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"
29 #include <data/case.h>
30 #include <data/casewriter-provider.h>
31 #include <data/casewriter.h>
32 #include <data/dictionary.h>
33 #include <data/file-handle-def.h>
34 #include <data/file-name.h>
35 #include <data/format.h>
36 #include <data/missing-values.h>
37 #include <data/short-names.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 /* Maximum width of a variable in a portable file. */
53 #define MAX_POR_WIDTH 255
55 /* Portable file writer. */
58 struct file_handle *fh; /* File handle. */
59 FILE *file; /* File stream. */
61 int lc; /* Number of characters on this line so far. */
63 size_t var_cnt; /* Number of variables. */
64 struct pfm_var *vars; /* Variables. */
66 int digits; /* Digits of precision. */
69 /* A variable to write to the portable file. */
72 int width; /* 0=numeric, otherwise string var width. */
73 int fv; /* Starting case index. */
76 static struct casewriter_class por_file_casewriter_class;
78 static bool close_writer (struct pfm_writer *);
79 static void buf_write (struct pfm_writer *, const void *, size_t);
80 static void write_header (struct pfm_writer *);
81 static void write_version_data (struct pfm_writer *);
82 static void write_variables (struct pfm_writer *, struct dictionary *);
83 static void write_value_labels (struct pfm_writer *,
84 const struct dictionary *);
85 static void write_documents (struct pfm_writer *,
86 const struct dictionary *);
88 static void format_trig_double (long double, int base_10_precision, char[]);
89 static char *format_trig_int (int, bool force_sign, char[]);
91 /* Returns default options for writing a portable file. */
92 struct pfm_write_options
93 pfm_writer_default_options (void)
95 struct pfm_write_options opts;
96 opts.create_writeable = true;
98 opts.digits = DBL_DIG;
102 /* Writes the dictionary DICT to portable file HANDLE according
103 to the given OPTS. Returns nonzero only if successful. DICT
104 will not be modified, except to assign short names. */
106 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
107 struct pfm_write_options opts)
109 struct pfm_writer *w = NULL;
114 /* Open file handle. */
115 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
119 mode = S_IRUSR | S_IRGRP | S_IROTH;
120 if (opts.create_writeable)
121 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
122 file = create_stream (fh_get_file_name (fh), "w", mode);
125 fh_close (fh, "portable file", "we");
126 msg (ME, _("An error occurred while opening \"%s\" for writing "
127 "as a portable file: %s."),
128 fh_get_file_name (fh), strerror (errno));
132 /* Initialize data structures. */
133 w = xmalloc (sizeof *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))
172 return casewriter_create (dict_get_next_value_idx (dict),
173 &por_file_casewriter_class, w);
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. The format is first resized to fit
289 a value of the given WIDTH, which is handy in case F
290 represents a string longer than 255 bytes and thus WIDTH is
291 truncated to 255 bytes. */
293 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
295 fmt_resize (&f, width);
296 write_int (w, fmt_to_io (f.type));
301 /* Write value V for variable VV to file H. */
303 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
305 if (var_is_numeric (vv))
306 write_float (w, v->f);
309 int width = MIN (var_get_width (vv), MAX_POR_WIDTH);
310 write_int (w, width);
311 buf_write (w, v->s, width);
315 /* Write variable records. */
317 write_variables (struct pfm_writer *w, struct dictionary *dict)
321 short_names_assign (dict);
323 if (dict_get_weight (dict) != NULL)
325 buf_write (w, "6", 1);
326 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
329 buf_write (w, "4", 1);
330 write_int (w, dict_get_var_cnt (dict));
333 for (i = 0; i < dict_get_var_cnt (dict); i++)
335 struct variable *v = dict_get_var (dict, i);
336 struct missing_values mv;
337 int width = MIN (var_get_width (v), MAX_POR_WIDTH);
339 buf_write (w, "7", 1);
340 write_int (w, width);
341 write_string (w, var_get_short_name (v, 0));
342 write_format (w, *var_get_print_format (v), width);
343 write_format (w, *var_get_write_format (v), width);
345 /* Write missing values. */
346 mv_copy (&mv, var_get_missing_values (v));
347 while (mv_has_range (&mv))
350 mv_pop_range (&mv, &x, &y);
353 buf_write (w, "9", 1);
356 else if (y == HIGHEST)
358 buf_write (w, "A", 1);
363 buf_write (w, "B", 1);
368 while (mv_has_value (&mv))
371 mv_pop_value (&mv, &value);
372 buf_write (w, "8", 1);
373 write_value (w, &value, v);
376 /* Write variable label. */
377 if (var_get_label (v) != NULL)
379 buf_write (w, "C", 1);
380 write_string (w, var_get_label (v));
385 /* Write value labels to disk. FIXME: Inefficient. */
387 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
391 for (i = 0; i < dict_get_var_cnt (dict); i++)
393 struct val_labs_iterator *j;
394 struct variable *v = dict_get_var (dict, i);
395 const struct val_labs *val_labs = var_get_value_labels (v);
398 if (val_labs == NULL)
401 buf_write (w, "D", 1);
403 write_string (w, var_get_short_name (v, 0));
404 write_int (w, val_labs_count (val_labs));
406 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
407 vl = val_labs_next (val_labs, &j))
409 write_value (w, &vl->value, v);
410 write_string (w, vl->label);
415 /* Write documents in DICT to portable file W. */
417 write_documents (struct pfm_writer *w, const struct dictionary *dict)
419 size_t line_cnt = dict_get_document_line_cnt (dict);
420 struct string line = DS_EMPTY_INITIALIZER;
423 buf_write (w, "E", 1);
424 write_int (w, line_cnt);
425 for (i = 0; i < line_cnt; i++)
427 dict_get_document_line (dict, i, &line);
428 write_string (w, ds_cstr (&line));
433 /* Writes case C to the portable file represented by WRITER. */
435 por_file_casewriter_write (struct casewriter *writer, void *w_,
438 struct pfm_writer *w = w_;
441 if (!ferror (w->file))
443 for (i = 0; i < w->var_cnt; i++)
445 struct pfm_var *v = &w->vars[i];
448 write_float (w, case_num_idx (c, v->fv));
451 write_int (w, v->width);
452 buf_write (w, case_str_idx (c, v->fv), v->width);
457 casewriter_force_error (writer);
463 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
465 struct pfm_writer *w = w_;
466 if (!close_writer (w))
467 casewriter_force_error (writer);
470 /* Closes a portable file after we're done with it.
471 Returns true if successful, false if an I/O error occurred. */
473 close_writer (struct pfm_writer *w)
484 memset (buf, 'Z', sizeof buf);
485 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
487 ok = !ferror (w->file);
488 if (fclose (w->file) == EOF)
492 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
493 fh_get_file_name (w->fh));
496 fh_close (w->fh, "portable file", "we");
504 /* Base-30 conversion.
506 Portable files represent numbers in base-30 format, so we need
507 to be able to convert real and integer number to that base.
508 Older versions of PSPP used libgmp to do so, but this added a
509 big library dependency to do just one thing. Now we do it
510 ourselves internally.
512 Important fact: base 30 is called "trigesimal". */
514 /* Conversion base. */
515 #define BASE 30 /* As an integer. */
516 #define LDBASE ((long double) BASE) /* As a long double. */
518 /* This is floor(log30(2**31)), the minimum number of trigesimal
519 digits that a `long int' can hold. */
522 /* pow_tab[i] = pow (30, pow (2, i)) */
523 static long double pow_tab[16];
525 /* Initializes pow_tab[]. */
529 static bool did_init = false;
533 /* Only initialize once. */
538 /* Set each element of pow_tab[] until we run out of numerical
541 for (power = 30.0L; power < DBL_MAX; power *= power)
543 assert (i < sizeof pow_tab / sizeof *pow_tab);
544 pow_tab[i++] = power;
548 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
550 pow30_nonnegative (int exponent)
555 assert (exponent >= 0);
556 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
559 for (i = 0; exponent > 0; exponent >>= 1, i++)
566 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
572 return pow30_nonnegative (exponent);
574 return 1.L / pow30_nonnegative (-exponent);
577 /* Returns the character corresponding to TRIG. */
579 trig_to_char (int trig)
581 assert (trig >= 0 && trig < 30);
582 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
585 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
586 null-terminated STRING. The trigesimal point is inserted
587 after TRIG_PLACES characters have been printed, if necessary
588 adding extra zeros at either end for correctness. Returns the
589 character after the formatted number. */
591 format_trig_digits (char *string,
592 const char trigs[], int trig_cnt, int trig_places)
597 while (trig_places++ < 0)
601 while (trig_cnt-- > 0)
603 if (trig_places-- == 0)
605 *string++ = trig_to_char (*trigs++);
607 while (trig_places-- > 0)
613 /* Helper function for format_trig_int() that formats VALUE as a
614 trigesimal integer at CP. VALUE must be nonnegative.
615 Returns the character following the formatted integer. */
617 recurse_format_trig_int (char *cp, int value)
619 int trig = value % BASE;
622 cp = recurse_format_trig_int (cp, value);
623 *cp++ = trig_to_char (trig);
627 /* Formats VALUE as a trigesimal integer in null-terminated
628 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
629 FORCE_SIGN is true, a sign is always inserted; otherwise, a
630 sign is only inserted if VALUE is negative. */
632 format_trig_int (int value, bool force_sign, char string[])
643 /* Format integer. */
644 string = recurse_format_trig_int (string, value);
649 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
650 rounding up or down. Returns true if TRIGS[] represents a
651 value greater than half, false if less than half. If TRIGS[]
652 is exactly half, examines TRIGS[-1] and returns true if odd,
653 false if even ("round to even"). */
655 should_round_up (const char trigs[], int trig_cnt)
657 assert (trig_cnt > 0);
659 if (*trigs < BASE / 2)
661 /* Less than half: round down. */
664 else if (*trigs > BASE / 2)
666 /* Greater than half: round up. */
671 /* Approximately half: look more closely. */
673 for (i = 1; i < trig_cnt; i++)
676 /* Slightly greater than half: round up. */
680 /* Exactly half: round to even. */
681 return trigs[-1] % 2;
685 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
686 carrying to the left as necessary. Returns true if
687 successful, false on failure (due to a carry out of the
688 leftmost position). */
690 try_round_up (char *trigs, int trig_cnt)
694 char *round_trig = trigs + --trig_cnt;
695 if (*round_trig != BASE - 1)
697 /* Round this trig up to the next value. */
702 /* Carry over to the next trig to the left. */
706 /* Ran out of trigs to carry. */
710 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
711 equivalent of at least BASE_10_PRECISION decimal digits of
712 precision. The output format may use conventional or
713 scientific notation. Missing, infinite, and extreme values
714 are represented with "*.". */
716 format_trig_double (long double value, int base_10_precision, char output[])
718 /* Original VALUE was negative? */
721 /* Number of significant trigesimals. */
722 int base_30_precision;
724 /* Base-2 significand and exponent for original VALUE. */
728 /* VALUE as a set of trigesimals. */
729 char buffer[DBL_DIG + 16];
733 /* Number of trigesimal places for trigs.
734 trigs[0] has coefficient 30**(trig_places - 1),
735 trigs[1] has coefficient 30**(trig_places - 2),
737 In other words, the trigesimal point is just before trigs[0].
741 /* Number of trigesimal places left to write into BUFFER. */
746 /* Handle special cases. */
752 /* Make VALUE positive. */
761 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
762 point left or right as necessary. We approximate the
763 base-30 exponent by obtaining the base-2 exponent, then
764 multiplying by log30(2). This approximation is sufficient
765 to ensure that the adjusted VALUE is always in the range
766 0...30**6, an invariant of the loop below. */
768 base_2_sig = frexp (value, &base_2_exp);
769 if (errno != 0 || !finite (base_2_sig))
771 if (base_2_exp == 0 && base_2_sig == 0.)
773 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
775 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
776 value *= pow30 (CHUNK_SIZE - trig_places);
778 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
781 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
783 trigs_to_output -= CHUNK_SIZE)
788 /* The current chunk is just the integer part of VALUE,
789 truncated to the nearest integer. The chunk fits in a
792 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
793 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
797 /* Append the chunk, in base 30, to trigs[]. */
798 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
800 trigs[trig_cnt + --trigs_left] = chunk % 30;
803 while (trigs_left > 0)
804 trigs[trig_cnt + --trigs_left] = 0;
805 trig_cnt += CHUNK_SIZE;
807 /* Proceed to the next chunk. */
810 value *= pow (LDBASE, CHUNK_SIZE);
813 /* Strip leading zeros. */
814 while (trig_cnt > 1 && *trigs == 0)
821 /* Round to requested precision, conservatively estimating the
822 required base-30 precision as 2/3 of the base-10 precision
823 (log30(10) = .68). */
824 assert (base_10_precision > 0);
825 if (base_10_precision > LDBL_DIG)
826 base_10_precision = LDBL_DIG;
827 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
828 if (trig_cnt > base_30_precision)
830 if (should_round_up (trigs + base_30_precision,
831 trig_cnt - base_30_precision))
833 /* Try to round up. */
834 if (try_round_up (trigs, base_30_precision))
836 /* Rounding up worked. */
837 trig_cnt = base_30_precision;
841 /* Couldn't round up because we ran out of trigs to
842 carry into. Do the carry here instead. */
851 trig_cnt = base_30_precision;
856 /* No rounding required: fewer digits available than
860 /* Strip trailing zeros. */
861 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
867 if (trig_places >= -1 && trig_places < trig_cnt + 3)
869 /* Use conventional notation. */
870 format_trig_digits (output, trigs, trig_cnt, trig_places);
874 /* Use scientific notation. */
876 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
877 op = format_trig_int (trig_places - trig_cnt, true, op);
882 strcpy (output, "0");
886 strcpy (output, "*.");
890 static struct casewriter_class por_file_casewriter_class =
892 por_file_casewriter_write,
893 por_file_casewriter_destroy,