1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011, 2012 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/file-name.h"
36 #include "data/format.h"
37 #include "data/make-file.h"
38 #include "data/missing-values.h"
39 #include "data/short-names.h"
40 #include "data/value-labels.h"
41 #include "data/variable.h"
42 #include "libpspp/message.h"
43 #include "libpspp/misc.h"
44 #include "libpspp/str.h"
45 #include "libpspp/version.h"
47 #include "gl/minmax.h"
48 #include "gl/xalloc.h"
51 #define _(msgid) gettext (msgid)
52 #define N_(msgid) (msgid)
54 /* Maximum width of a variable in a portable file. */
55 #define MAX_POR_WIDTH 255
57 /* Portable file writer. */
60 struct file_handle *fh; /* File handle. */
61 struct fh_lock *lock; /* Lock on file handle. */
62 FILE *file; /* File stream. */
63 struct replace_file *rf; /* Ticket for replacing output file. */
65 int lc; /* Number of characters on this line so far. */
67 size_t var_cnt; /* Number of variables. */
68 struct pfm_var *vars; /* Variables. */
70 int digits; /* Digits of precision. */
73 /* A variable to write to the portable file. */
76 int width; /* 0=numeric, otherwise string var width. */
77 int fv; /* Starting case index. */
80 static const struct casewriter_class por_file_casewriter_class;
82 static bool close_writer (struct pfm_writer *);
83 static void buf_write (struct pfm_writer *, const void *, size_t);
84 static void write_header (struct pfm_writer *);
85 static void write_version_data (struct pfm_writer *);
86 static void write_variables (struct pfm_writer *, struct dictionary *);
87 static void write_value_labels (struct pfm_writer *,
88 const struct dictionary *);
89 static void write_documents (struct pfm_writer *,
90 const struct dictionary *);
92 static void format_trig_double (long double, int base_10_precision, char[]);
93 static char *format_trig_int (int, bool force_sign, char[]);
95 /* Returns default options for writing a portable file. */
96 struct pfm_write_options
97 pfm_writer_default_options (void)
99 struct pfm_write_options opts;
100 opts.create_writeable = true;
101 opts.type = PFM_COMM;
102 opts.digits = DBL_DIG;
106 /* Writes the dictionary DICT to portable file HANDLE according
107 to the given OPTS. Returns nonzero only if successful. DICT
108 will not be modified, except to assign short names. */
110 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
111 struct pfm_write_options opts)
113 struct pfm_writer *w = NULL;
117 /* Initialize data structures. */
118 w = xmalloc (sizeof *w);
127 w->var_cnt = dict_get_var_cnt (dict);
128 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
129 for (i = 0; i < w->var_cnt; i++)
131 const struct variable *dv = dict_get_var (dict, i);
132 struct pfm_var *pv = &w->vars[i];
133 pv->width = MIN (var_get_width (dv), MAX_POR_WIDTH);
134 pv->fv = var_get_case_index (dv);
137 w->digits = opts.digits;
140 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
146 /* TRANSLATORS: this fragment will be interpolated into
147 messages in fh_lock() that identify types of files. */
148 w->lock = fh_lock (fh, FH_REF_FILE, N_("portable file"), FH_ACC_WRITE, true);
154 if (opts.create_writeable)
156 w->rf = replace_file_start (fh_get_file_name (fh), "w", mode,
160 msg (ME, _("Error opening `%s' for writing as a portable file: %s."),
161 fh_get_file_name (fh), strerror (errno));
165 /* Write file header. */
167 write_version_data (w);
168 write_variables (w, dict);
169 write_value_labels (w, dict);
170 if (dict_get_document_line_cnt (dict) > 0)
171 write_documents (w, dict);
172 buf_write (w, "F", 1);
173 if (ferror (w->file))
175 return casewriter_create (dict_get_proto (dict),
176 &por_file_casewriter_class, w);
183 /* Write NBYTES starting at BUF to the portable file represented by
184 H. Break lines properly every 80 characters. */
186 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
188 const char *buf = buf_;
190 if (ferror (w->file))
193 assert (buf != NULL);
194 while (nbytes + w->lc >= 80)
196 size_t n = 80 - w->lc;
199 fwrite (buf, n, 1, w->file);
200 fwrite ("\r\n", 2, 1, w->file);
206 fwrite (buf, nbytes, 1, w->file);
211 /* Write D to the portable file as a floating-point field. */
213 write_float (struct pfm_writer *w, double d)
216 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
217 buf_write (w, buffer, strlen (buffer));
219 buf_write (w, "/", 1);
222 /* Write N to the portable file as an integer field. */
224 write_int (struct pfm_writer *w, int n)
227 format_trig_int (n, false, buffer);
228 buf_write (w, buffer, strlen (buffer));
229 buf_write (w, "/", 1);
232 /* Write S to the portable file as a string field. */
234 write_string (struct pfm_writer *w, const char *s)
236 size_t n = strlen (s);
237 write_int (w, (int) n);
241 /* Write file header. */
243 write_header (struct pfm_writer *w)
245 static const char spss2ascii[256] =
247 "0000000000000000000000000000000000000000000000000000000000000000"
248 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
249 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
250 "0000000000000000000000000000000000000000000000000000000000000000"
254 for (i = 0; i < 5; i++)
255 buf_write (w, "ASCII SPSS PORT FILE ", 40);
257 buf_write (w, spss2ascii, 256);
258 buf_write (w, "SPSSPORT", 8);
261 /* Writes version, date, and identification records. */
263 write_version_data (struct pfm_writer *w)
271 if ((time_t) -1 == time (&t))
273 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
278 tmp = localtime (&t);
280 sprintf (date_str, "%04d%02d%02d",
281 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
282 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
283 buf_write (w, "A", 1);
284 write_string (w, date_str);
285 write_string (w, time_str);
287 /* Product identification. */
288 buf_write (w, "1", 1);
289 write_string (w, version);
291 /* Subproduct identification. */
292 buf_write (w, "3", 1);
293 write_string (w, host_system);
296 /* Write format F to file H. The format is first resized to fit
297 a value of the given WIDTH, which is handy in case F
298 represents a string longer than 255 bytes and thus WIDTH is
299 truncated to 255 bytes. */
301 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
303 fmt_resize (&f, width);
304 write_int (w, fmt_to_io (f.type));
309 /* Write value V with width WIDTH to file H. */
311 write_value (struct pfm_writer *w, const union value *v, int width)
314 write_float (w, v->f);
317 width = MIN (width, MAX_POR_WIDTH);
318 write_int (w, width);
319 buf_write (w, value_str (v, width), width);
323 /* Write variable records. */
325 write_variables (struct pfm_writer *w, struct dictionary *dict)
329 short_names_assign (dict);
331 if (dict_get_weight (dict) != NULL)
333 buf_write (w, "6", 1);
334 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
337 buf_write (w, "4", 1);
338 write_int (w, dict_get_var_cnt (dict));
341 for (i = 0; i < dict_get_var_cnt (dict); i++)
343 struct variable *v = dict_get_var (dict, i);
344 struct missing_values mv;
345 int width = MIN (var_get_width (v), MAX_POR_WIDTH);
348 buf_write (w, "7", 1);
349 write_int (w, width);
350 write_string (w, var_get_short_name (v, 0));
351 write_format (w, *var_get_print_format (v), width);
352 write_format (w, *var_get_write_format (v), width);
354 /* Write missing values. */
355 mv_copy (&mv, var_get_missing_values (v));
356 if (var_get_width (v) > 8)
358 if (mv_has_range (&mv))
361 mv_get_range (&mv, &x, &y);
364 buf_write (w, "9", 1);
367 else if (y == HIGHEST)
369 buf_write (w, "A", 1);
374 buf_write (w, "B", 1);
379 for (j = 0; j < mv_n_values (&mv); j++)
381 buf_write (w, "8", 1);
382 write_value (w, mv_get_value (&mv, j), mv_get_width (&mv));
386 /* Write variable label. */
387 if (var_get_label (v) != NULL)
389 buf_write (w, "C", 1);
390 write_string (w, var_get_label (v));
395 /* Write value labels to disk. FIXME: Inefficient. */
397 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
401 for (i = 0; i < dict_get_var_cnt (dict); i++)
403 struct variable *v = dict_get_var (dict, i);
404 const struct val_labs *val_labs = var_get_value_labels (v);
405 size_t n_labels = val_labs_count (val_labs);
406 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 (j = 0; j < n_labels; j++)
421 const struct val_lab *vl = labels[j];
422 write_value (w, val_lab_get_value (vl), var_get_width (v));
423 write_string (w, val_lab_get_escaped_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++)
440 write_string (w, dict_get_document_line (dict, i));
444 /* Writes case C to the portable file represented by WRITER. */
446 por_file_casewriter_write (struct casewriter *writer, void *w_,
449 struct pfm_writer *w = w_;
452 if (!ferror (w->file))
454 for (i = 0; i < w->var_cnt; i++)
456 struct pfm_var *v = &w->vars[i];
459 write_float (w, case_num_idx (c, v->fv));
462 write_int (w, v->width);
463 buf_write (w, case_str_idx (c, v->fv), v->width);
468 casewriter_force_error (writer);
474 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
476 struct pfm_writer *w = w_;
477 if (!close_writer (w))
478 casewriter_force_error (writer);
481 /* Closes a portable file after we're done with it.
482 Returns true if successful, false if an I/O error occurred. */
484 close_writer (struct pfm_writer *w)
495 memset (buf, 'Z', sizeof buf);
496 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
498 ok = !ferror (w->file);
499 if (fclose (w->file) == EOF)
503 msg (ME, _("An I/O error occurred writing portable file `%s'."),
504 fh_get_file_name (w->fh));
506 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
519 /* Base-30 conversion.
521 Portable files represent numbers in base-30 format, so we need
522 to be able to convert real and integer number to that base.
523 Older versions of PSPP used libgmp to do so, but this added a
524 big library dependency to do just one thing. Now we do it
525 ourselves internally.
527 Important fact: base 30 is called "trigesimal". */
529 /* Conversion base. */
530 #define BASE 30 /* As an integer. */
531 #define LDBASE ((long double) BASE) /* As a long double. */
533 /* This is floor(log30(2**31)), the minimum number of trigesimal
534 digits that a `long int' can hold. */
537 /* pow_tab[i] = pow (30, pow (2, i)) */
538 static long double pow_tab[16];
540 /* Initializes pow_tab[]. */
544 static bool did_init = false;
548 /* Only initialize once. */
553 /* Set each element of pow_tab[] until we run out of numerical
556 for (power = 30.0L; power < DBL_MAX; power *= power)
558 assert (i < sizeof pow_tab / sizeof *pow_tab);
559 pow_tab[i++] = power;
563 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
565 pow30_nonnegative (int exponent)
570 assert (exponent >= 0);
571 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
574 for (i = 0; exponent > 0; exponent >>= 1, i++)
581 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
587 return pow30_nonnegative (exponent);
589 return 1.L / pow30_nonnegative (-exponent);
592 /* Returns the character corresponding to TRIG. */
594 trig_to_char (int trig)
596 assert (trig >= 0 && trig < 30);
597 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
600 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
601 null-terminated STRING. The trigesimal point is inserted
602 after TRIG_PLACES characters have been printed, if necessary
603 adding extra zeros at either end for correctness. Returns the
604 character after the formatted number. */
606 format_trig_digits (char *string,
607 const char trigs[], int trig_cnt, int trig_places)
612 while (trig_places++ < 0)
616 while (trig_cnt-- > 0)
618 if (trig_places-- == 0)
620 *string++ = trig_to_char (*trigs++);
622 while (trig_places-- > 0)
628 /* Helper function for format_trig_int() that formats VALUE as a
629 trigesimal integer at CP. VALUE must be nonnegative.
630 Returns the character following the formatted integer. */
632 recurse_format_trig_int (char *cp, int value)
634 int trig = value % BASE;
637 cp = recurse_format_trig_int (cp, value);
638 *cp++ = trig_to_char (trig);
642 /* Formats VALUE as a trigesimal integer in null-terminated
643 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
644 FORCE_SIGN is true, a sign is always inserted; otherwise, a
645 sign is only inserted if VALUE is negative. */
647 format_trig_int (int value, bool force_sign, char string[])
658 /* Format integer. */
659 string = recurse_format_trig_int (string, value);
664 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
665 rounding up or down. Returns true if TRIGS[] represents a
666 value greater than half, false if less than half. If TRIGS[]
667 is exactly half, examines TRIGS[-1] and returns true if odd,
668 false if even ("round to even"). */
670 should_round_up (const char trigs[], int trig_cnt)
672 assert (trig_cnt > 0);
674 if (*trigs < BASE / 2)
676 /* Less than half: round down. */
679 else if (*trigs > BASE / 2)
681 /* Greater than half: round up. */
686 /* Approximately half: look more closely. */
688 for (i = 1; i < trig_cnt; i++)
691 /* Slightly greater than half: round up. */
695 /* Exactly half: round to even. */
696 return trigs[-1] % 2;
700 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
701 carrying to the left as necessary. Returns true if
702 successful, false on failure (due to a carry out of the
703 leftmost position). */
705 try_round_up (char *trigs, int trig_cnt)
709 char *round_trig = trigs + --trig_cnt;
710 if (*round_trig != BASE - 1)
712 /* Round this trig up to the next value. */
717 /* Carry over to the next trig to the left. */
721 /* Ran out of trigs to carry. */
725 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
726 equivalent of at least BASE_10_PRECISION decimal digits of
727 precision. The output format may use conventional or
728 scientific notation. Missing, infinite, and extreme values
729 are represented with "*.". */
731 format_trig_double (long double value, int base_10_precision, char output[])
733 /* Original VALUE was negative? */
736 /* Number of significant trigesimals. */
737 int base_30_precision;
739 /* Base-2 significand and exponent for original VALUE. */
743 /* VALUE as a set of trigesimals. */
744 char buffer[DBL_DIG + 16];
748 /* Number of trigesimal places for trigs.
749 trigs[0] has coefficient 30**(trig_places - 1),
750 trigs[1] has coefficient 30**(trig_places - 2),
752 In other words, the trigesimal point is just before trigs[0].
756 /* Number of trigesimal places left to write into BUFFER. */
761 /* Handle special cases. */
767 /* Make VALUE positive. */
776 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
777 point left or right as necessary. We approximate the
778 base-30 exponent by obtaining the base-2 exponent, then
779 multiplying by log30(2). This approximation is sufficient
780 to ensure that the adjusted VALUE is always in the range
781 0...30**6, an invariant of the loop below. */
783 base_2_sig = frexp (value, &base_2_exp);
784 if (errno != 0 || !isfinite (base_2_sig))
786 if (base_2_exp == 0 && base_2_sig == 0.)
788 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
790 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
791 value *= pow30 (CHUNK_SIZE - trig_places);
793 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
796 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
798 trigs_to_output -= CHUNK_SIZE)
803 /* The current chunk is just the integer part of VALUE,
804 truncated to the nearest integer. The chunk fits in a
807 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
808 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
812 /* Append the chunk, in base 30, to trigs[]. */
813 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
815 trigs[trig_cnt + --trigs_left] = chunk % 30;
818 while (trigs_left > 0)
819 trigs[trig_cnt + --trigs_left] = 0;
820 trig_cnt += CHUNK_SIZE;
822 /* Proceed to the next chunk. */
825 value *= pow (LDBASE, CHUNK_SIZE);
828 /* Strip leading zeros. */
829 while (trig_cnt > 1 && *trigs == 0)
836 /* Round to requested precision, conservatively estimating the
837 required base-30 precision as 2/3 of the base-10 precision
838 (log30(10) = .68). */
839 assert (base_10_precision > 0);
840 if (base_10_precision > LDBL_DIG)
841 base_10_precision = LDBL_DIG;
842 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
843 if (trig_cnt > base_30_precision)
845 if (should_round_up (trigs + base_30_precision,
846 trig_cnt - base_30_precision))
848 /* Try to round up. */
849 if (try_round_up (trigs, base_30_precision))
851 /* Rounding up worked. */
852 trig_cnt = base_30_precision;
856 /* Couldn't round up because we ran out of trigs to
857 carry into. Do the carry here instead. */
866 trig_cnt = base_30_precision;
871 /* No rounding required: fewer digits available than
875 /* Strip trailing zeros. */
876 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
882 if (trig_places >= -1 && trig_places < trig_cnt + 3)
884 /* Use conventional notation. */
885 format_trig_digits (output, trigs, trig_cnt, trig_places);
889 /* Use scientific notation. */
891 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
892 op = format_trig_int (trig_places - trig_cnt, true, op);
897 strcpy (output, "0");
901 strcpy (output, "*.");
905 static const struct casewriter_class por_file_casewriter_class =
907 por_file_casewriter_write,
908 por_file_casewriter_destroy,