1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011, 2012, 2013 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/>. */
19 #include "data/por-file-writer.h"
30 #include "data/case.h"
31 #include "data/casewriter-provider.h"
32 #include "data/casewriter.h"
33 #include "data/dictionary.h"
34 #include "data/file-handle-def.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"
41 #include "libpspp/message.h"
42 #include "libpspp/misc.h"
43 #include "libpspp/str.h"
44 #include "libpspp/version.h"
46 #include "gl/minmax.h"
47 #include "gl/xalloc.h"
50 #define _(msgid) gettext (msgid)
51 #define N_(msgid) (msgid)
53 /* Maximum width of a variable in a portable file. */
54 #define MAX_POR_WIDTH 255
56 /* Portable file writer. */
59 struct file_handle *fh; /* File handle. */
60 struct fh_lock *lock; /* Lock on file handle. */
61 FILE *file; /* File stream. */
62 struct replace_file *rf; /* Ticket for replacing output file. */
64 int lc; /* Number of characters on this line so far. */
66 size_t var_cnt; /* Number of variables. */
67 struct pfm_var *vars; /* Variables. */
69 int digits; /* Digits of precision. */
72 /* A variable to write to the portable file. */
75 int width; /* 0=numeric, otherwise string var width. */
76 int case_index; /* Index in case. */
79 static const struct casewriter_class por_file_casewriter_class;
81 static bool close_writer (struct pfm_writer *);
82 static void buf_write (struct pfm_writer *, const void *, size_t);
83 static void write_header (struct pfm_writer *);
84 static void write_version_data (struct pfm_writer *);
85 static void write_variables (struct pfm_writer *, struct dictionary *);
86 static void write_value_labels (struct pfm_writer *,
87 const struct dictionary *);
88 static void write_documents (struct pfm_writer *,
89 const struct dictionary *);
91 static void format_trig_double (long double, int base_10_precision, char[]);
92 static char *format_trig_int (int, bool force_sign, char[]);
94 /* Returns default options for writing a portable file. */
95 struct pfm_write_options
96 pfm_writer_default_options (void)
98 struct pfm_write_options opts;
99 opts.create_writeable = true;
100 opts.type = PFM_COMM;
101 opts.digits = DBL_DIG;
105 /* Writes the dictionary DICT to portable file HANDLE according
106 to the given OPTS. Returns nonzero only if successful. DICT
107 will not be modified, except to assign short names. */
109 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
110 struct pfm_write_options opts)
112 struct pfm_writer *w = NULL;
116 /* Initialize data structures. */
117 w = xmalloc (sizeof *w);
126 w->var_cnt = dict_get_var_cnt (dict);
127 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
128 for (i = 0; i < w->var_cnt; i++)
130 const struct variable *dv = dict_get_var (dict, i);
131 struct pfm_var *pv = &w->vars[i];
132 pv->width = MIN (var_get_width (dv), MAX_POR_WIDTH);
133 pv->case_index = var_get_case_index (dv);
136 w->digits = opts.digits;
139 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
145 /* TRANSLATORS: this fragment will be interpolated into
146 messages in fh_lock() that identify types of files. */
147 w->lock = fh_lock (fh, FH_REF_FILE, N_("portable file"), FH_ACC_WRITE, true);
153 if (opts.create_writeable)
155 w->rf = replace_file_start (fh, "w", mode,
159 msg (ME, _("Error opening `%s' for writing as a portable file: %s."),
160 fh_get_file_name (fh), strerror (errno));
164 /* Write file header. */
166 write_version_data (w);
167 write_variables (w, dict);
168 write_value_labels (w, dict);
169 if (dict_get_document_line_cnt (dict) > 0)
170 write_documents (w, dict);
171 buf_write (w, "F", 1);
172 if (ferror (w->file))
174 return casewriter_create (dict_get_proto (dict),
175 &por_file_casewriter_class, w);
182 /* Write NBYTES starting at BUF to the portable file represented by
183 H. Break lines properly every 80 characters. */
185 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
187 const char *buf = buf_;
189 if (ferror (w->file))
192 assert (buf != NULL);
193 while (nbytes + w->lc >= 80)
195 size_t n = 80 - w->lc;
198 fwrite (buf, n, 1, w->file);
199 fwrite ("\r\n", 2, 1, w->file);
205 fwrite (buf, nbytes, 1, w->file);
210 /* Write D to the portable file as a floating-point field. */
212 write_float (struct pfm_writer *w, double d)
215 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
216 buf_write (w, buffer, strlen (buffer));
218 buf_write (w, "/", 1);
221 /* Write N to the portable file as an integer field. */
223 write_int (struct pfm_writer *w, int n)
226 format_trig_int (n, false, buffer);
227 buf_write (w, buffer, strlen (buffer));
228 buf_write (w, "/", 1);
231 /* Write S to the portable file as a string field. */
233 write_string (struct pfm_writer *w, const char *s)
235 size_t n = strlen (s);
236 write_int (w, (int) n);
240 /* Write file header. */
242 write_header (struct pfm_writer *w)
244 static const char spss2ascii[256] =
246 "0000000000000000000000000000000000000000000000000000000000000000"
247 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
248 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
249 "0000000000000000000000000000000000000000000000000000000000000000"
253 for (i = 0; i < 5; i++)
254 buf_write (w, "ASCII SPSS PORT FILE ", 40);
256 buf_write (w, spss2ascii, 256);
257 buf_write (w, "SPSSPORT", 8);
260 /* Writes version, date, and identification records. */
262 write_version_data (struct pfm_writer *w)
270 if ((time_t) -1 == time (&t))
272 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
277 tmp = localtime (&t);
279 sprintf (date_str, "%04d%02d%02d",
280 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
281 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
282 buf_write (w, "A", 1);
283 write_string (w, date_str);
284 write_string (w, time_str);
286 /* Product identification. */
287 buf_write (w, "1", 1);
288 write_string (w, version);
290 /* Subproduct identification. */
291 buf_write (w, "3", 1);
292 write_string (w, host_system);
295 /* Write format F to file H. The format is first resized to fit
296 a value of the given WIDTH, which is handy in case F
297 represents a string longer than 255 bytes and thus WIDTH is
298 truncated to 255 bytes. */
300 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
302 fmt_resize (&f, width);
303 write_int (w, fmt_to_io (f.type));
308 /* Write value V with width WIDTH to file H. */
310 write_value (struct pfm_writer *w, const union value *v, int width)
313 write_float (w, v->f);
316 width = MIN (width, MAX_POR_WIDTH);
317 write_int (w, width);
318 buf_write (w, value_str (v, width), width);
322 /* Write variable records. */
324 write_variables (struct pfm_writer *w, struct dictionary *dict)
328 short_names_assign (dict);
330 if (dict_get_weight (dict) != NULL)
332 buf_write (w, "6", 1);
333 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
336 buf_write (w, "4", 1);
337 write_int (w, dict_get_var_cnt (dict));
339 buf_write (w, "5", 1);
340 write_int (w, ceil (w->digits * (log (10) / log (30))));
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;
413 buf_write (w, "D", 1);
415 write_string (w, var_get_short_name (v, 0));
416 write_int (w, val_labs_count (val_labs));
418 n_labels = val_labs_count (val_labs);
419 labels = val_labs_sorted (val_labs);
420 for (j = 0; j < n_labels; j++)
422 const struct val_lab *vl = labels[j];
423 write_value (w, val_lab_get_value (vl), var_get_width (v));
424 write_string (w, val_lab_get_escaped_label (vl));
430 /* Write documents in DICT to portable file W. */
432 write_documents (struct pfm_writer *w, const struct dictionary *dict)
434 size_t line_cnt = dict_get_document_line_cnt (dict);
435 struct string line = DS_EMPTY_INITIALIZER;
438 buf_write (w, "E", 1);
439 write_int (w, line_cnt);
440 for (i = 0; i < line_cnt; i++)
441 write_string (w, dict_get_document_line (dict, i));
445 /* Writes case C to the portable file represented by WRITER. */
447 por_file_casewriter_write (struct casewriter *writer, void *w_,
450 struct pfm_writer *w = w_;
453 if (!ferror (w->file))
455 for (i = 0; i < w->var_cnt; i++)
457 struct pfm_var *v = &w->vars[i];
460 write_float (w, case_num_idx (c, v->case_index));
463 write_int (w, v->width);
464 buf_write (w, case_str_idx (c, v->case_index), v->width);
469 casewriter_force_error (writer);
475 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
477 struct pfm_writer *w = w_;
478 if (!close_writer (w))
479 casewriter_force_error (writer);
482 /* Closes a portable file after we're done with it.
483 Returns true if successful, false if an I/O error occurred. */
485 close_writer (struct pfm_writer *w)
496 memset (buf, 'Z', sizeof buf);
497 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
499 ok = !ferror (w->file);
500 if (fclose (w->file) == EOF)
504 msg (ME, _("An I/O error occurred writing portable file `%s'."),
505 fh_get_file_name (w->fh));
507 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
520 /* Base-30 conversion.
522 Portable files represent numbers in base-30 format, so we need
523 to be able to convert real and integer number to that base.
524 Older versions of PSPP used libgmp to do so, but this added a
525 big library dependency to do just one thing. Now we do it
526 ourselves internally.
528 Important fact: base 30 is called "trigesimal". */
530 /* Conversion base. */
531 #define BASE 30 /* As an integer. */
532 #define LDBASE ((long double) BASE) /* As a long double. */
534 /* This is floor(log30(2**31)), the minimum number of trigesimal
535 digits that a `long int' can hold. */
538 /* pow_tab[i] = pow (30, pow (2, i)) */
539 static long double pow_tab[16];
541 /* Initializes pow_tab[]. */
545 static bool did_init = false;
549 /* Only initialize once. */
554 /* Set each element of pow_tab[] until we run out of numerical
557 for (power = 30.0L; power < DBL_MAX; power *= power)
559 assert (i < sizeof pow_tab / sizeof *pow_tab);
560 pow_tab[i++] = power;
564 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
566 pow30_nonnegative (int exponent)
571 assert (exponent >= 0);
572 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
575 for (i = 0; exponent > 0; exponent >>= 1, i++)
582 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
588 return pow30_nonnegative (exponent);
590 return 1.L / pow30_nonnegative (-exponent);
593 /* Returns the character corresponding to TRIG. */
595 trig_to_char (int trig)
597 assert (trig >= 0 && trig < 30);
598 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
601 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
602 null-terminated STRING. The trigesimal point is inserted
603 after TRIG_PLACES characters have been printed, if necessary
604 adding extra zeros at either end for correctness. Returns the
605 character after the formatted number. */
607 format_trig_digits (char *string,
608 const char trigs[], int trig_cnt, int trig_places)
613 while (trig_places++ < 0)
617 while (trig_cnt-- > 0)
619 if (trig_places-- == 0)
621 *string++ = trig_to_char (*trigs++);
623 while (trig_places-- > 0)
629 /* Helper function for format_trig_int() that formats VALUE as a
630 trigesimal integer at CP. VALUE must be nonnegative.
631 Returns the character following the formatted integer. */
633 recurse_format_trig_int (char *cp, int value)
635 int trig = value % BASE;
638 cp = recurse_format_trig_int (cp, value);
639 *cp++ = trig_to_char (trig);
643 /* Formats VALUE as a trigesimal integer in null-terminated
644 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
645 FORCE_SIGN is true, a sign is always inserted; otherwise, a
646 sign is only inserted if VALUE is negative. */
648 format_trig_int (int value, bool force_sign, char string[])
659 /* Format integer. */
660 string = recurse_format_trig_int (string, value);
665 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
666 rounding up or down. Returns true if TRIGS[] represents a
667 value greater than half, false if less than half. If TRIGS[]
668 is exactly half, examines TRIGS[-1] and returns true if odd,
669 false if even ("round to even"). */
671 should_round_up (const char trigs[], int trig_cnt)
673 assert (trig_cnt > 0);
675 if (*trigs < BASE / 2)
677 /* Less than half: round down. */
680 else if (*trigs > BASE / 2)
682 /* Greater than half: round up. */
687 /* Approximately half: look more closely. */
689 for (i = 1; i < trig_cnt; i++)
692 /* Slightly greater than half: round up. */
696 /* Exactly half: round to even. */
697 return trigs[-1] % 2;
701 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
702 carrying to the left as necessary. Returns true if
703 successful, false on failure (due to a carry out of the
704 leftmost position). */
706 try_round_up (char *trigs, int trig_cnt)
710 char *round_trig = trigs + --trig_cnt;
711 if (*round_trig != BASE - 1)
713 /* Round this trig up to the next value. */
718 /* Carry over to the next trig to the left. */
722 /* Ran out of trigs to carry. */
726 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
727 equivalent of at least BASE_10_PRECISION decimal digits of
728 precision. The output format may use conventional or
729 scientific notation. Missing, infinite, and extreme values
730 are represented with "*.". */
732 format_trig_double (long double value, int base_10_precision, char output[])
734 /* Original VALUE was negative? */
737 /* Number of significant trigesimals. */
738 int base_30_precision;
740 /* Base-2 significand and exponent for original VALUE. */
744 /* VALUE as a set of trigesimals. */
745 char buffer[DBL_DIG + 16];
749 /* Number of trigesimal places for trigs.
750 trigs[0] has coefficient 30**(trig_places - 1),
751 trigs[1] has coefficient 30**(trig_places - 2),
753 In other words, the trigesimal point is just before trigs[0].
757 /* Number of trigesimal places left to write into BUFFER. */
762 /* Handle special cases. */
768 /* Make VALUE positive. */
777 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
778 point left or right as necessary. We approximate the
779 base-30 exponent by obtaining the base-2 exponent, then
780 multiplying by log30(2). This approximation is sufficient
781 to ensure that the adjusted VALUE is always in the range
782 0...30**6, an invariant of the loop below. */
784 base_2_sig = frexp (value, &base_2_exp);
785 if (errno != 0 || !isfinite (base_2_sig))
787 if (base_2_exp == 0 && base_2_sig == 0.)
789 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
791 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
792 value *= pow30 (CHUNK_SIZE - trig_places);
794 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
797 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
799 trigs_to_output -= CHUNK_SIZE)
804 /* The current chunk is just the integer part of VALUE,
805 truncated to the nearest integer. The chunk fits in a
808 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
809 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
813 /* Append the chunk, in base 30, to trigs[]. */
814 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
816 trigs[trig_cnt + --trigs_left] = chunk % 30;
819 while (trigs_left > 0)
820 trigs[trig_cnt + --trigs_left] = 0;
821 trig_cnt += CHUNK_SIZE;
823 /* Proceed to the next chunk. */
826 value *= pow (LDBASE, CHUNK_SIZE);
829 /* Strip leading zeros. */
830 while (trig_cnt > 1 && *trigs == 0)
837 /* Round to requested precision, conservatively estimating the
838 required base-30 precision as 2/3 of the base-10 precision
839 (log30(10) = .68). */
840 assert (base_10_precision > 0);
841 if (base_10_precision > LDBL_DIG)
842 base_10_precision = LDBL_DIG;
843 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
844 if (trig_cnt > base_30_precision)
846 if (should_round_up (trigs + base_30_precision,
847 trig_cnt - base_30_precision))
849 /* Try to round up. */
850 if (try_round_up (trigs, base_30_precision))
852 /* Rounding up worked. */
853 trig_cnt = base_30_precision;
857 /* Couldn't round up because we ran out of trigs to
858 carry into. Do the carry here instead. */
867 trig_cnt = base_30_precision;
872 /* No rounding required: fewer digits available than
876 /* Strip trailing zeros. */
877 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
883 if (trig_places >= -1 && trig_places < trig_cnt + 3)
885 /* Use conventional notation. */
886 format_trig_digits (output, trigs, trig_cnt, trig_places);
890 /* Use scientific notation. */
892 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
893 op = format_trig_int (trig_places - trig_cnt, true, op);
898 strcpy (output, "0");
902 strcpy (output, "*.");
906 static const struct casewriter_class por_file_casewriter_class =
908 por_file_casewriter_write,
909 por_file_casewriter_destroy,