2 * Copyright (C) 1994-1999, Index Data
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.34 1999-05-26 07:49:13 adam
10 * Revision 1.33 1999/05/15 14:36:38 adam
11 * Updated dictionary. Implemented "compression" of dictionary.
13 * Revision 1.32 1999/05/12 13:08:06 adam
14 * First version of ISAMS.
16 * Revision 1.31 1999/02/02 14:50:56 adam
17 * Updated WIN32 code specific sections. Changed header.
19 * Revision 1.30 1998/10/28 10:53:57 adam
20 * Added type cast to prevent warning.
22 * Revision 1.29 1998/06/11 15:41:39 adam
25 * Revision 1.28 1998/03/05 08:45:12 adam
26 * New result set model and modular ranking system. Moved towards
27 * descent server API. System information stored as "SGML" records.
29 * Revision 1.27 1998/02/17 10:32:52 adam
30 * Fixed bug: binary files weren't opened with flag b on NT.
32 * Revision 1.26 1998/01/29 13:39:13 adam
33 * Compress ISAM is default.
35 * Revision 1.25 1997/09/17 12:19:14 adam
36 * Zebra version corresponds to YAZ version 1.4.
37 * Changed Zebra server so that it doesn't depend on global common_resource.
39 * Revision 1.24 1997/09/09 13:38:07 adam
40 * Partial port to WIN95/NT.
42 * Revision 1.23 1997/09/04 13:57:39 adam
43 * Added O_BINARY for open calls.
45 * Revision 1.22 1997/02/12 20:39:45 adam
46 * Implemented options -f <n> that limits the log to the first <n>
48 * Changed some log messages also.
50 * Revision 1.21 1996/11/08 11:10:23 adam
51 * Buffers used during file match got bigger.
52 * Compressed ISAM support everywhere.
53 * Bug fixes regarding masking characters in queries.
54 * Redesigned Regexp-2 queries.
56 * Revision 1.20 1996/11/01 08:58:41 adam
57 * Interface to isamc system now includes update and delete.
59 * Revision 1.19 1996/10/29 14:09:46 adam
60 * Use of cisam system - enabled if setting isamc is 1.
62 * Revision 1.18 1996/06/04 10:18:59 adam
63 * Minor changes - removed include of ctype.h.
65 * Revision 1.17 1996/05/14 15:47:07 adam
66 * Cleanup of various buffer size entities.
68 * Revision 1.16 1996/04/09 10:05:20 adam
69 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
71 * Revision 1.15 1996/03/21 14:50:09 adam
72 * File update uses modify-time instead of change-time.
74 * Revision 1.14 1996/02/07 14:06:37 adam
75 * Better progress report during register merge.
76 * New command: clean - removes temporary shadow files.
78 * Revision 1.13 1996/02/05 12:30:00 adam
79 * Logging reduced a bit.
80 * The remaining running time is estimated during register merge.
82 * Revision 1.12 1995/12/06 17:49:19 adam
83 * Uses dict_delete now.
85 * Revision 1.11 1995/12/06 16:06:43 adam
86 * Better diagnostics. Work on 'real' dictionary deletion.
88 * Revision 1.10 1995/12/06 12:41:22 adam
89 * New command 'stat' for the index program.
90 * Filenames can be read from stdin by specifying '-'.
91 * Bug fix/enhancement of the transformation from terms to regular
92 * expressons in the search engine.
94 * Revision 1.9 1995/10/10 12:24:39 adam
95 * Temporary sort files are compressed.
97 * Revision 1.8 1995/10/04 16:57:19 adam
98 * Key input and merge sort in one pass.
100 * Revision 1.7 1995/10/02 15:18:52 adam
101 * New member in recRetrieveCtrl: diagnostic.
103 * Revision 1.6 1995/09/29 15:51:56 adam
104 * First work on multi-way read.
106 * Revision 1.5 1995/09/29 14:01:43 adam
109 * Revision 1.4 1995/09/28 14:22:57 adam
110 * Sort uses smaller temporary files.
112 * Revision 1.3 1995/09/06 16:11:17 adam
113 * Option: only one word key per file.
115 * Revision 1.2 1995/09/04 12:33:42 adam
116 * Various cleanup. YAZ util used instead.
118 * Revision 1.1 1995/09/04 09:10:37 adam
119 * More work on index add/del/update.
120 * Merge sort implemented.
121 * Initial work on z39 server.
138 #define KEY_SIZE (1+sizeof(struct it_key))
139 #define INP_NAME_MAX 768
140 #define INP_BUF_START 60000
141 #define INP_BUF_ADD 400000
143 static int no_diffs = 0;
144 static int no_updates = 0;
145 static int no_deletions = 0;
146 static int no_insertions = 0;
147 static int no_iterations = 0;
150 int no; /* file no */
151 off_t offset; /* file offset */
152 unsigned char *buf; /* buffer block */
153 size_t buf_size; /* number of read bytes in block */
154 size_t chunk; /* number of bytes allocated */
155 size_t buf_ptr; /* current position in buffer */
156 char *prev_name; /* last word read */
157 int sysno; /* last sysno */
158 int seqno; /* last seqno */
159 off_t length; /* length of file */
160 /* handler invoked in each read */
161 void (*readHandler)(struct key_file *keyp, void *rinfo);
165 void getFnameTmp (char *fname, int no)
169 pre = res_get_def (common_resource, "keyTmpDir", ".");
170 sprintf (fname, "%s/key%d.tmp", pre, no);
173 void key_file_chunk_read (struct key_file *f)
175 int nr = 0, r = 0, fd;
177 getFnameTmp (fname, f->no);
178 fd = open (fname, O_BINARY|O_RDONLY);
181 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
186 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
188 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
192 if (lseek (fd, f->offset, SEEK_SET) == -1)
194 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
197 while (f->chunk - nr > 0)
199 r = read (fd, f->buf + nr, f->chunk - nr);
206 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
212 (*f->readHandler)(f, f->readInfo);
216 struct key_file *key_file_init (int no, int chunk)
220 f = (struct key_file *) xmalloc (sizeof(*f));
227 f->readHandler = NULL;
228 f->buf = (unsigned char *) xmalloc (f->chunk);
229 f->prev_name = (char *) xmalloc (INP_NAME_MAX);
230 *f->prev_name = '\0';
231 key_file_chunk_read (f);
235 int key_file_getc (struct key_file *f)
237 if (f->buf_ptr < f->buf_size)
238 return f->buf[(f->buf_ptr)++];
239 if (f->buf_size < f->chunk)
241 f->offset += f->buf_size;
242 key_file_chunk_read (f);
243 if (f->buf_ptr < f->buf_size)
244 return f->buf[(f->buf_ptr)++];
249 int key_file_decode (struct key_file *f)
253 c = key_file_getc (f);
260 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
263 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
264 d = (d << 8) + (key_file_getc (f) & 0xff);
267 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
268 d = (d << 8) + (key_file_getc (f) & 0xff);
269 d = (d << 8) + (key_file_getc (f) & 0xff);
275 int key_file_read (struct key_file *f, char *key)
280 c = key_file_getc (f);
283 strcpy (key, f->prev_name);
292 while ((key[i++] = key_file_getc (f)))
294 strcpy (f->prev_name, key);
297 d = key_file_decode (f);
300 itkey.sysno = d + f->sysno;
303 f->sysno = itkey.sysno;
306 d = key_file_decode (f);
307 itkey.seqno = d + f->seqno;
308 f->seqno = itkey.seqno;
309 memcpy (key + i, &itkey, sizeof(struct it_key));
310 return i + sizeof (struct it_key);
315 struct key_file **file;
320 int (*cmp)(const void *p1, const void *p2);
327 struct heap_info *key_heap_init (int nkeys,
328 int (*cmp)(const void *p1, const void *p2))
330 struct heap_info *hi;
333 hi = (struct heap_info *) xmalloc (sizeof(*hi));
334 hi->info.file = (struct key_file **)
335 xmalloc (sizeof(*hi->info.file) * (1+nkeys));
336 hi->info.buf = (char **) xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
338 hi->ptr = (int *) xmalloc (sizeof(*hi->ptr) * (1+nkeys));
340 for (i = 0; i<= nkeys; i++)
343 hi->info.buf[i] = (char *) xmalloc (INP_NAME_MAX);
348 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
353 hi->ptr[i1] = hi->ptr[i2];
358 static void key_heap_delete (struct heap_info *hi)
360 int cur = 1, child = 2;
362 assert (hi->heapnum > 0);
364 key_heap_swap (hi, 1, hi->heapnum);
366 while (child <= hi->heapnum) {
367 if (child < hi->heapnum &&
368 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
369 &hi->info.buf[hi->ptr[child+1]]) > 0)
371 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
372 &hi->info.buf[hi->ptr[child]]) > 0)
374 key_heap_swap (hi, cur, child);
383 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
388 cur = ++(hi->heapnum);
389 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
390 hi->info.file[hi->ptr[cur]] = kf;
393 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
394 &hi->info.buf[hi->ptr[cur]]) > 0)
396 key_heap_swap (hi, cur, parent);
402 static int heap_read_one (struct heap_info *hi, char *name, char *key)
405 char rbuf[INP_NAME_MAX];
411 strcpy (name, hi->info.buf[n]);
412 kf = hi->info.file[n];
414 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
415 key_heap_delete (hi);
416 if ((r = key_file_read (kf, rbuf)))
417 key_heap_insert (hi, rbuf, r, kf);
422 struct heap_cread_info {
423 char prev_name[INP_NAME_MAX];
424 char cur_name[INP_NAME_MAX];
426 struct heap_info *hi;
431 int heap_cread_item (void *vp, char **dst, int *insertMode)
433 struct heap_cread_info *p = (struct heap_cread_info *) vp;
434 struct heap_info *hi = p->hi;
438 *insertMode = p->key[0];
439 memcpy (*dst, p->key+1, sizeof(struct it_key));
440 (*dst) += sizeof(struct it_key);
444 strcpy (p->prev_name, p->cur_name);
445 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
447 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
452 *insertMode = p->key[0];
453 memcpy (*dst, p->key+1, sizeof(struct it_key));
454 (*dst) += sizeof(struct it_key);
458 int heap_inpc (struct heap_info *hi)
460 struct heap_cread_info hci;
461 ISAMC_I isamc_i = (ISAMC_I) xmalloc (sizeof(*isamc_i));
463 hci.key = (char *) xmalloc (KEY_SIZE);
466 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
468 isamc_i->clientData = &hci;
469 isamc_i->read_item = heap_cread_item;
473 char this_name[INP_NAME_MAX];
474 ISAMC_P isamc_p, isamc_p2;
477 strcpy (this_name, hci.cur_name);
478 assert (hci.cur_name[1]);
480 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
482 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
483 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
487 if (!dict_delete (hi->dict, this_name))
493 if (isamc_p2 != isamc_p)
494 dict_insert (hi->dict, this_name,
495 sizeof(ISAMC_P), &isamc_p2);
500 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
502 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
509 int heap_inps (struct heap_info *hi)
511 struct heap_cread_info hci;
512 ISAMS_I isams_i = (ISAMS_I) xmalloc (sizeof(*isams_i));
514 hci.key = (char *) xmalloc (KEY_SIZE);
517 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
519 isams_i->clientData = &hci;
520 isams_i->read_item = heap_cread_item;
524 char this_name[INP_NAME_MAX];
528 strcpy (this_name, hci.cur_name);
529 assert (hci.cur_name[1]);
531 if (!(dict_info = dict_lookup (hi->dict, hci.cur_name)))
533 isams_p = isams_merge (hi->isams, isams_i);
535 dict_insert (hi->dict, this_name, sizeof(ISAMS_P), &isams_p);
545 int heap_inp (struct heap_info *hi)
548 char next_name[INP_NAME_MAX];
549 char cur_name[INP_NAME_MAX];
550 int key_buf_size = INP_BUF_START;
556 next_key = (char *) xmalloc (KEY_SIZE);
557 key_buf = (char *) xmalloc (key_buf_size);
558 more = heap_read_one (hi, cur_name, key_buf);
559 while (more) /* EOF ? */
562 key_buf_ptr = KEY_SIZE;
565 if (!(more = heap_read_one (hi, next_name, next_key)))
567 if (*next_name && strcmp (next_name, cur_name))
569 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
570 key_buf_ptr += KEY_SIZE;
571 if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size)
574 new_key_buf = (char *) xmalloc (key_buf_size + INP_BUF_ADD);
575 memcpy (new_key_buf, key_buf, key_buf_size);
576 key_buf_size += INP_BUF_ADD;
578 key_buf = new_key_buf;
582 nmemb = key_buf_ptr / KEY_SIZE;
583 assert (nmemb * (int) KEY_SIZE == key_buf_ptr);
584 if ((info = dict_lookup (hi->dict, cur_name)))
586 ISAM_P isam_p, isam_p2;
587 memcpy (&isam_p, info+1, sizeof(ISAM_P));
588 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
592 if (!dict_delete (hi->dict, cur_name))
598 if (isam_p2 != isam_p)
599 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
606 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
607 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
609 memcpy (key_buf, next_key, KEY_SIZE);
610 strcpy (cur_name, next_name);
615 struct progressInfo {
622 void progressFunc (struct key_file *keyp, void *info)
624 struct progressInfo *p = (struct progressInfo *) info;
625 time_t now, remaining;
627 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
631 if (now >= p->lastTime+10)
634 remaining = (time_t) ((now - p->startTime)*
635 ((double) p->totalBytes/p->totalOffset - 1.0));
636 if (remaining <= 130)
637 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
638 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
640 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
641 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
643 p->totalOffset += keyp->buf_size;
650 void key_input (BFiles bfs, int nkeys, int cache)
657 struct key_file **kf;
660 struct heap_info *hi;
661 struct progressInfo progressInfo;
669 getFnameTmp (fname, nkeys+1);
670 if (access (fname, R_OK) == -1)
677 dict = dict_open (bfs, FNAME_DICT, cache, 1, 0);
680 logf (LOG_FATAL, "dict_open fail");
683 if (res_get_match (common_resource, "isam", "s", NULL))
685 isams = isams_open (bfs, FNAME_ISAMS, 1,
686 key_isams_m (common_resource));
689 logf (LOG_FATAL, "isams_open fail");
693 else if (res_get_match (common_resource, "isam", "i", NULL))
695 isam = is_open (bfs, FNAME_ISAM, key_compare, 1,
696 sizeof(struct it_key), common_resource);
699 logf (LOG_FATAL, "is_open fail");
705 isamc = isc_open (bfs, FNAME_ISAMC, 1,
706 key_isamc_m (common_resource));
709 logf (LOG_FATAL, "isc_open fail");
713 kf = (struct key_file **) xmalloc ((1+nkeys) * sizeof(*kf));
714 progressInfo.totalBytes = 0;
715 progressInfo.totalOffset = 0;
716 time (&progressInfo.startTime);
717 time (&progressInfo.lastTime);
718 for (i = 1; i<=nkeys; i++)
720 kf[i] = key_file_init (i, 32768);
721 kf[i]->readHandler = progressFunc;
722 kf[i]->readInfo = &progressInfo;
723 progressInfo.totalBytes += kf[i]->length;
724 progressInfo.totalOffset += kf[i]->buf_size;
726 hi = key_heap_init (nkeys, key_qsort_compare);
732 for (i = 1; i<=nkeys; i++)
733 if ((r = key_file_read (kf[i], rbuf)))
734 key_heap_insert (hi, rbuf, r, kf[i]);
749 for (i = 1; i<=nkeys; i++)
751 getFnameTmp (rbuf, i);
754 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
755 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
756 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
757 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
758 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);