2 * Copyright (C) 1994-1996, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.20 1996-11-01 08:58:41 adam
8 * Interface to isamc system now includes update and delete.
10 * Revision 1.19 1996/10/29 14:09:46 adam
11 * Use of cisam system - enabled if setting isamc is 1.
13 * Revision 1.18 1996/06/04 10:18:59 adam
14 * Minor changes - removed include of ctype.h.
16 * Revision 1.17 1996/05/14 15:47:07 adam
17 * Cleanup of various buffer size entities.
19 * Revision 1.16 1996/04/09 10:05:20 adam
20 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
22 * Revision 1.15 1996/03/21 14:50:09 adam
23 * File update uses modify-time instead of change-time.
25 * Revision 1.14 1996/02/07 14:06:37 adam
26 * Better progress report during register merge.
27 * New command: clean - removes temporary shadow files.
29 * Revision 1.13 1996/02/05 12:30:00 adam
30 * Logging reduced a bit.
31 * The remaining running time is estimated during register merge.
33 * Revision 1.12 1995/12/06 17:49:19 adam
34 * Uses dict_delete now.
36 * Revision 1.11 1995/12/06 16:06:43 adam
37 * Better diagnostics. Work on 'real' dictionary deletion.
39 * Revision 1.10 1995/12/06 12:41:22 adam
40 * New command 'stat' for the index program.
41 * Filenames can be read from stdin by specifying '-'.
42 * Bug fix/enhancement of the transformation from terms to regular
43 * expressons in the search engine.
45 * Revision 1.9 1995/10/10 12:24:39 adam
46 * Temporary sort files are compressed.
48 * Revision 1.8 1995/10/04 16:57:19 adam
49 * Key input and merge sort in one pass.
51 * Revision 1.7 1995/10/02 15:18:52 adam
52 * New member in recRetrieveCtrl: diagnostic.
54 * Revision 1.6 1995/09/29 15:51:56 adam
55 * First work on multi-way read.
57 * Revision 1.5 1995/09/29 14:01:43 adam
60 * Revision 1.4 1995/09/28 14:22:57 adam
61 * Sort uses smaller temporary files.
63 * Revision 1.3 1995/09/06 16:11:17 adam
64 * Option: only one word key per file.
66 * Revision 1.2 1995/09/04 12:33:42 adam
67 * Various cleanup. YAZ util used instead.
69 * Revision 1.1 1995/09/04 09:10:37 adam
70 * More work on index add/del/update.
71 * Merge sort implemented.
72 * Initial work on z39 server.
85 #define KEY_SIZE (1+sizeof(struct it_key))
86 #define INP_NAME_MAX 768
87 #define INP_BUF_START 60000
88 #define INP_BUF_ADD 400000
90 static int no_diffs = 0;
91 static int no_updates = 0;
92 static int no_deletions = 0;
93 static int no_insertions = 0;
94 static int no_iterations = 0;
98 off_t offset; /* file offset */
99 unsigned char *buf; /* buffer block */
100 size_t buf_size; /* number of read bytes in block */
101 size_t chunk; /* number of bytes allocated */
102 size_t buf_ptr; /* current position in buffer */
103 char *prev_name; /* last word read */
104 int sysno; /* last sysno */
105 int seqno; /* last seqno */
106 off_t length; /* length of file */
107 /* handler invoked in each read */
108 void (*readHandler)(struct key_file *keyp, void *rinfo);
112 void getFnameTmp (char *fname, int no)
116 pre = res_get_def (common_resource, "keyTmpDir", ".");
117 sprintf (fname, "%s/key%d.tmp", pre, no);
120 void key_file_chunk_read (struct key_file *f)
124 getFnameTmp (fname, f->no);
125 fd = open (fname, O_RDONLY);
128 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
133 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
135 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
139 if (lseek (fd, f->offset, SEEK_SET) == -1)
141 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
144 while (f->chunk - nr > 0)
146 r = read (fd, f->buf + nr, f->chunk - nr);
153 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
159 (*f->readHandler)(f, f->readInfo);
163 struct key_file *key_file_init (int no, int chunk)
167 f = xmalloc (sizeof(*f));
174 f->readHandler = NULL;
175 f->buf = xmalloc (f->chunk);
176 f->prev_name = xmalloc (INP_NAME_MAX);
177 *f->prev_name = '\0';
178 key_file_chunk_read (f);
182 int key_file_getc (struct key_file *f)
184 if (f->buf_ptr < f->buf_size)
185 return f->buf[(f->buf_ptr)++];
186 if (f->buf_size < f->chunk)
188 f->offset += f->buf_size;
189 key_file_chunk_read (f);
190 if (f->buf_ptr < f->buf_size)
191 return f->buf[(f->buf_ptr)++];
196 int key_file_decode (struct key_file *f)
200 c = key_file_getc (f);
207 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
210 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
211 d = (d << 8) + (key_file_getc (f) & 0xff);
214 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
215 d = (d << 8) + (key_file_getc (f) & 0xff);
216 d = (d << 8) + (key_file_getc (f) & 0xff);
222 int key_file_read (struct key_file *f, char *key)
227 c = key_file_getc (f);
230 strcpy (key, f->prev_name);
239 while ((key[i++] = key_file_getc (f)))
241 strcpy (f->prev_name, key);
244 d = key_file_decode (f);
247 itkey.sysno = d + f->sysno;
250 f->sysno = itkey.sysno;
253 d = key_file_decode (f);
254 itkey.seqno = d + f->seqno;
255 f->seqno = itkey.seqno;
256 memcpy (key + i, &itkey, sizeof(struct it_key));
257 return i + sizeof (struct it_key);
262 struct key_file **file;
267 int (*cmp)(const void *p1, const void *p2);
273 struct heap_info *key_heap_init (int nkeys,
274 int (*cmp)(const void *p1, const void *p2))
276 struct heap_info *hi;
279 hi = xmalloc (sizeof(*hi));
280 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
281 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
283 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
285 for (i = 0; i<= nkeys; i++)
288 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
293 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
298 hi->ptr[i1] = hi->ptr[i2];
303 static void key_heap_delete (struct heap_info *hi)
305 int cur = 1, child = 2;
307 assert (hi->heapnum > 0);
309 key_heap_swap (hi, 1, hi->heapnum);
311 while (child <= hi->heapnum) {
312 if (child < hi->heapnum &&
313 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
314 &hi->info.buf[hi->ptr[child+1]]) > 0)
316 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
317 &hi->info.buf[hi->ptr[child]]) > 0)
319 key_heap_swap (hi, cur, child);
328 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
333 cur = ++(hi->heapnum);
334 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
335 hi->info.file[hi->ptr[cur]] = kf;
338 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
339 &hi->info.buf[hi->ptr[cur]]) > 0)
341 key_heap_swap (hi, cur, parent);
347 static int heap_read_one (struct heap_info *hi, char *name, char *key)
350 char rbuf[INP_NAME_MAX];
356 strcpy (name, hi->info.buf[n]);
357 kf = hi->info.file[n];
359 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
360 key_heap_delete (hi);
361 if ((r = key_file_read (kf, rbuf)))
362 key_heap_insert (hi, rbuf, r, kf);
367 struct heap_cread_info {
368 char prev_name[INP_NAME_MAX];
369 char cur_name[INP_NAME_MAX];
371 struct heap_info *hi;
376 int heap_cread_item (void *vp, char **dst, int *insertMode)
378 struct heap_cread_info *p = vp;
379 struct heap_info *hi = p->hi;
383 *insertMode = p->key[0];
384 memcpy (*dst, p->key+1, sizeof(struct it_key));
385 (*dst) += sizeof(struct it_key);
389 strcpy (p->prev_name, p->cur_name);
390 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
392 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
397 *insertMode = p->key[0];
398 memcpy (*dst, p->key+1, sizeof(struct it_key));
399 (*dst) += sizeof(struct it_key);
403 int heap_inpc (struct heap_info *hi)
405 struct heap_cread_info hci;
406 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
408 hci.key = xmalloc (KEY_SIZE);
411 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
413 isamc_i->clientData = &hci;
414 isamc_i->read_item = heap_cread_item;
418 char this_name[INP_NAME_MAX];
419 ISAMC_P isamc_p, isamc_p2;
422 strcpy (this_name, hci.cur_name);
423 logf (LOG_DEBUG, "inserting %s", 1+hci.cur_name);
424 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
426 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
427 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
431 if (!dict_delete (hi->dict, this_name))
437 if (isamc_p2 != isamc_p)
438 dict_insert (hi->dict, this_name,
439 sizeof(ISAMC_P), &isamc_p2);
445 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
447 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
454 int heap_inp (struct heap_info *hi)
457 char next_name[INP_NAME_MAX];
458 char cur_name[INP_NAME_MAX];
459 int key_buf_size = INP_BUF_START;
465 next_key = xmalloc (KEY_SIZE);
466 key_buf = xmalloc (key_buf_size);
467 more = heap_read_one (hi, cur_name, key_buf);
468 while (more) /* EOF ? */
471 key_buf_ptr = KEY_SIZE;
474 if (!(more = heap_read_one (hi, next_name, next_key)))
476 if (*next_name && strcmp (next_name, cur_name))
478 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
479 key_buf_ptr += KEY_SIZE;
480 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
483 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
484 memcpy (new_key_buf, key_buf, key_buf_size);
485 key_buf_size += INP_BUF_ADD;
487 key_buf = new_key_buf;
491 nmemb = key_buf_ptr / KEY_SIZE;
492 assert (nmemb*KEY_SIZE == key_buf_ptr);
493 if ((info = dict_lookup (hi->dict, cur_name)))
495 ISAM_P isam_p, isam_p2;
496 logf (LOG_DEBUG, "updating %s", 1+cur_name);
497 memcpy (&isam_p, info+1, sizeof(ISAM_P));
498 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
502 if (!dict_delete (hi->dict, cur_name))
508 if (isam_p2 != isam_p)
509 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
515 logf (LOG_DEBUG, "inserting %s", 1+cur_name);
517 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
518 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
520 memcpy (key_buf, next_key, KEY_SIZE);
521 strcpy (cur_name, next_name);
526 struct progressInfo {
533 void progressFunc (struct key_file *keyp, void *info)
535 struct progressInfo *p = info;
536 time_t now, remaining;
538 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
542 if (now >= p->lastTime+10)
545 remaining = (now - p->startTime)*
546 ((double) p->totalBytes/p->totalOffset - 1.0);
547 if (remaining <= 130)
548 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
549 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
551 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
552 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
554 p->totalOffset += keyp->buf_size;
557 void key_input (int nkeys, int cache)
563 struct key_file **kf;
566 struct heap_info *hi;
567 struct progressInfo progressInfo;
575 getFnameTmp (fname, nkeys+1);
576 if (access (fname, R_OK) == -1)
583 dict = dict_open (FNAME_DICT, cache, 1);
586 logf (LOG_FATAL, "dict_open fail");
589 if (res_get_match (common_resource, "isam", "c", NULL))
591 isamc = isc_open (FNAME_ISAMC, 1, key_isamc_m ());
594 logf (LOG_FATAL, "isc_open fail");
600 isam = is_open (FNAME_ISAM, key_compare, 1, sizeof(struct it_key));
603 logf (LOG_FATAL, "is_open fail");
607 kf = xmalloc ((1+nkeys) * sizeof(*kf));
608 progressInfo.totalBytes = 0;
609 progressInfo.totalOffset = 0;
610 time (&progressInfo.startTime);
611 time (&progressInfo.lastTime);
612 for (i = 1; i<=nkeys; i++)
614 kf[i] = key_file_init (i, 32768);
615 kf[i]->readHandler = progressFunc;
616 kf[i]->readInfo = &progressInfo;
617 progressInfo.totalBytes += kf[i]->length;
618 progressInfo.totalOffset += kf[i]->buf_size;
620 hi = key_heap_init (nkeys, key_qsort_compare);
625 for (i = 1; i<=nkeys; i++)
626 if ((r = key_file_read (kf[i], rbuf)))
627 key_heap_insert (hi, rbuf, r, kf[i]);
638 for (i = 1; i<=nkeys; i++)
640 getFnameTmp (rbuf, i);
643 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
644 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
645 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
646 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
647 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);