1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000, 2006 Free Software Foundation, Inc.
4 This program is free software; you can redistribute it and/or
5 modify it under the terms of the GNU General Public License as
6 published by the Free Software Foundation; either version 2 of the
7 License, or (at your option) any later version.
9 This program is distributed in the hope that it will be useful, but
10 WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 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, write to the Free Software
16 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
20 #include "por-file-writer.h"
33 #include <data/case.h>
34 #include <data/casewriter-provider.h>
35 #include <data/casewriter.h>
36 #include <data/dictionary.h>
37 #include <data/file-handle-def.h>
38 #include <data/format.h>
39 #include <data/missing-values.h>
40 #include <data/value-labels.h>
41 #include <data/variable.h>
43 #include <libpspp/alloc.h>
44 #include <libpspp/hash.h>
45 #include <libpspp/magic.h>
46 #include <libpspp/message.h>
47 #include <libpspp/misc.h>
48 #include <libpspp/str.h>
49 #include <libpspp/version.h>
52 #define _(msgid) gettext (msgid)
54 /* Portable file writer. */
57 struct file_handle *fh; /* File handle. */
58 FILE *file; /* File stream. */
60 int lc; /* Number of characters on this line so far. */
62 size_t var_cnt; /* Number of variables. */
63 struct pfm_var *vars; /* Variables. */
65 int digits; /* Digits of precision. */
68 /* A variable to write to the portable file. */
71 int width; /* 0=numeric, otherwise string var width. */
72 int fv; /* Starting case index. */
75 static struct casewriter_class por_file_casewriter_class;
77 static bool close_writer (struct pfm_writer *);
78 static void buf_write (struct pfm_writer *, const void *, size_t);
79 static void write_header (struct pfm_writer *);
80 static void write_version_data (struct pfm_writer *);
81 static void write_variables (struct pfm_writer *, struct dictionary *);
82 static void write_value_labels (struct pfm_writer *,
83 const struct dictionary *);
85 static void format_trig_double (long double, int base_10_precision, char[]);
86 static char *format_trig_int (int, bool force_sign, char[]);
88 /* Returns default options for writing a portable file. */
89 struct pfm_write_options
90 pfm_writer_default_options (void)
92 struct pfm_write_options opts;
93 opts.create_writeable = true;
95 opts.digits = DBL_DIG;
99 /* Writes the dictionary DICT to portable file HANDLE according
100 to the given OPTS. Returns nonzero only if successful. DICT
101 will not be modified, except to assign short names. */
103 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
104 struct pfm_write_options opts)
106 struct pfm_writer *w = NULL;
112 mode = S_IRUSR | S_IRGRP | S_IROTH;
113 if (opts.create_writeable)
114 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
115 fd = open (fh_get_file_name (fh), O_WRONLY | O_CREAT | O_TRUNC, mode);
119 /* Open file handle. */
120 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
123 /* Initialize data structures. */
124 w = xmalloc (sizeof *w);
126 w->file = fdopen (fd, "w");
137 w->var_cnt = dict_get_var_cnt (dict);
138 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
139 for (i = 0; i < w->var_cnt; i++)
141 const struct variable *dv = dict_get_var (dict, i);
142 struct pfm_var *pv = &w->vars[i];
143 pv->width = var_get_width (dv);
144 pv->fv = var_get_case_index (dv);
147 w->digits = opts.digits;
150 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
155 /* Write file header. */
157 write_version_data (w);
158 write_variables (w, dict);
159 write_value_labels (w, dict);
160 buf_write (w, "F", 1);
161 if (ferror (w->file))
163 return casewriter_create (&por_file_casewriter_class, w);
170 msg (ME, _("An error occurred while opening \"%s\" for writing "
171 "as a portable file: %s."),
172 fh_get_file_name (fh), strerror (errno));
176 /* Write NBYTES starting at BUF to the portable file represented by
177 H. Break lines properly every 80 characters. */
179 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
181 const char *buf = buf_;
183 if (ferror (w->file))
186 assert (buf != NULL);
187 while (nbytes + w->lc >= 80)
189 size_t n = 80 - w->lc;
192 fwrite (buf, n, 1, w->file);
193 fwrite ("\r\n", 2, 1, w->file);
199 fwrite (buf, nbytes, 1, w->file);
204 /* Write D to the portable file as a floating-point field. */
206 write_float (struct pfm_writer *w, double d)
209 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
210 buf_write (w, buffer, strlen (buffer));
211 buf_write (w, "/", 1);
214 /* Write N to the portable file as an integer field. */
216 write_int (struct pfm_writer *w, int n)
219 format_trig_int (n, false, buffer);
220 buf_write (w, buffer, strlen (buffer));
221 buf_write (w, "/", 1);
224 /* Write S to the portable file as a string field. */
226 write_string (struct pfm_writer *w, const char *s)
228 size_t n = strlen (s);
229 write_int (w, (int) n);
233 /* Write file header. */
235 write_header (struct pfm_writer *w)
237 static const char spss2ascii[256] =
239 "0000000000000000000000000000000000000000000000000000000000000000"
240 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
241 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
242 "0000000000000000000000000000000000000000000000000000000000000000"
246 for (i = 0; i < 5; i++)
247 buf_write (w, "ASCII SPSS PORT FILE ", 40);
249 buf_write (w, spss2ascii, 256);
250 buf_write (w, "SPSSPORT", 8);
253 /* Writes version, date, and identification records. */
255 write_version_data (struct pfm_writer *w)
263 if ((time_t) -1 == time (&t))
265 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
270 tmp = localtime (&t);
272 sprintf (date_str, "%04d%02d%02d",
273 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
274 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
275 buf_write (w, "A", 1);
276 write_string (w, date_str);
277 write_string (w, time_str);
279 /* Product identification. */
280 buf_write (w, "1", 1);
281 write_string (w, version);
283 /* Subproduct identification. */
284 buf_write (w, "3", 1);
285 write_string (w, host_system);
288 /* Write format F to file H. */
290 write_format (struct pfm_writer *w, const struct fmt_spec *f)
292 write_int (w, fmt_to_io (f->type));
297 /* Write value V for variable VV to file H. */
299 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
301 if (var_is_numeric (vv))
302 write_float (w, v->f);
305 write_int (w, var_get_width (vv));
306 buf_write (w, v->s, var_get_width (vv));
310 /* Write variable records. */
312 write_variables (struct pfm_writer *w, struct dictionary *dict)
316 dict_assign_short_names (dict);
318 buf_write (w, "4", 1);
319 write_int (w, dict_get_var_cnt (dict));
322 for (i = 0; i < dict_get_var_cnt (dict); i++)
324 struct variable *v = dict_get_var (dict, i);
325 struct missing_values mv;
327 buf_write (w, "7", 1);
328 write_int (w, var_get_width (v));
329 write_string (w, var_get_short_name (v));
330 write_format (w, var_get_print_format (v));
331 write_format (w, var_get_write_format (v));
333 /* Write missing values. */
334 mv_copy (&mv, var_get_missing_values (v));
335 while (mv_has_range (&mv))
338 mv_pop_range (&mv, &x, &y);
341 buf_write (w, "9", 1);
344 else if (y == HIGHEST)
346 buf_write (w, "A", 1);
351 buf_write (w, "B", 1);
356 while (mv_has_value (&mv))
359 mv_pop_value (&mv, &value);
360 buf_write (w, "8", 1);
361 write_value (w, &value, v);
364 /* Write variable label. */
365 if (var_get_label (v) != NULL)
367 buf_write (w, "C", 1);
368 write_string (w, var_get_label (v));
373 /* Write value labels to disk. FIXME: Inefficient. */
375 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
379 for (i = 0; i < dict_get_var_cnt (dict); i++)
381 struct val_labs_iterator *j;
382 struct variable *v = dict_get_var (dict, i);
383 const struct val_labs *val_labs = var_get_value_labels (v);
386 if (val_labs == NULL)
389 buf_write (w, "D", 1);
391 write_string (w, var_get_short_name (v));
392 write_int (w, val_labs_count (val_labs));
394 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
395 vl = val_labs_next (val_labs, &j))
397 write_value (w, &vl->value, v);
398 write_string (w, vl->label);
403 /* Writes case C to the portable file represented by H. */
405 por_file_casewriter_write (struct casewriter *writer, void *w_,
408 struct pfm_writer *w = w_;
411 if (!ferror (w->file))
413 for (i = 0; i < w->var_cnt; i++)
415 struct pfm_var *v = &w->vars[i];
418 write_float (w, case_num_idx (c, v->fv));
421 write_int (w, v->width);
422 buf_write (w, case_str_idx (c, v->fv), v->width);
427 casewriter_force_error (writer);
433 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
435 struct pfm_writer *w = w_;
436 if (!close_writer (w))
437 casewriter_force_error (writer);
440 /* Closes a portable file after we're done with it.
441 Returns true if successful, false if an I/O error occurred. */
443 close_writer (struct pfm_writer *w)
454 memset (buf, 'Z', sizeof buf);
455 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
457 ok = !ferror (w->file);
458 if (fclose (w->file) == EOF)
462 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
463 fh_get_file_name (w->fh));
466 fh_close (w->fh, "portable file", "we");
474 /* Base-30 conversion.
476 Portable files represent numbers in base-30 format, so we need
477 to be able to convert real and integer number to that base.
478 Older versions of PSPP used libgmp to do so, but this added a
479 big library dependency to do just one thing. Now we do it
480 ourselves internally.
482 Important fact: base 30 is called "trigesimal". */
484 /* Conversion base. */
485 #define BASE 30 /* As an integer. */
486 #define LDBASE ((long double) BASE) /* As a long double. */
488 /* This is floor(log30(2**31)), the minimum number of trigesimal
489 digits that a `long int' can hold. */
492 /* pow_tab[i] = pow (30, pow (2, i)) */
493 static long double pow_tab[16];
495 /* Initializes pow_tab[]. */
499 static bool did_init = false;
503 /* Only initialize once. */
508 /* Set each element of pow_tab[] until we run out of numerical
511 for (power = 30.0L; power < DBL_MAX; power *= power)
513 assert (i < sizeof pow_tab / sizeof *pow_tab);
514 pow_tab[i++] = power;
518 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
520 pow30_nonnegative (int exponent)
525 assert (exponent >= 0);
526 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
529 for (i = 0; exponent > 0; exponent >>= 1, i++)
536 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
542 return pow30_nonnegative (exponent);
544 return 1.L / pow30_nonnegative (-exponent);
547 /* Returns the character corresponding to TRIG. */
549 trig_to_char (int trig)
551 assert (trig >= 0 && trig < 30);
552 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
555 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
556 null-terminated STRING. The trigesimal point is inserted
557 after TRIG_PLACES characters have been printed, if necessary
558 adding extra zeros at either end for correctness. Returns the
559 character after the formatted number. */
561 format_trig_digits (char *string,
562 const char trigs[], int trig_cnt, int trig_places)
567 while (trig_places++ < 0)
571 while (trig_cnt-- > 0)
573 if (trig_places-- == 0)
575 *string++ = trig_to_char (*trigs++);
577 while (trig_places-- > 0)
583 /* Helper function for format_trig_int() that formats VALUE as a
584 trigesimal integer at CP. VALUE must be nonnegative.
585 Returns the character following the formatted integer. */
587 recurse_format_trig_int (char *cp, int value)
589 int trig = value % BASE;
592 cp = recurse_format_trig_int (cp, value);
593 *cp++ = trig_to_char (trig);
597 /* Formats VALUE as a trigesimal integer in null-terminated
598 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
599 FORCE_SIGN is true, a sign is always inserted; otherwise, a
600 sign is only inserted if VALUE is negative. */
602 format_trig_int (int value, bool force_sign, char string[])
613 /* Format integer. */
614 string = recurse_format_trig_int (string, value);
619 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
620 rounding up or down. Returns true if TRIGS[] represents a
621 value greater than half, false if less than half. If TRIGS[]
622 is exactly half, examines TRIGS[-1] and returns true if odd,
623 false if even ("round to even"). */
625 should_round_up (const char trigs[], int trig_cnt)
627 assert (trig_cnt > 0);
629 if (*trigs < BASE / 2)
631 /* Less than half: round down. */
634 else if (*trigs > BASE / 2)
636 /* Greater than half: round up. */
641 /* Approximately half: look more closely. */
643 for (i = 1; i < trig_cnt; i++)
646 /* Slightly greater than half: round up. */
650 /* Exactly half: round to even. */
651 return trigs[-1] % 2;
655 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
656 carrying to the left as necessary. Returns true if
657 successful, false on failure (due to a carry out of the
658 leftmost position). */
660 try_round_up (char *trigs, int trig_cnt)
664 char *round_trig = trigs + --trig_cnt;
665 if (*round_trig != BASE - 1)
667 /* Round this trig up to the next value. */
672 /* Carry over to the next trig to the left. */
676 /* Ran out of trigs to carry. */
680 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
681 equivalent of at least BASE_10_PRECISION decimal digits of
682 precision. The output format may use conventional or
683 scientific notation. Missing, infinite, and extreme values
684 are represented with "*.". */
686 format_trig_double (long double value, int base_10_precision, char output[])
688 /* Original VALUE was negative? */
691 /* Number of significant trigesimals. */
692 int base_30_precision;
694 /* Base-2 significand and exponent for original VALUE. */
698 /* VALUE as a set of trigesimals. */
699 char buffer[DBL_DIG + 16];
703 /* Number of trigesimal places for trigs.
704 trigs[0] has coefficient 30**(trig_places - 1),
705 trigs[1] has coefficient 30**(trig_places - 2),
707 In other words, the trigesimal point is just before trigs[0].
711 /* Number of trigesimal places left to write into BUFFER. */
716 /* Handle special cases. */
722 /* Make VALUE positive. */
731 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
732 point left or right as necessary. We approximate the
733 base-30 exponent by obtaining the base-2 exponent, then
734 multiplying by log30(2). This approximation is sufficient
735 to ensure that the adjusted VALUE is always in the range
736 0...30**6, an invariant of the loop below. */
738 base_2_sig = frexp (value, &base_2_exp);
739 if (errno != 0 || !finite (base_2_sig))
741 if (base_2_exp == 0 && base_2_sig == 0.)
743 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
745 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
746 value *= pow30 (CHUNK_SIZE - trig_places);
748 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
751 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
753 trigs_to_output -= CHUNK_SIZE)
758 /* The current chunk is just the integer part of VALUE,
759 truncated to the nearest integer. The chunk fits in a
762 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
763 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
767 /* Append the chunk, in base 30, to trigs[]. */
768 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
770 trigs[trig_cnt + --trigs_left] = chunk % 30;
773 while (trigs_left > 0)
774 trigs[trig_cnt + --trigs_left] = 0;
775 trig_cnt += CHUNK_SIZE;
777 /* Proceed to the next chunk. */
780 value *= pow (LDBASE, CHUNK_SIZE);
783 /* Strip leading zeros. */
784 while (trig_cnt > 1 && *trigs == 0)
791 /* Round to requested precision, conservatively estimating the
792 required base-30 precision as 2/3 of the base-10 precision
793 (log30(10) = .68). */
794 assert (base_10_precision > 0);
795 if (base_10_precision > LDBL_DIG)
796 base_10_precision = LDBL_DIG;
797 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
798 if (trig_cnt > base_30_precision)
800 if (should_round_up (trigs + base_30_precision,
801 trig_cnt - base_30_precision))
803 /* Try to round up. */
804 if (try_round_up (trigs, base_30_precision))
806 /* Rounding up worked. */
807 trig_cnt = base_30_precision;
811 /* Couldn't round up because we ran out of trigs to
812 carry into. Do the carry here instead. */
821 trig_cnt = base_30_precision;
826 /* No rounding required: fewer digits available than
830 /* Strip trailing zeros. */
831 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
837 if (trig_places >= -1 && trig_places < trig_cnt + 3)
839 /* Use conventional notation. */
840 format_trig_digits (output, trigs, trig_cnt, trig_places);
844 /* Use scientific notation. */
846 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
847 op = format_trig_int (trig_places - trig_cnt, true, op);
852 strcpy (output, "0");
856 strcpy (output, "*.");
860 static struct casewriter_class por_file_casewriter_class =
862 por_file_casewriter_write,
863 por_file_casewriter_destroy,