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/value-labels.h>
39 #include <data/variable.h>
41 #include <libpspp/alloc.h>
42 #include <libpspp/hash.h>
43 #include <libpspp/magic.h>
44 #include <libpspp/message.h>
45 #include <libpspp/misc.h>
46 #include <libpspp/str.h>
47 #include <libpspp/version.h>
50 #define _(msgid) gettext (msgid)
52 /* Portable file writer. */
55 struct file_handle *fh; /* File handle. */
56 FILE *file; /* File stream. */
58 int lc; /* Number of characters on this line so far. */
60 size_t var_cnt; /* Number of variables. */
61 struct pfm_var *vars; /* Variables. */
63 int digits; /* Digits of precision. */
66 /* A variable to write to the portable file. */
69 int width; /* 0=numeric, otherwise string var width. */
70 int fv; /* Starting case index. */
73 static struct casewriter_class por_file_casewriter_class;
75 static bool close_writer (struct pfm_writer *);
76 static void buf_write (struct pfm_writer *, const void *, size_t);
77 static void write_header (struct pfm_writer *);
78 static void write_version_data (struct pfm_writer *);
79 static void write_variables (struct pfm_writer *, struct dictionary *);
80 static void write_value_labels (struct pfm_writer *,
81 const struct dictionary *);
83 static void format_trig_double (long double, int base_10_precision, char[]);
84 static char *format_trig_int (int, bool force_sign, char[]);
86 /* Returns default options for writing a portable file. */
87 struct pfm_write_options
88 pfm_writer_default_options (void)
90 struct pfm_write_options opts;
91 opts.create_writeable = true;
93 opts.digits = DBL_DIG;
97 /* Writes the dictionary DICT to portable file HANDLE according
98 to the given OPTS. Returns nonzero only if successful. DICT
99 will not be modified, except to assign short names. */
101 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
102 struct pfm_write_options opts)
104 struct pfm_writer *w = NULL;
110 mode = S_IRUSR | S_IRGRP | S_IROTH;
111 if (opts.create_writeable)
112 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
113 fd = open (fh_get_file_name (fh), O_WRONLY | O_CREAT | O_TRUNC, mode);
117 /* Open file handle. */
118 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
121 /* Initialize data structures. */
122 w = xmalloc (sizeof *w);
124 w->file = fdopen (fd, "w");
135 w->var_cnt = dict_get_var_cnt (dict);
136 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
137 for (i = 0; i < w->var_cnt; i++)
139 const struct variable *dv = dict_get_var (dict, i);
140 struct pfm_var *pv = &w->vars[i];
141 pv->width = var_get_width (dv);
142 pv->fv = var_get_case_index (dv);
145 w->digits = opts.digits;
148 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
153 /* Write file header. */
155 write_version_data (w);
156 write_variables (w, dict);
157 write_value_labels (w, dict);
158 buf_write (w, "F", 1);
159 if (ferror (w->file))
161 return casewriter_create (&por_file_casewriter_class, w);
168 msg (ME, _("An error occurred while opening \"%s\" for writing "
169 "as a portable file: %s."),
170 fh_get_file_name (fh), strerror (errno));
174 /* Write NBYTES starting at BUF to the portable file represented by
175 H. Break lines properly every 80 characters. */
177 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
179 const char *buf = buf_;
181 if (ferror (w->file))
184 assert (buf != NULL);
185 while (nbytes + w->lc >= 80)
187 size_t n = 80 - w->lc;
190 fwrite (buf, n, 1, w->file);
191 fwrite ("\r\n", 2, 1, w->file);
197 fwrite (buf, nbytes, 1, w->file);
202 /* Write D to the portable file as a floating-point field. */
204 write_float (struct pfm_writer *w, double d)
207 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
208 buf_write (w, buffer, strlen (buffer));
209 buf_write (w, "/", 1);
212 /* Write N to the portable file as an integer field. */
214 write_int (struct pfm_writer *w, int n)
217 format_trig_int (n, false, buffer);
218 buf_write (w, buffer, strlen (buffer));
219 buf_write (w, "/", 1);
222 /* Write S to the portable file as a string field. */
224 write_string (struct pfm_writer *w, const char *s)
226 size_t n = strlen (s);
227 write_int (w, (int) n);
231 /* Write file header. */
233 write_header (struct pfm_writer *w)
235 static const char spss2ascii[256] =
237 "0000000000000000000000000000000000000000000000000000000000000000"
238 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
239 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
240 "0000000000000000000000000000000000000000000000000000000000000000"
244 for (i = 0; i < 5; i++)
245 buf_write (w, "ASCII SPSS PORT FILE ", 40);
247 buf_write (w, spss2ascii, 256);
248 buf_write (w, "SPSSPORT", 8);
251 /* Writes version, date, and identification records. */
253 write_version_data (struct pfm_writer *w)
261 if ((time_t) -1 == time (&t))
263 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
268 tmp = localtime (&t);
270 sprintf (date_str, "%04d%02d%02d",
271 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
272 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
273 buf_write (w, "A", 1);
274 write_string (w, date_str);
275 write_string (w, time_str);
277 /* Product identification. */
278 buf_write (w, "1", 1);
279 write_string (w, version);
281 /* Subproduct identification. */
282 buf_write (w, "3", 1);
283 write_string (w, host_system);
286 /* Write format F to file H. */
288 write_format (struct pfm_writer *w, const struct fmt_spec *f)
290 write_int (w, fmt_to_io (f->type));
295 /* Write value V for variable VV to file H. */
297 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
299 if (var_is_numeric (vv))
300 write_float (w, v->f);
303 write_int (w, var_get_width (vv));
304 buf_write (w, v->s, var_get_width (vv));
308 /* Write variable records. */
310 write_variables (struct pfm_writer *w, struct dictionary *dict)
314 dict_assign_short_names (dict);
316 buf_write (w, "4", 1);
317 write_int (w, dict_get_var_cnt (dict));
320 for (i = 0; i < dict_get_var_cnt (dict); i++)
322 struct variable *v = dict_get_var (dict, i);
323 struct missing_values mv;
325 buf_write (w, "7", 1);
326 write_int (w, var_get_width (v));
327 write_string (w, var_get_short_name (v));
328 write_format (w, var_get_print_format (v));
329 write_format (w, var_get_write_format (v));
331 /* Write missing values. */
332 mv_copy (&mv, var_get_missing_values (v));
333 while (mv_has_range (&mv))
336 mv_pop_range (&mv, &x, &y);
339 buf_write (w, "9", 1);
342 else if (y == HIGHEST)
344 buf_write (w, "A", 1);
349 buf_write (w, "B", 1);
354 while (mv_has_value (&mv))
357 mv_pop_value (&mv, &value);
358 buf_write (w, "8", 1);
359 write_value (w, &value, v);
362 /* Write variable label. */
363 if (var_get_label (v) != NULL)
365 buf_write (w, "C", 1);
366 write_string (w, var_get_label (v));
371 /* Write value labels to disk. FIXME: Inefficient. */
373 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
377 for (i = 0; i < dict_get_var_cnt (dict); i++)
379 struct val_labs_iterator *j;
380 struct variable *v = dict_get_var (dict, i);
381 const struct val_labs *val_labs = var_get_value_labels (v);
384 if (val_labs == NULL)
387 buf_write (w, "D", 1);
389 write_string (w, var_get_short_name (v));
390 write_int (w, val_labs_count (val_labs));
392 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
393 vl = val_labs_next (val_labs, &j))
395 write_value (w, &vl->value, v);
396 write_string (w, vl->label);
401 /* Writes case C to the portable file represented by H. */
403 por_file_casewriter_write (struct casewriter *writer, void *w_,
406 struct pfm_writer *w = w_;
409 if (!ferror (w->file))
411 for (i = 0; i < w->var_cnt; i++)
413 struct pfm_var *v = &w->vars[i];
416 write_float (w, case_num_idx (c, v->fv));
419 write_int (w, v->width);
420 buf_write (w, case_str_idx (c, v->fv), v->width);
425 casewriter_force_error (writer);
431 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
433 struct pfm_writer *w = w_;
434 if (!close_writer (w))
435 casewriter_force_error (writer);
438 /* Closes a portable file after we're done with it.
439 Returns true if successful, false if an I/O error occurred. */
441 close_writer (struct pfm_writer *w)
452 memset (buf, 'Z', sizeof buf);
453 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
455 ok = !ferror (w->file);
456 if (fclose (w->file) == EOF)
460 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
461 fh_get_file_name (w->fh));
464 fh_close (w->fh, "portable file", "we");
472 /* Base-30 conversion.
474 Portable files represent numbers in base-30 format, so we need
475 to be able to convert real and integer number to that base.
476 Older versions of PSPP used libgmp to do so, but this added a
477 big library dependency to do just one thing. Now we do it
478 ourselves internally.
480 Important fact: base 30 is called "trigesimal". */
482 /* Conversion base. */
483 #define BASE 30 /* As an integer. */
484 #define LDBASE ((long double) BASE) /* As a long double. */
486 /* This is floor(log30(2**31)), the minimum number of trigesimal
487 digits that a `long int' can hold. */
490 /* pow_tab[i] = pow (30, pow (2, i)) */
491 static long double pow_tab[16];
493 /* Initializes pow_tab[]. */
497 static bool did_init = false;
501 /* Only initialize once. */
506 /* Set each element of pow_tab[] until we run out of numerical
509 for (power = 30.0L; power < DBL_MAX; power *= power)
511 assert (i < sizeof pow_tab / sizeof *pow_tab);
512 pow_tab[i++] = power;
516 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
518 pow30_nonnegative (int exponent)
523 assert (exponent >= 0);
524 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
527 for (i = 0; exponent > 0; exponent >>= 1, i++)
534 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
540 return pow30_nonnegative (exponent);
542 return 1.L / pow30_nonnegative (-exponent);
545 /* Returns the character corresponding to TRIG. */
547 trig_to_char (int trig)
549 assert (trig >= 0 && trig < 30);
550 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
553 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
554 null-terminated STRING. The trigesimal point is inserted
555 after TRIG_PLACES characters have been printed, if necessary
556 adding extra zeros at either end for correctness. Returns the
557 character after the formatted number. */
559 format_trig_digits (char *string,
560 const char trigs[], int trig_cnt, int trig_places)
565 while (trig_places++ < 0)
569 while (trig_cnt-- > 0)
571 if (trig_places-- == 0)
573 *string++ = trig_to_char (*trigs++);
575 while (trig_places-- > 0)
581 /* Helper function for format_trig_int() that formats VALUE as a
582 trigesimal integer at CP. VALUE must be nonnegative.
583 Returns the character following the formatted integer. */
585 recurse_format_trig_int (char *cp, int value)
587 int trig = value % BASE;
590 cp = recurse_format_trig_int (cp, value);
591 *cp++ = trig_to_char (trig);
595 /* Formats VALUE as a trigesimal integer in null-terminated
596 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
597 FORCE_SIGN is true, a sign is always inserted; otherwise, a
598 sign is only inserted if VALUE is negative. */
600 format_trig_int (int value, bool force_sign, char string[])
611 /* Format integer. */
612 string = recurse_format_trig_int (string, value);
617 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
618 rounding up or down. Returns true if TRIGS[] represents a
619 value greater than half, false if less than half. If TRIGS[]
620 is exactly half, examines TRIGS[-1] and returns true if odd,
621 false if even ("round to even"). */
623 should_round_up (const char trigs[], int trig_cnt)
625 assert (trig_cnt > 0);
627 if (*trigs < BASE / 2)
629 /* Less than half: round down. */
632 else if (*trigs > BASE / 2)
634 /* Greater than half: round up. */
639 /* Approximately half: look more closely. */
641 for (i = 1; i < trig_cnt; i++)
644 /* Slightly greater than half: round up. */
648 /* Exactly half: round to even. */
649 return trigs[-1] % 2;
653 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
654 carrying to the left as necessary. Returns true if
655 successful, false on failure (due to a carry out of the
656 leftmost position). */
658 try_round_up (char *trigs, int trig_cnt)
662 char *round_trig = trigs + --trig_cnt;
663 if (*round_trig != BASE - 1)
665 /* Round this trig up to the next value. */
670 /* Carry over to the next trig to the left. */
674 /* Ran out of trigs to carry. */
678 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
679 equivalent of at least BASE_10_PRECISION decimal digits of
680 precision. The output format may use conventional or
681 scientific notation. Missing, infinite, and extreme values
682 are represented with "*.". */
684 format_trig_double (long double value, int base_10_precision, char output[])
686 /* Original VALUE was negative? */
689 /* Number of significant trigesimals. */
690 int base_30_precision;
692 /* Base-2 significand and exponent for original VALUE. */
696 /* VALUE as a set of trigesimals. */
697 char buffer[DBL_DIG + 16];
701 /* Number of trigesimal places for trigs.
702 trigs[0] has coefficient 30**(trig_places - 1),
703 trigs[1] has coefficient 30**(trig_places - 2),
705 In other words, the trigesimal point is just before trigs[0].
709 /* Number of trigesimal places left to write into BUFFER. */
714 /* Handle special cases. */
720 /* Make VALUE positive. */
729 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
730 point left or right as necessary. We approximate the
731 base-30 exponent by obtaining the base-2 exponent, then
732 multiplying by log30(2). This approximation is sufficient
733 to ensure that the adjusted VALUE is always in the range
734 0...30**6, an invariant of the loop below. */
736 base_2_sig = frexp (value, &base_2_exp);
737 if (errno != 0 || !finite (base_2_sig))
739 if (base_2_exp == 0 && base_2_sig == 0.)
741 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
743 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
744 value *= pow30 (CHUNK_SIZE - trig_places);
746 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
749 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
751 trigs_to_output -= CHUNK_SIZE)
756 /* The current chunk is just the integer part of VALUE,
757 truncated to the nearest integer. The chunk fits in a
760 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
761 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
765 /* Append the chunk, in base 30, to trigs[]. */
766 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
768 trigs[trig_cnt + --trigs_left] = chunk % 30;
771 while (trigs_left > 0)
772 trigs[trig_cnt + --trigs_left] = 0;
773 trig_cnt += CHUNK_SIZE;
775 /* Proceed to the next chunk. */
778 value *= pow (LDBASE, CHUNK_SIZE);
781 /* Strip leading zeros. */
782 while (trig_cnt > 1 && *trigs == 0)
789 /* Round to requested precision, conservatively estimating the
790 required base-30 precision as 2/3 of the base-10 precision
791 (log30(10) = .68). */
792 assert (base_10_precision > 0);
793 if (base_10_precision > LDBL_DIG)
794 base_10_precision = LDBL_DIG;
795 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
796 if (trig_cnt > base_30_precision)
798 if (should_round_up (trigs + base_30_precision,
799 trig_cnt - base_30_precision))
801 /* Try to round up. */
802 if (try_round_up (trigs, base_30_precision))
804 /* Rounding up worked. */
805 trig_cnt = base_30_precision;
809 /* Couldn't round up because we ran out of trigs to
810 carry into. Do the carry here instead. */
819 trig_cnt = base_30_precision;
824 /* No rounding required: fewer digits available than
828 /* Strip trailing zeros. */
829 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
835 if (trig_places >= -1 && trig_places < trig_cnt + 3)
837 /* Use conventional notation. */
838 format_trig_digits (output, trigs, trig_cnt, trig_places);
842 /* Use scientific notation. */
844 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
845 op = format_trig_int (trig_places - trig_cnt, true, op);
850 strcpy (output, "0");
854 strcpy (output, "*.");
858 static struct casewriter_class por_file_casewriter_class =
860 por_file_casewriter_write,
861 por_file_casewriter_destroy,