2 * Copyright (C) 1994-1996, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.24 1997-09-09 13:38:07 adam
8 * Partial port to WIN95/NT.
10 * Revision 1.23 1997/09/04 13:57:39 adam
11 * Added O_BINARY for open calls.
13 * Revision 1.22 1997/02/12 20:39:45 adam
14 * Implemented options -f <n> that limits the log to the first <n>
16 * Changed some log messages also.
18 * Revision 1.21 1996/11/08 11:10:23 adam
19 * Buffers used during file match got bigger.
20 * Compressed ISAM support everywhere.
21 * Bug fixes regarding masking characters in queries.
22 * Redesigned Regexp-2 queries.
24 * Revision 1.20 1996/11/01 08:58:41 adam
25 * Interface to isamc system now includes update and delete.
27 * Revision 1.19 1996/10/29 14:09:46 adam
28 * Use of cisam system - enabled if setting isamc is 1.
30 * Revision 1.18 1996/06/04 10:18:59 adam
31 * Minor changes - removed include of ctype.h.
33 * Revision 1.17 1996/05/14 15:47:07 adam
34 * Cleanup of various buffer size entities.
36 * Revision 1.16 1996/04/09 10:05:20 adam
37 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
39 * Revision 1.15 1996/03/21 14:50:09 adam
40 * File update uses modify-time instead of change-time.
42 * Revision 1.14 1996/02/07 14:06:37 adam
43 * Better progress report during register merge.
44 * New command: clean - removes temporary shadow files.
46 * Revision 1.13 1996/02/05 12:30:00 adam
47 * Logging reduced a bit.
48 * The remaining running time is estimated during register merge.
50 * Revision 1.12 1995/12/06 17:49:19 adam
51 * Uses dict_delete now.
53 * Revision 1.11 1995/12/06 16:06:43 adam
54 * Better diagnostics. Work on 'real' dictionary deletion.
56 * Revision 1.10 1995/12/06 12:41:22 adam
57 * New command 'stat' for the index program.
58 * Filenames can be read from stdin by specifying '-'.
59 * Bug fix/enhancement of the transformation from terms to regular
60 * expressons in the search engine.
62 * Revision 1.9 1995/10/10 12:24:39 adam
63 * Temporary sort files are compressed.
65 * Revision 1.8 1995/10/04 16:57:19 adam
66 * Key input and merge sort in one pass.
68 * Revision 1.7 1995/10/02 15:18:52 adam
69 * New member in recRetrieveCtrl: diagnostic.
71 * Revision 1.6 1995/09/29 15:51:56 adam
72 * First work on multi-way read.
74 * Revision 1.5 1995/09/29 14:01:43 adam
77 * Revision 1.4 1995/09/28 14:22:57 adam
78 * Sort uses smaller temporary files.
80 * Revision 1.3 1995/09/06 16:11:17 adam
81 * Option: only one word key per file.
83 * Revision 1.2 1995/09/04 12:33:42 adam
84 * Various cleanup. YAZ util used instead.
86 * Revision 1.1 1995/09/04 09:10:37 adam
87 * More work on index add/del/update.
88 * Merge sort implemented.
89 * Initial work on z39 server.
106 #define KEY_SIZE (1+sizeof(struct it_key))
107 #define INP_NAME_MAX 768
108 #define INP_BUF_START 60000
109 #define INP_BUF_ADD 400000
111 static int no_diffs = 0;
112 static int no_updates = 0;
113 static int no_deletions = 0;
114 static int no_insertions = 0;
115 static int no_iterations = 0;
118 int no; /* file no */
119 off_t offset; /* file offset */
120 unsigned char *buf; /* buffer block */
121 size_t buf_size; /* number of read bytes in block */
122 size_t chunk; /* number of bytes allocated */
123 size_t buf_ptr; /* current position in buffer */
124 char *prev_name; /* last word read */
125 int sysno; /* last sysno */
126 int seqno; /* last seqno */
127 off_t length; /* length of file */
128 /* handler invoked in each read */
129 void (*readHandler)(struct key_file *keyp, void *rinfo);
133 void getFnameTmp (char *fname, int no)
137 pre = res_get_def (common_resource, "keyTmpDir", ".");
138 sprintf (fname, "%s/key%d.tmp", pre, no);
141 void key_file_chunk_read (struct key_file *f)
145 getFnameTmp (fname, f->no);
146 fd = open (fname, O_BINARY|O_RDONLY);
149 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
154 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
156 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
160 if (lseek (fd, f->offset, SEEK_SET) == -1)
162 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
165 while (f->chunk - nr > 0)
167 r = read (fd, f->buf + nr, f->chunk - nr);
174 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
180 (*f->readHandler)(f, f->readInfo);
184 struct key_file *key_file_init (int no, int chunk)
188 f = xmalloc (sizeof(*f));
195 f->readHandler = NULL;
196 f->buf = xmalloc (f->chunk);
197 f->prev_name = xmalloc (INP_NAME_MAX);
198 *f->prev_name = '\0';
199 key_file_chunk_read (f);
203 int key_file_getc (struct key_file *f)
205 if (f->buf_ptr < f->buf_size)
206 return f->buf[(f->buf_ptr)++];
207 if (f->buf_size < f->chunk)
209 f->offset += f->buf_size;
210 key_file_chunk_read (f);
211 if (f->buf_ptr < f->buf_size)
212 return f->buf[(f->buf_ptr)++];
217 int key_file_decode (struct key_file *f)
221 c = key_file_getc (f);
228 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
231 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
232 d = (d << 8) + (key_file_getc (f) & 0xff);
235 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
236 d = (d << 8) + (key_file_getc (f) & 0xff);
237 d = (d << 8) + (key_file_getc (f) & 0xff);
243 int key_file_read (struct key_file *f, char *key)
248 c = key_file_getc (f);
251 strcpy (key, f->prev_name);
260 while ((key[i++] = key_file_getc (f)))
262 strcpy (f->prev_name, key);
265 d = key_file_decode (f);
268 itkey.sysno = d + f->sysno;
271 f->sysno = itkey.sysno;
274 d = key_file_decode (f);
275 itkey.seqno = d + f->seqno;
276 f->seqno = itkey.seqno;
277 memcpy (key + i, &itkey, sizeof(struct it_key));
278 return i + sizeof (struct it_key);
283 struct key_file **file;
288 int (*cmp)(const void *p1, const void *p2);
294 struct heap_info *key_heap_init (int nkeys,
295 int (*cmp)(const void *p1, const void *p2))
297 struct heap_info *hi;
300 hi = xmalloc (sizeof(*hi));
301 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
302 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
304 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
306 for (i = 0; i<= nkeys; i++)
309 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
314 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
319 hi->ptr[i1] = hi->ptr[i2];
324 static void key_heap_delete (struct heap_info *hi)
326 int cur = 1, child = 2;
328 assert (hi->heapnum > 0);
330 key_heap_swap (hi, 1, hi->heapnum);
332 while (child <= hi->heapnum) {
333 if (child < hi->heapnum &&
334 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
335 &hi->info.buf[hi->ptr[child+1]]) > 0)
337 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
338 &hi->info.buf[hi->ptr[child]]) > 0)
340 key_heap_swap (hi, cur, child);
349 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
354 cur = ++(hi->heapnum);
355 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
356 hi->info.file[hi->ptr[cur]] = kf;
359 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
360 &hi->info.buf[hi->ptr[cur]]) > 0)
362 key_heap_swap (hi, cur, parent);
368 static int heap_read_one (struct heap_info *hi, char *name, char *key)
371 char rbuf[INP_NAME_MAX];
377 strcpy (name, hi->info.buf[n]);
378 kf = hi->info.file[n];
380 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
381 key_heap_delete (hi);
382 if ((r = key_file_read (kf, rbuf)))
383 key_heap_insert (hi, rbuf, r, kf);
388 struct heap_cread_info {
389 char prev_name[INP_NAME_MAX];
390 char cur_name[INP_NAME_MAX];
392 struct heap_info *hi;
397 int heap_cread_item (void *vp, char **dst, int *insertMode)
399 struct heap_cread_info *p = vp;
400 struct heap_info *hi = p->hi;
404 *insertMode = p->key[0];
405 memcpy (*dst, p->key+1, sizeof(struct it_key));
406 (*dst) += sizeof(struct it_key);
410 strcpy (p->prev_name, p->cur_name);
411 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
413 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
418 *insertMode = p->key[0];
419 memcpy (*dst, p->key+1, sizeof(struct it_key));
420 (*dst) += sizeof(struct it_key);
424 int heap_inpc (struct heap_info *hi)
426 struct heap_cread_info hci;
427 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
429 hci.key = xmalloc (KEY_SIZE);
432 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
434 isamc_i->clientData = &hci;
435 isamc_i->read_item = heap_cread_item;
439 char this_name[INP_NAME_MAX];
440 ISAMC_P isamc_p, isamc_p2;
443 strcpy (this_name, hci.cur_name);
444 logf (LOG_DEBUG, "inserting %s", 1+hci.cur_name);
446 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
448 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
449 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
453 if (!dict_delete (hi->dict, this_name))
459 if (isamc_p2 != isamc_p)
460 dict_insert (hi->dict, this_name,
461 sizeof(ISAMC_P), &isamc_p2);
466 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
468 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
475 int heap_inp (struct heap_info *hi)
478 char next_name[INP_NAME_MAX];
479 char cur_name[INP_NAME_MAX];
480 int key_buf_size = INP_BUF_START;
486 next_key = xmalloc (KEY_SIZE);
487 key_buf = xmalloc (key_buf_size);
488 more = heap_read_one (hi, cur_name, key_buf);
489 while (more) /* EOF ? */
492 key_buf_ptr = KEY_SIZE;
495 if (!(more = heap_read_one (hi, next_name, next_key)))
497 if (*next_name && strcmp (next_name, cur_name))
499 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
500 key_buf_ptr += KEY_SIZE;
501 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
504 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
505 memcpy (new_key_buf, key_buf, key_buf_size);
506 key_buf_size += INP_BUF_ADD;
508 key_buf = new_key_buf;
512 nmemb = key_buf_ptr / KEY_SIZE;
513 assert (nmemb*KEY_SIZE == key_buf_ptr);
514 if ((info = dict_lookup (hi->dict, cur_name)))
516 ISAM_P isam_p, isam_p2;
517 logf (LOG_DEBUG, "updating %s", 1+cur_name);
518 memcpy (&isam_p, info+1, sizeof(ISAM_P));
519 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
523 if (!dict_delete (hi->dict, cur_name))
529 if (isam_p2 != isam_p)
530 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
536 logf (LOG_DEBUG, "inserting %s", 1+cur_name);
538 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
539 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
541 memcpy (key_buf, next_key, KEY_SIZE);
542 strcpy (cur_name, next_name);
547 struct progressInfo {
554 void progressFunc (struct key_file *keyp, void *info)
556 struct progressInfo *p = info;
557 time_t now, remaining;
559 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
563 if (now >= p->lastTime+10)
566 remaining = (now - p->startTime)*
567 ((double) p->totalBytes/p->totalOffset - 1.0);
568 if (remaining <= 130)
569 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
570 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
572 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
573 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
575 p->totalOffset += keyp->buf_size;
582 void key_input (int nkeys, int cache)
588 struct key_file **kf;
591 struct heap_info *hi;
592 struct progressInfo progressInfo;
600 getFnameTmp (fname, nkeys+1);
601 if (access (fname, R_OK) == -1)
608 dict = dict_open (FNAME_DICT, cache, 1);
611 logf (LOG_FATAL, "dict_open fail");
614 if (res_get_match (common_resource, "isam", "c", NULL))
616 isamc = isc_open (FNAME_ISAMC, 1, key_isamc_m ());
619 logf (LOG_FATAL, "isc_open fail");
625 isam = is_open (FNAME_ISAM, key_compare, 1, sizeof(struct it_key));
628 logf (LOG_FATAL, "is_open fail");
632 kf = xmalloc ((1+nkeys) * sizeof(*kf));
633 progressInfo.totalBytes = 0;
634 progressInfo.totalOffset = 0;
635 time (&progressInfo.startTime);
636 time (&progressInfo.lastTime);
637 for (i = 1; i<=nkeys; i++)
639 kf[i] = key_file_init (i, 32768);
640 kf[i]->readHandler = progressFunc;
641 kf[i]->readInfo = &progressInfo;
642 progressInfo.totalBytes += kf[i]->length;
643 progressInfo.totalOffset += kf[i]->buf_size;
645 hi = key_heap_init (nkeys, key_qsort_compare);
650 for (i = 1; i<=nkeys; i++)
651 if ((r = key_file_read (kf[i], rbuf)))
652 key_heap_insert (hi, rbuf, r, kf[i]);
663 for (i = 1; i<=nkeys; i++)
665 getFnameTmp (rbuf, i);
668 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
669 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
670 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
671 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
672 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);