1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2009 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/make-file.h>
37 #include <data/missing-values.h>
38 #include <data/short-names.h>
39 #include <data/value-labels.h>
40 #include <data/variable.h>
42 #include <libpspp/hash.h>
43 #include <libpspp/message.h>
44 #include <libpspp/misc.h>
45 #include <libpspp/str.h>
46 #include <libpspp/version.h>
52 #define _(msgid) gettext (msgid)
53 #define N_(msgid) (msgid)
55 /* Maximum width of a variable in a portable file. */
56 #define MAX_POR_WIDTH 255
58 /* Portable file writer. */
61 struct file_handle *fh; /* File handle. */
62 struct fh_lock *lock; /* Lock on file handle. */
63 FILE *file; /* File stream. */
64 struct replace_file *rf; /* Ticket for replacing output file. */
66 int lc; /* Number of characters on this line so far. */
68 size_t var_cnt; /* Number of variables. */
69 struct pfm_var *vars; /* Variables. */
71 int digits; /* Digits of precision. */
74 /* A variable to write to the portable file. */
77 int width; /* 0=numeric, otherwise string var width. */
78 int fv; /* Starting case index. */
81 static const struct casewriter_class por_file_casewriter_class;
83 static bool close_writer (struct pfm_writer *);
84 static void buf_write (struct pfm_writer *, const void *, size_t);
85 static void write_header (struct pfm_writer *);
86 static void write_version_data (struct pfm_writer *);
87 static void write_variables (struct pfm_writer *, struct dictionary *);
88 static void write_value_labels (struct pfm_writer *,
89 const struct dictionary *);
90 static void write_documents (struct pfm_writer *,
91 const struct dictionary *);
93 static void format_trig_double (long double, int base_10_precision, char[]);
94 static char *format_trig_int (int, bool force_sign, char[]);
96 /* Returns default options for writing a portable file. */
97 struct pfm_write_options
98 pfm_writer_default_options (void)
100 struct pfm_write_options opts;
101 opts.create_writeable = true;
102 opts.type = PFM_COMM;
103 opts.digits = DBL_DIG;
107 /* Writes the dictionary DICT to portable file HANDLE according
108 to the given OPTS. Returns nonzero only if successful. DICT
109 will not be modified, except to assign short names. */
111 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
112 struct pfm_write_options opts)
114 struct pfm_writer *w = NULL;
118 /* Initialize data structures. */
119 w = xmalloc (sizeof *w);
128 w->var_cnt = dict_get_var_cnt (dict);
129 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
130 for (i = 0; i < w->var_cnt; i++)
132 const struct variable *dv = dict_get_var (dict, i);
133 struct pfm_var *pv = &w->vars[i];
134 pv->width = MIN (var_get_width (dv), MAX_POR_WIDTH);
135 pv->fv = var_get_case_index (dv);
138 w->digits = opts.digits;
141 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
147 /* TRANSLATORS: this fragment will be interpolated into
148 messages in fh_lock() that identify types of files. */
149 w->lock = fh_lock (fh, FH_REF_FILE, N_("portable file"), FH_ACC_WRITE, true);
154 mode = S_IRUSR | S_IRGRP | S_IROTH;
155 if (opts.create_writeable)
156 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
157 w->rf = replace_file_start (fh_get_file_name (fh), "w", mode,
161 msg (ME, _("Error opening \"%s\" for writing as a portable file: %s."),
162 fh_get_file_name (fh), strerror (errno));
166 /* Write file header. */
168 write_version_data (w);
169 write_variables (w, dict);
170 write_value_labels (w, dict);
171 if (dict_get_document_line_cnt (dict) > 0)
172 write_documents (w, dict);
173 buf_write (w, "F", 1);
174 if (ferror (w->file))
176 return casewriter_create (dict_get_proto (dict),
177 &por_file_casewriter_class, w);
184 /* Write NBYTES starting at BUF to the portable file represented by
185 H. Break lines properly every 80 characters. */
187 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
189 const char *buf = buf_;
191 if (ferror (w->file))
194 assert (buf != NULL);
195 while (nbytes + w->lc >= 80)
197 size_t n = 80 - w->lc;
200 fwrite (buf, n, 1, w->file);
201 fwrite ("\r\n", 2, 1, w->file);
207 fwrite (buf, nbytes, 1, w->file);
212 /* Write D to the portable file as a floating-point field. */
214 write_float (struct pfm_writer *w, double d)
217 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
218 buf_write (w, buffer, strlen (buffer));
220 buf_write (w, "/", 1);
223 /* Write N to the portable file as an integer field. */
225 write_int (struct pfm_writer *w, int n)
228 format_trig_int (n, false, buffer);
229 buf_write (w, buffer, strlen (buffer));
230 buf_write (w, "/", 1);
233 /* Write S to the portable file as a string field. */
235 write_string (struct pfm_writer *w, const char *s)
237 size_t n = strlen (s);
238 write_int (w, (int) n);
242 /* Write file header. */
244 write_header (struct pfm_writer *w)
246 static const char spss2ascii[256] =
248 "0000000000000000000000000000000000000000000000000000000000000000"
249 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
250 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
251 "0000000000000000000000000000000000000000000000000000000000000000"
255 for (i = 0; i < 5; i++)
256 buf_write (w, "ASCII SPSS PORT FILE ", 40);
258 buf_write (w, spss2ascii, 256);
259 buf_write (w, "SPSSPORT", 8);
262 /* Writes version, date, and identification records. */
264 write_version_data (struct pfm_writer *w)
272 if ((time_t) -1 == time (&t))
274 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
279 tmp = localtime (&t);
281 sprintf (date_str, "%04d%02d%02d",
282 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
283 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
284 buf_write (w, "A", 1);
285 write_string (w, date_str);
286 write_string (w, time_str);
288 /* Product identification. */
289 buf_write (w, "1", 1);
290 write_string (w, version);
292 /* Subproduct identification. */
293 buf_write (w, "3", 1);
294 write_string (w, host_system);
297 /* Write format F to file H. The format is first resized to fit
298 a value of the given WIDTH, which is handy in case F
299 represents a string longer than 255 bytes and thus WIDTH is
300 truncated to 255 bytes. */
302 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
304 fmt_resize (&f, width);
305 write_int (w, fmt_to_io (f.type));
310 /* Write value V with width WIDTH to file H. */
312 write_value (struct pfm_writer *w, const union value *v, int width)
315 write_float (w, v->f);
318 width = MIN (width, MAX_POR_WIDTH);
319 write_int (w, width);
320 buf_write (w, value_str (v, width), width);
324 /* Write variable records. */
326 write_variables (struct pfm_writer *w, struct dictionary *dict)
330 short_names_assign (dict);
332 if (dict_get_weight (dict) != NULL)
334 buf_write (w, "6", 1);
335 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
338 buf_write (w, "4", 1);
339 write_int (w, dict_get_var_cnt (dict));
342 for (i = 0; i < dict_get_var_cnt (dict); i++)
344 struct variable *v = dict_get_var (dict, i);
345 struct missing_values mv;
346 int width = MIN (var_get_width (v), MAX_POR_WIDTH);
349 buf_write (w, "7", 1);
350 write_int (w, width);
351 write_string (w, var_get_short_name (v, 0));
352 write_format (w, *var_get_print_format (v), width);
353 write_format (w, *var_get_write_format (v), width);
355 /* Write missing values. */
356 mv_copy (&mv, var_get_missing_values (v));
357 if (var_get_width (v) > 8)
359 if (mv_has_range (&mv))
362 mv_get_range (&mv, &x, &y);
365 buf_write (w, "9", 1);
368 else if (y == HIGHEST)
370 buf_write (w, "A", 1);
375 buf_write (w, "B", 1);
380 for (j = 0; j < mv_n_values (&mv); j++)
382 buf_write (w, "8", 1);
383 write_value (w, mv_get_value (&mv, j), mv_get_width (&mv));
387 /* Write variable label. */
388 if (var_get_label (v) != NULL)
390 buf_write (w, "C", 1);
391 write_string (w, var_get_label (v));
396 /* Write value labels to disk. FIXME: Inefficient. */
398 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
402 for (i = 0; i < dict_get_var_cnt (dict); i++)
404 struct variable *v = dict_get_var (dict, i);
405 const struct val_labs *val_labs = var_get_value_labels (v);
406 size_t n_labels = val_labs_count (val_labs);
407 const struct val_lab **labels;
412 buf_write (w, "D", 1);
414 write_string (w, var_get_short_name (v, 0));
415 write_int (w, val_labs_count (val_labs));
417 n_labels = val_labs_count (val_labs);
418 labels = val_labs_sorted (val_labs);
419 for (i = 0; i < n_labels; i++)
421 const struct val_lab *vl = labels[i];
422 write_value (w, val_lab_get_value (vl), var_get_width (v));
423 write_string (w, val_lab_get_label (vl));
429 /* Write documents in DICT to portable file W. */
431 write_documents (struct pfm_writer *w, const struct dictionary *dict)
433 size_t line_cnt = dict_get_document_line_cnt (dict);
434 struct string line = DS_EMPTY_INITIALIZER;
437 buf_write (w, "E", 1);
438 write_int (w, line_cnt);
439 for (i = 0; i < line_cnt; i++)
441 dict_get_document_line (dict, i, &line);
442 write_string (w, ds_cstr (&line));
447 /* Writes case C to the portable file represented by WRITER. */
449 por_file_casewriter_write (struct casewriter *writer, void *w_,
452 struct pfm_writer *w = w_;
455 if (!ferror (w->file))
457 for (i = 0; i < w->var_cnt; i++)
459 struct pfm_var *v = &w->vars[i];
462 write_float (w, case_num_idx (c, v->fv));
465 write_int (w, v->width);
466 buf_write (w, case_str_idx (c, v->fv), v->width);
471 casewriter_force_error (writer);
477 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
479 struct pfm_writer *w = w_;
480 if (!close_writer (w))
481 casewriter_force_error (writer);
484 /* Closes a portable file after we're done with it.
485 Returns true if successful, false if an I/O error occurred. */
487 close_writer (struct pfm_writer *w)
498 memset (buf, 'Z', sizeof buf);
499 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
501 ok = !ferror (w->file);
502 if (fclose (w->file) == EOF)
506 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
507 fh_get_file_name (w->fh));
509 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
522 /* Base-30 conversion.
524 Portable files represent numbers in base-30 format, so we need
525 to be able to convert real and integer number to that base.
526 Older versions of PSPP used libgmp to do so, but this added a
527 big library dependency to do just one thing. Now we do it
528 ourselves internally.
530 Important fact: base 30 is called "trigesimal". */
532 /* Conversion base. */
533 #define BASE 30 /* As an integer. */
534 #define LDBASE ((long double) BASE) /* As a long double. */
536 /* This is floor(log30(2**31)), the minimum number of trigesimal
537 digits that a `long int' can hold. */
540 /* pow_tab[i] = pow (30, pow (2, i)) */
541 static long double pow_tab[16];
543 /* Initializes pow_tab[]. */
547 static bool did_init = false;
551 /* Only initialize once. */
556 /* Set each element of pow_tab[] until we run out of numerical
559 for (power = 30.0L; power < DBL_MAX; power *= power)
561 assert (i < sizeof pow_tab / sizeof *pow_tab);
562 pow_tab[i++] = power;
566 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
568 pow30_nonnegative (int exponent)
573 assert (exponent >= 0);
574 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
577 for (i = 0; exponent > 0; exponent >>= 1, i++)
584 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
590 return pow30_nonnegative (exponent);
592 return 1.L / pow30_nonnegative (-exponent);
595 /* Returns the character corresponding to TRIG. */
597 trig_to_char (int trig)
599 assert (trig >= 0 && trig < 30);
600 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
603 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
604 null-terminated STRING. The trigesimal point is inserted
605 after TRIG_PLACES characters have been printed, if necessary
606 adding extra zeros at either end for correctness. Returns the
607 character after the formatted number. */
609 format_trig_digits (char *string,
610 const char trigs[], int trig_cnt, int trig_places)
615 while (trig_places++ < 0)
619 while (trig_cnt-- > 0)
621 if (trig_places-- == 0)
623 *string++ = trig_to_char (*trigs++);
625 while (trig_places-- > 0)
631 /* Helper function for format_trig_int() that formats VALUE as a
632 trigesimal integer at CP. VALUE must be nonnegative.
633 Returns the character following the formatted integer. */
635 recurse_format_trig_int (char *cp, int value)
637 int trig = value % BASE;
640 cp = recurse_format_trig_int (cp, value);
641 *cp++ = trig_to_char (trig);
645 /* Formats VALUE as a trigesimal integer in null-terminated
646 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
647 FORCE_SIGN is true, a sign is always inserted; otherwise, a
648 sign is only inserted if VALUE is negative. */
650 format_trig_int (int value, bool force_sign, char string[])
661 /* Format integer. */
662 string = recurse_format_trig_int (string, value);
667 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
668 rounding up or down. Returns true if TRIGS[] represents a
669 value greater than half, false if less than half. If TRIGS[]
670 is exactly half, examines TRIGS[-1] and returns true if odd,
671 false if even ("round to even"). */
673 should_round_up (const char trigs[], int trig_cnt)
675 assert (trig_cnt > 0);
677 if (*trigs < BASE / 2)
679 /* Less than half: round down. */
682 else if (*trigs > BASE / 2)
684 /* Greater than half: round up. */
689 /* Approximately half: look more closely. */
691 for (i = 1; i < trig_cnt; i++)
694 /* Slightly greater than half: round up. */
698 /* Exactly half: round to even. */
699 return trigs[-1] % 2;
703 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
704 carrying to the left as necessary. Returns true if
705 successful, false on failure (due to a carry out of the
706 leftmost position). */
708 try_round_up (char *trigs, int trig_cnt)
712 char *round_trig = trigs + --trig_cnt;
713 if (*round_trig != BASE - 1)
715 /* Round this trig up to the next value. */
720 /* Carry over to the next trig to the left. */
724 /* Ran out of trigs to carry. */
728 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
729 equivalent of at least BASE_10_PRECISION decimal digits of
730 precision. The output format may use conventional or
731 scientific notation. Missing, infinite, and extreme values
732 are represented with "*.". */
734 format_trig_double (long double value, int base_10_precision, char output[])
736 /* Original VALUE was negative? */
739 /* Number of significant trigesimals. */
740 int base_30_precision;
742 /* Base-2 significand and exponent for original VALUE. */
746 /* VALUE as a set of trigesimals. */
747 char buffer[DBL_DIG + 16];
751 /* Number of trigesimal places for trigs.
752 trigs[0] has coefficient 30**(trig_places - 1),
753 trigs[1] has coefficient 30**(trig_places - 2),
755 In other words, the trigesimal point is just before trigs[0].
759 /* Number of trigesimal places left to write into BUFFER. */
764 /* Handle special cases. */
770 /* Make VALUE positive. */
779 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
780 point left or right as necessary. We approximate the
781 base-30 exponent by obtaining the base-2 exponent, then
782 multiplying by log30(2). This approximation is sufficient
783 to ensure that the adjusted VALUE is always in the range
784 0...30**6, an invariant of the loop below. */
786 base_2_sig = frexp (value, &base_2_exp);
787 if (errno != 0 || !isfinite (base_2_sig))
789 if (base_2_exp == 0 && base_2_sig == 0.)
791 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
793 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
794 value *= pow30 (CHUNK_SIZE - trig_places);
796 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
799 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
801 trigs_to_output -= CHUNK_SIZE)
806 /* The current chunk is just the integer part of VALUE,
807 truncated to the nearest integer. The chunk fits in a
810 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
811 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
815 /* Append the chunk, in base 30, to trigs[]. */
816 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
818 trigs[trig_cnt + --trigs_left] = chunk % 30;
821 while (trigs_left > 0)
822 trigs[trig_cnt + --trigs_left] = 0;
823 trig_cnt += CHUNK_SIZE;
825 /* Proceed to the next chunk. */
828 value *= pow (LDBASE, CHUNK_SIZE);
831 /* Strip leading zeros. */
832 while (trig_cnt > 1 && *trigs == 0)
839 /* Round to requested precision, conservatively estimating the
840 required base-30 precision as 2/3 of the base-10 precision
841 (log30(10) = .68). */
842 assert (base_10_precision > 0);
843 if (base_10_precision > LDBL_DIG)
844 base_10_precision = LDBL_DIG;
845 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
846 if (trig_cnt > base_30_precision)
848 if (should_round_up (trigs + base_30_precision,
849 trig_cnt - base_30_precision))
851 /* Try to round up. */
852 if (try_round_up (trigs, base_30_precision))
854 /* Rounding up worked. */
855 trig_cnt = base_30_precision;
859 /* Couldn't round up because we ran out of trigs to
860 carry into. Do the carry here instead. */
869 trig_cnt = base_30_precision;
874 /* No rounding required: fewer digits available than
878 /* Strip trailing zeros. */
879 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
885 if (trig_places >= -1 && trig_places < trig_cnt + 3)
887 /* Use conventional notation. */
888 format_trig_digits (output, trigs, trig_cnt, trig_places);
892 /* Use scientific notation. */
894 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
895 op = format_trig_int (trig_places - trig_cnt, true, op);
900 strcpy (output, "0");
904 strcpy (output, "*.");
908 static const struct casewriter_class por_file_casewriter_class =
910 por_file_casewriter_write,
911 por_file_casewriter_destroy,