1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
3 Written by Ben Pfaff <blp@gnu.org>.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 #include "por-file-writer.h"
22 #include <libpspp/message.h>
33 #include <libpspp/alloc.h>
35 #include "dictionary.h"
36 #include <libpspp/message.h>
37 #include "file-handle-def.h"
38 #include <libpspp/hash.h>
39 #include <libpspp/magic.h>
40 #include <libpspp/misc.h>
41 #include "stat-macros.h"
42 #include <libpspp/str.h>
43 #include "value-labels.h"
45 #include <libpspp/version.h>
48 #define _(msgid) gettext (msgid)
50 /* Portable file writer. */
53 struct file_handle *fh; /* File handle. */
54 FILE *file; /* File stream. */
56 int lc; /* Number of characters on this line so far. */
58 size_t var_cnt; /* Number of variables. */
59 struct pfm_var *vars; /* Variables. */
61 int digits; /* Digits of precision. */
64 /* A variable to write to the portable file. */
67 int width; /* 0=numeric, otherwise string var width. */
68 int fv; /* Starting case index. */
71 static void buf_write (struct pfm_writer *, const void *, size_t);
72 static void write_header (struct pfm_writer *);
73 static void write_version_data (struct pfm_writer *);
74 static void write_variables (struct pfm_writer *, struct dictionary *);
75 static void write_value_labels (struct pfm_writer *,
76 const struct dictionary *);
78 static void format_trig_double (long double, int base_10_precision, char[]);
79 static char *format_trig_int (int, bool force_sign, char[]);
81 /* Returns default options for writing a portable file. */
82 struct pfm_write_options
83 pfm_writer_default_options (void)
85 struct pfm_write_options opts;
86 opts.create_writeable = true;
88 opts.digits = DBL_DIG;
92 /* Writes the dictionary DICT to portable file HANDLE according
93 to the given OPTS. Returns nonzero only if successful. DICT
94 will not be modified, except to assign short names. */
96 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
97 struct pfm_write_options opts)
99 struct pfm_writer *w = NULL;
105 mode = S_IRUSR | S_IRGRP | S_IROTH;
106 if (opts.create_writeable)
107 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
108 fd = open (fh_get_file_name (fh), O_WRONLY | O_CREAT | O_TRUNC, mode);
112 /* Open file handle. */
113 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
116 /* Initialize data structures. */
117 w = xmalloc (sizeof *w);
119 w->file = fdopen (fd, "w");
130 w->var_cnt = dict_get_var_cnt (dict);
131 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
132 for (i = 0; i < w->var_cnt; i++)
134 const struct variable *dv = dict_get_var (dict, i);
135 struct pfm_var *pv = &w->vars[i];
136 pv->width = dv->width;
140 w->digits = opts.digits;
143 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
148 /* Write file header. */
150 write_version_data (w);
151 write_variables (w, dict);
152 write_value_labels (w, dict);
153 buf_write (w, "F", 1);
154 if (pfm_write_error (w))
159 pfm_close_writer (w);
163 msg (ME, _("An error occurred while opening \"%s\" for writing "
164 "as a portable file: %s."),
165 fh_get_file_name (fh), strerror (errno));
169 /* Write NBYTES starting at BUF to the portable file represented by
170 H. Break lines properly every 80 characters. */
172 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
174 const char *buf = buf_;
176 if (ferror (w->file))
179 assert (buf != NULL);
180 while (nbytes + w->lc >= 80)
182 size_t n = 80 - w->lc;
185 fwrite (buf, n, 1, w->file);
186 fwrite ("\r\n", 2, 1, w->file);
192 fwrite (buf, nbytes, 1, w->file);
197 /* Write D to the portable file as a floating-point field. */
199 write_float (struct pfm_writer *w, double d)
202 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
203 buf_write (w, buffer, strlen (buffer));
204 buf_write (w, "/", 1);
207 /* Write N to the portable file as an integer field. */
209 write_int (struct pfm_writer *w, int n)
212 format_trig_int (n, false, buffer);
213 buf_write (w, buffer, strlen (buffer));
214 buf_write (w, "/", 1);
217 /* Write S to the portable file as a string field. */
219 write_string (struct pfm_writer *w, const char *s)
221 size_t n = strlen (s);
222 write_int (w, (int) n);
226 /* Write file header. */
228 write_header (struct pfm_writer *w)
230 static const char spss2ascii[256] =
232 "0000000000000000000000000000000000000000000000000000000000000000"
233 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
234 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
235 "0000000000000000000000000000000000000000000000000000000000000000"
239 for (i = 0; i < 5; i++)
240 buf_write (w, "ASCII SPSS PORT FILE ", 40);
242 buf_write (w, spss2ascii, 256);
243 buf_write (w, "SPSSPORT", 8);
246 /* Writes version, date, and identification records. */
248 write_version_data (struct pfm_writer *w)
256 if ((time_t) -1 == time (&t))
258 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
263 tmp = localtime (&t);
265 sprintf (date_str, "%04d%02d%02d",
266 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
267 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
268 buf_write (w, "A", 1);
269 write_string (w, date_str);
270 write_string (w, time_str);
272 /* Product identification. */
273 buf_write (w, "1", 1);
274 write_string (w, version);
276 /* Subproduct identification. */
277 buf_write (w, "3", 1);
278 write_string (w, host_system);
281 /* Write format F to file H. */
283 write_format (struct pfm_writer *w, struct fmt_spec *f)
285 write_int (w, formats[f->type].spss);
290 /* Write value V for variable VV to file H. */
292 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
294 if (vv->type == NUMERIC)
295 write_float (w, v->f);
298 write_int (w, vv->width);
299 buf_write (w, v->s, vv->width);
303 /* Write variable records. */
305 write_variables (struct pfm_writer *w, struct dictionary *dict)
309 dict_assign_short_names (dict);
311 buf_write (w, "4", 1);
312 write_int (w, dict_get_var_cnt (dict));
315 for (i = 0; i < dict_get_var_cnt (dict); i++)
317 struct variable *v = dict_get_var (dict, i);
318 struct missing_values mv;
320 buf_write (w, "7", 1);
321 write_int (w, v->width);
322 write_string (w, v->short_name);
323 write_format (w, &v->print);
324 write_format (w, &v->write);
326 /* Write missing values. */
327 mv_copy (&mv, &v->miss);
328 while (mv_has_range (&mv))
331 mv_pop_range (&mv, &x, &y);
334 buf_write (w, "9", 1);
337 else if (y == HIGHEST)
339 buf_write (w, "A", 1);
344 buf_write (w, "B", 1);
349 while (mv_has_value (&mv))
352 mv_pop_value (&mv, &value);
353 buf_write (w, "8", 1);
354 write_value (w, &value, v);
359 buf_write (w, "C", 1);
360 write_string (w, v->label);
365 /* Write value labels to disk. FIXME: Inefficient. */
367 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
371 for (i = 0; i < dict_get_var_cnt (dict); i++)
373 struct val_labs_iterator *j;
374 struct variable *v = dict_get_var (dict, i);
377 if (!val_labs_count (v->val_labs))
380 buf_write (w, "D", 1);
382 write_string (w, v->short_name);
383 write_int (w, val_labs_count (v->val_labs));
385 for (vl = val_labs_first_sorted (v->val_labs, &j); vl != NULL;
386 vl = val_labs_next (v->val_labs, &j))
388 write_value (w, &vl->value, v);
389 write_string (w, vl->label);
394 /* Writes case ELEM to the portable file represented by H. */
396 pfm_write_case (struct pfm_writer *w, const struct ccase *c)
400 if (ferror (w->file))
403 for (i = 0; i < w->var_cnt; i++)
405 struct pfm_var *v = &w->vars[i];
408 write_float (w, case_num (c, v->fv));
411 write_int (w, v->width);
412 buf_write (w, case_str (c, v->fv), v->width);
416 return !pfm_write_error (w);
420 pfm_write_error (const struct pfm_writer *w)
422 return ferror (w->file);
425 /* Closes a portable file after we're done with it.
426 Returns true if successful, false if an I/O error occurred. */
428 pfm_close_writer (struct pfm_writer *w)
439 memset (buf, 'Z', sizeof buf);
440 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
442 ok = !pfm_write_error (w);
443 if (fclose (w->file) == EOF)
447 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
448 fh_get_file_name (w->fh));
451 fh_close (w->fh, "portable file", "we");
459 /* Base-30 conversion.
461 Portable files represent numbers in base-30 format, so we need
462 to be able to convert real and integer number to that base.
463 Older versions of PSPP used libgmp to do so, but this added a
464 big library dependency to do just one thing. Now we do it
465 ourselves internally.
467 Important fact: base 30 is called "trigesimal". */
469 /* Conversion base. */
470 #define BASE 30 /* As an integer. */
471 #define LDBASE ((long double) BASE) /* As a long double. */
473 /* This is floor(log30(2**31)), the minimum number of trigesimal
474 digits that a `long int' can hold. */
477 /* pow_tab[i] = pow (30, pow (2, i)) */
478 static long double pow_tab[16];
480 /* Initializes pow_tab[]. */
484 static bool did_init = false;
488 /* Only initialize once. */
493 /* Set each element of pow_tab[] until we run out of numerical
496 for (power = 30.0L; power < DBL_MAX; power *= power)
498 assert (i < sizeof pow_tab / sizeof *pow_tab);
499 pow_tab[i++] = power;
503 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
505 pow30_nonnegative (int exponent)
510 assert (exponent >= 0);
511 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
514 for (i = 0; exponent > 0; exponent >>= 1, i++)
521 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
527 return pow30_nonnegative (exponent);
529 return 1.L / pow30_nonnegative (-exponent);
532 /* Returns the character corresponding to TRIG. */
534 trig_to_char (int trig)
536 assert (trig >= 0 && trig < 30);
537 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
540 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
541 null-terminated STRING. The trigesimal point is inserted
542 after TRIG_PLACES characters have been printed, if necessary
543 adding extra zeros at either end for correctness. Returns the
544 character after the formatted number. */
546 format_trig_digits (char *string,
547 const char trigs[], int trig_cnt, int trig_places)
552 while (trig_places++ < 0)
556 while (trig_cnt-- > 0)
558 if (trig_places-- == 0)
560 *string++ = trig_to_char (*trigs++);
562 while (trig_places-- > 0)
568 /* Helper function for format_trig_int() that formats VALUE as a
569 trigesimal integer at CP. VALUE must be nonnegative.
570 Returns the character following the formatted integer. */
572 recurse_format_trig_int (char *cp, int value)
574 int trig = value % BASE;
577 cp = recurse_format_trig_int (cp, value);
578 *cp++ = trig_to_char (trig);
582 /* Formats VALUE as a trigesimal integer in null-terminated
583 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
584 FORCE_SIGN is true, a sign is always inserted; otherwise, a
585 sign is only inserted if VALUE is negative. */
587 format_trig_int (int value, bool force_sign, char string[])
598 /* Format integer. */
599 string = recurse_format_trig_int (string, value);
604 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
605 rounding up or down. Returns true if TRIGS[] represents a
606 value greater than half, false if less than half. If TRIGS[]
607 is exactly half, examines TRIGS[-1] and returns true if odd,
608 false if even ("round to even"). */
610 should_round_up (const char trigs[], int trig_cnt)
612 assert (trig_cnt > 0);
614 if (*trigs < BASE / 2)
616 /* Less than half: round down. */
619 else if (*trigs > BASE / 2)
621 /* Greater than half: round up. */
626 /* Approximately half: look more closely. */
628 for (i = 1; i < trig_cnt; i++)
631 /* Slightly greater than half: round up. */
635 /* Exactly half: round to even. */
636 return trigs[-1] % 2;
640 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
641 carrying to the left as necessary. Returns true if
642 successful, false on failure (due to a carry out of the
643 leftmost position). */
645 try_round_up (char *trigs, int trig_cnt)
649 char *round_trig = trigs + --trig_cnt;
650 if (*round_trig != BASE - 1)
652 /* Round this trig up to the next value. */
657 /* Carry over to the next trig to the left. */
661 /* Ran out of trigs to carry. */
665 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
666 equivalent of at least BASE_10_PRECISION decimal digits of
667 precision. The output format may use conventional or
668 scientific notation. Missing, infinite, and extreme values
669 are represented with "*.". */
671 format_trig_double (long double value, int base_10_precision, char output[])
673 /* Original VALUE was negative? */
676 /* Number of significant trigesimals. */
677 int base_30_precision;
679 /* Base-2 significand and exponent for original VALUE. */
683 /* VALUE as a set of trigesimals. */
684 char buffer[DBL_DIG + 16];
688 /* Number of trigesimal places for trigs.
689 trigs[0] has coefficient 30**(trig_places - 1),
690 trigs[1] has coefficient 30**(trig_places - 2),
692 In other words, the trigesimal point is just before trigs[0].
696 /* Number of trigesimal places left to write into BUFFER. */
701 /* Handle special cases. */
707 /* Make VALUE positive. */
716 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
717 point left or right as necessary. We approximate the
718 base-30 exponent by obtaining the base-2 exponent, then
719 multiplying by log30(2). This approximation is sufficient
720 to ensure that the adjusted VALUE is always in the range
721 0...30**6, an invariant of the loop below. */
723 base_2_sig = frexp (value, &base_2_exp);
724 if (errno != 0 || !finite (base_2_sig))
726 if (base_2_exp == 0 && base_2_sig == 0.)
728 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
730 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
731 value *= pow30 (CHUNK_SIZE - trig_places);
733 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
736 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
738 trigs_to_output -= CHUNK_SIZE)
743 /* The current chunk is just the integer part of VALUE,
744 truncated to the nearest integer. The chunk fits in a
747 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
748 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
752 /* Append the chunk, in base 30, to trigs[]. */
753 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
755 trigs[trig_cnt + --trigs_left] = chunk % 30;
758 while (trigs_left > 0)
759 trigs[trig_cnt + --trigs_left] = 0;
760 trig_cnt += CHUNK_SIZE;
762 /* Proceed to the next chunk. */
765 value *= pow (LDBASE, CHUNK_SIZE);
768 /* Strip leading zeros. */
769 while (trig_cnt > 1 && *trigs == 0)
776 /* Round to requested precision, conservatively estimating the
777 required base-30 precision as 2/3 of the base-10 precision
778 (log30(10) = .68). */
779 assert (base_10_precision > 0);
780 if (base_10_precision > LDBL_DIG)
781 base_10_precision = LDBL_DIG;
782 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
783 if (trig_cnt > base_30_precision)
785 if (should_round_up (trigs + base_30_precision,
786 trig_cnt - base_30_precision))
788 /* Try to round up. */
789 if (try_round_up (trigs, base_30_precision))
791 /* Rounding up worked. */
792 trig_cnt = base_30_precision;
796 /* Couldn't round up because we ran out of trigs to
797 carry into. Do the carry here instead. */
806 trig_cnt = base_30_precision;
811 /* No rounding required: fewer digits available than
815 /* Strip trailing zeros. */
816 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
822 if (trig_places >= -1 && trig_places < trig_cnt + 3)
824 /* Use conventional notation. */
825 format_trig_digits (output, trigs, trig_cnt, trig_places);
829 /* Use scientific notation. */
831 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
832 op = format_trig_int (trig_places - trig_cnt, true, op);
837 strcpy (output, "0");
841 strcpy (output, "*.");