, "0");
892 strcpy (output, "*.");
896 static struct casewriter_class por_file_casewriter_class =
898 por_file_casewriter_write,
899 por_file_casewriter_destroy,