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"
34 #include "dictionary.h"
35 #include "file-handle-def.h"
37 #include "missing-values.h"
38 #include "stat-macros.h"
39 #include "value-labels.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 /* Portable file writer. */
56 struct file_handle *fh; /* File handle. */
57 FILE *file; /* File stream. */
59 int lc; /* Number of characters on this line so far. */
61 size_t var_cnt; /* Number of variables. */
62 struct pfm_var *vars; /* Variables. */
64 int digits; /* Digits of precision. */
67 /* A variable to write to the portable file. */
70 int width; /* 0=numeric, otherwise string var width. */
71 int fv; /* Starting case index. */
74 static void buf_write (struct pfm_writer *, const void *, size_t);
75 static void write_header (struct pfm_writer *);
76 static void write_version_data (struct pfm_writer *);
77 static void write_variables (struct pfm_writer *, struct dictionary *);
78 static void write_value_labels (struct pfm_writer *,
79 const struct dictionary *);
81 static void format_trig_double (long double, int base_10_precision, char[]);
82 static char *format_trig_int (int, bool force_sign, char[]);
84 /* Returns default options for writing a portable file. */
85 struct pfm_write_options
86 pfm_writer_default_options (void)
88 struct pfm_write_options opts;
89 opts.create_writeable = true;
91 opts.digits = DBL_DIG;
95 /* Writes the dictionary DICT to portable file HANDLE according
96 to the given OPTS. Returns nonzero only if successful. DICT
97 will not be modified, except to assign short names. */
99 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
100 struct pfm_write_options opts)
102 struct pfm_writer *w = NULL;
108 mode = S_IRUSR | S_IRGRP | S_IROTH;
109 if (opts.create_writeable)
110 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
111 fd = open (fh_get_file_name (fh), O_WRONLY | O_CREAT | O_TRUNC, mode);
115 /* Open file handle. */
116 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
119 /* Initialize data structures. */
120 w = xmalloc (sizeof *w);
122 w->file = fdopen (fd, "w");
133 w->var_cnt = dict_get_var_cnt (dict);
134 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
135 for (i = 0; i < w->var_cnt; i++)
137 const struct variable *dv = dict_get_var (dict, i);
138 struct pfm_var *pv = &w->vars[i];
139 pv->width = var_get_width (dv);
140 pv->fv = var_get_case_index (dv);
143 w->digits = opts.digits;
146 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
151 /* Write file header. */
153 write_version_data (w);
154 write_variables (w, dict);
155 write_value_labels (w, dict);
156 buf_write (w, "F", 1);
157 if (pfm_write_error (w))
162 pfm_close_writer (w);
166 msg (ME, _("An error occurred while opening \"%s\" for writing "
167 "as a portable file: %s."),
168 fh_get_file_name (fh), strerror (errno));
172 /* Write NBYTES starting at BUF to the portable file represented by
173 H. Break lines properly every 80 characters. */
175 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
177 const char *buf = buf_;
179 if (ferror (w->file))
182 assert (buf != NULL);
183 while (nbytes + w->lc >= 80)
185 size_t n = 80 - w->lc;
188 fwrite (buf, n, 1, w->file);
189 fwrite ("\r\n", 2, 1, w->file);
195 fwrite (buf, nbytes, 1, w->file);
200 /* Write D to the portable file as a floating-point field. */
202 write_float (struct pfm_writer *w, double d)
205 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
206 buf_write (w, buffer, strlen (buffer));
207 buf_write (w, "/", 1);
210 /* Write N to the portable file as an integer field. */
212 write_int (struct pfm_writer *w, int n)
215 format_trig_int (n, false, buffer);
216 buf_write (w, buffer, strlen (buffer));
217 buf_write (w, "/", 1);
220 /* Write S to the portable file as a string field. */
222 write_string (struct pfm_writer *w, const char *s)
224 size_t n = strlen (s);
225 write_int (w, (int) n);
229 /* Write file header. */
231 write_header (struct pfm_writer *w)
233 static const char spss2ascii[256] =
235 "0000000000000000000000000000000000000000000000000000000000000000"
236 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
237 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
238 "0000000000000000000000000000000000000000000000000000000000000000"
242 for (i = 0; i < 5; i++)
243 buf_write (w, "ASCII SPSS PORT FILE ", 40);
245 buf_write (w, spss2ascii, 256);
246 buf_write (w, "SPSSPORT", 8);
249 /* Writes version, date, and identification records. */
251 write_version_data (struct pfm_writer *w)
259 if ((time_t) -1 == time (&t))
261 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
266 tmp = localtime (&t);
268 sprintf (date_str, "%04d%02d%02d",
269 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
270 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
271 buf_write (w, "A", 1);
272 write_string (w, date_str);
273 write_string (w, time_str);
275 /* Product identification. */
276 buf_write (w, "1", 1);
277 write_string (w, version);
279 /* Subproduct identification. */
280 buf_write (w, "3", 1);
281 write_string (w, host_system);
284 /* Write format F to file H. */
286 write_format (struct pfm_writer *w, const struct fmt_spec *f)
288 write_int (w, fmt_to_io (f->type));
293 /* Write value V for variable VV to file H. */
295 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
297 if (var_is_numeric (vv))
298 write_float (w, v->f);
301 write_int (w, var_get_width (vv));
302 buf_write (w, v->s, var_get_width (vv));
306 /* Write variable records. */
308 write_variables (struct pfm_writer *w, struct dictionary *dict)
312 dict_assign_short_names (dict);
314 buf_write (w, "4", 1);
315 write_int (w, dict_get_var_cnt (dict));
318 for (i = 0; i < dict_get_var_cnt (dict); i++)
320 struct variable *v = dict_get_var (dict, i);
321 struct missing_values mv;
323 buf_write (w, "7", 1);
324 write_int (w, var_get_width (v));
325 write_string (w, var_get_short_name (v));
326 write_format (w, var_get_print_format (v));
327 write_format (w, var_get_write_format (v));
329 /* Write missing values. */
330 mv_copy (&mv, var_get_missing_values (v));
331 while (mv_has_range (&mv))
334 mv_pop_range (&mv, &x, &y);
337 buf_write (w, "9", 1);
340 else if (y == HIGHEST)
342 buf_write (w, "A", 1);
347 buf_write (w, "B", 1);
352 while (mv_has_value (&mv))
355 mv_pop_value (&mv, &value);
356 buf_write (w, "8", 1);
357 write_value (w, &value, v);
360 if (var_get_label (v) != NULL)
362 buf_write (w, "C", 1);
363 write_string (w, var_get_label (v));
368 /* Write value labels to disk. FIXME: Inefficient. */
370 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
374 for (i = 0; i < dict_get_var_cnt (dict); i++)
376 struct val_labs_iterator *j;
377 struct variable *v = dict_get_var (dict, i);
378 const struct val_labs *val_labs = var_get_value_labels (v);
381 if (val_labs == NULL)
384 buf_write (w, "D", 1);
386 write_string (w, var_get_short_name (v));
387 write_int (w, val_labs_count (val_labs));
389 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
390 vl = val_labs_next (val_labs, &j))
392 write_value (w, &vl->value, v);
393 write_string (w, vl->label);
398 /* Writes case ELEM to the portable file represented by H. */
400 pfm_write_case (struct pfm_writer *w, const struct ccase *c)
404 if (ferror (w->file))
407 for (i = 0; i < w->var_cnt; i++)
409 struct pfm_var *v = &w->vars[i];
412 write_float (w, case_num_idx (c, v->fv));
415 write_int (w, v->width);
416 buf_write (w, case_str_idx (c, v->fv), v->width);
420 return !pfm_write_error (w);
424 pfm_write_error (const struct pfm_writer *w)
426 return ferror (w->file);
429 /* Closes a portable file after we're done with it.
430 Returns true if successful, false if an I/O error occurred. */
432 pfm_close_writer (struct pfm_writer *w)
443 memset (buf, 'Z', sizeof buf);
444 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
446 ok = !pfm_write_error (w);
447 if (fclose (w->file) == EOF)
451 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
452 fh_get_file_name (w->fh));
455 fh_close (w->fh, "portable file", "we");
463 /* Base-30 conversion.
465 Portable files represent numbers in base-30 format, so we need
466 to be able to convert real and integer number to that base.
467 Older versions of PSPP used libgmp to do so, but this added a
468 big library dependency to do just one thing. Now we do it
469 ourselves internally.
471 Important fact: base 30 is called "trigesimal". */
473 /* Conversion base. */
474 #define BASE 30 /* As an integer. */
475 #define LDBASE ((long double) BASE) /* As a long double. */
477 /* This is floor(log30(2**31)), the minimum number of trigesimal
478 digits that a `long int' can hold. */
481 /* pow_tab[i] = pow (30, pow (2, i)) */
482 static long double pow_tab[16];
484 /* Initializes pow_tab[]. */
488 static bool did_init = false;
492 /* Only initialize once. */
497 /* Set each element of pow_tab[] until we run out of numerical
500 for (power = 30.0L; power < DBL_MAX; power *= power)
502 assert (i < sizeof pow_tab / sizeof *pow_tab);
503 pow_tab[i++] = power;
507 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
509 pow30_nonnegative (int exponent)
514 assert (exponent >= 0);
515 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
518 for (i = 0; exponent > 0; exponent >>= 1, i++)
525 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
531 return pow30_nonnegative (exponent);
533 return 1.L / pow30_nonnegative (-exponent);
536 /* Returns the character corresponding to TRIG. */
538 trig_to_char (int trig)
540 assert (trig >= 0 && trig < 30);
541 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
544 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
545 null-terminated STRING. The trigesimal point is inserted
546 after TRIG_PLACES characters have been printed, if necessary
547 adding extra zeros at either end for correctness. Returns the
548 character after the formatted number. */
550 format_trig_digits (char *string,
551 const char trigs[], int trig_cnt, int trig_places)
556 while (trig_places++ < 0)
560 while (trig_cnt-- > 0)
562 if (trig_places-- == 0)
564 *string++ = trig_to_char (*trigs++);
566 while (trig_places-- > 0)
572 /* Helper function for format_trig_int() that formats VALUE as a
573 trigesimal integer at CP. VALUE must be nonnegative.
574 Returns the character following the formatted integer. */
576 recurse_format_trig_int (char *cp, int value)
578 int trig = value % BASE;
581 cp = recurse_format_trig_int (cp, value);
582 *cp++ = trig_to_char (trig);
586 /* Formats VALUE as a trigesimal integer in null-terminated
587 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
588 FORCE_SIGN is true, a sign is always inserted; otherwise, a
589 sign is only inserted if VALUE is negative. */
591 format_trig_int (int value, bool force_sign, char string[])
602 /* Format integer. */
603 string = recurse_format_trig_int (string, value);
608 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
609 rounding up or down. Returns true if TRIGS[] represents a
610 value greater than half, false if less than half. If TRIGS[]
611 is exactly half, examines TRIGS[-1] and returns true if odd,
612 false if even ("round to even"). */
614 should_round_up (const char trigs[], int trig_cnt)
616 assert (trig_cnt > 0);
618 if (*trigs < BASE / 2)
620 /* Less than half: round down. */
623 else if (*trigs > BASE / 2)
625 /* Greater than half: round up. */
630 /* Approximately half: look more closely. */
632 for (i = 1; i < trig_cnt; i++)
635 /* Slightly greater than half: round up. */
639 /* Exactly half: round to even. */
640 return trigs[-1] % 2;
644 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
645 carrying to the left as necessary. Returns true if
646 successful, false on failure (due to a carry out of the
647 leftmost position). */
649 try_round_up (char *trigs, int trig_cnt)
653 char *round_trig = trigs + --trig_cnt;
654 if (*round_trig != BASE - 1)
656 /* Round this trig up to the next value. */
661 /* Carry over to the next trig to the left. */
665 /* Ran out of trigs to carry. */
669 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
670 equivalent of at least BASE_10_PRECISION decimal digits of
671 precision. The output format may use conventional or
672 scientific notation. Missing, infinite, and extreme values
673 are represented with "*.". */
675 format_trig_double (long double value, int base_10_precision, char output[])
677 /* Original VALUE was negative? */
680 /* Number of significant trigesimals. */
681 int base_30_precision;
683 /* Base-2 significand and exponent for original VALUE. */
687 /* VALUE as a set of trigesimals. */
688 char buffer[DBL_DIG + 16];
692 /* Number of trigesimal places for trigs.
693 trigs[0] has coefficient 30**(trig_places - 1),
694 trigs[1] has coefficient 30**(trig_places - 2),
696 In other words, the trigesimal point is just before trigs[0].
700 /* Number of trigesimal places left to write into BUFFER. */
705 /* Handle special cases. */
711 /* Make VALUE positive. */
720 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
721 point left or right as necessary. We approximate the
722 base-30 exponent by obtaining the base-2 exponent, then
723 multiplying by log30(2). This approximation is sufficient
724 to ensure that the adjusted VALUE is always in the range
725 0...30**6, an invariant of the loop below. */
727 base_2_sig = frexp (value, &base_2_exp);
728 if (errno != 0 || !finite (base_2_sig))
730 if (base_2_exp == 0 && base_2_sig == 0.)
732 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
734 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
735 value *= pow30 (CHUNK_SIZE - trig_places);
737 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
740 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
742 trigs_to_output -= CHUNK_SIZE)
747 /* The current chunk is just the integer part of VALUE,
748 truncated to the nearest integer. The chunk fits in a
751 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
752 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
756 /* Append the chunk, in base 30, to trigs[]. */
757 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
759 trigs[trig_cnt + --trigs_left] = chunk % 30;
762 while (trigs_left > 0)
763 trigs[trig_cnt + --trigs_left] = 0;
764 trig_cnt += CHUNK_SIZE;
766 /* Proceed to the next chunk. */
769 value *= pow (LDBASE, CHUNK_SIZE);
772 /* Strip leading zeros. */
773 while (trig_cnt > 1 && *trigs == 0)
780 /* Round to requested precision, conservatively estimating the
781 required base-30 precision as 2/3 of the base-10 precision
782 (log30(10) = .68). */
783 assert (base_10_precision > 0);
784 if (base_10_precision > LDBL_DIG)
785 base_10_precision = LDBL_DIG;
786 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
787 if (trig_cnt > base_30_precision)
789 if (should_round_up (trigs + base_30_precision,
790 trig_cnt - base_30_precision))
792 /* Try to round up. */
793 if (try_round_up (trigs, base_30_precision))
795 /* Rounding up worked. */
796 trig_cnt = base_30_precision;
800 /* Couldn't round up because we ran out of trigs to
801 carry into. Do the carry here instead. */
810 trig_cnt = base_30_precision;
815 /* No rounding required: fewer digits available than
819 /* Strip trailing zeros. */
820 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
826 if (trig_places >= -1 && trig_places < trig_cnt + 3)
828 /* Use conventional notation. */
829 format_trig_digits (output, trigs, trig_cnt, trig_places);
833 /* Use scientific notation. */
835 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
836 op = format_trig_int (trig_places - trig_cnt, true, op);
841 strcpy (output, "0");
845 strcpy (output, "*.");