2 * Copyright (C) 1994-1996, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.23 1997-09-04 13:57:39 adam
8 * Added O_BINARY for open calls.
10 * Revision 1.22 1997/02/12 20:39:45 adam
11 * Implemented options -f <n> that limits the log to the first <n>
13 * Changed some log messages also.
15 * Revision 1.21 1996/11/08 11:10:23 adam
16 * Buffers used during file match got bigger.
17 * Compressed ISAM support everywhere.
18 * Bug fixes regarding masking characters in queries.
19 * Redesigned Regexp-2 queries.
21 * Revision 1.20 1996/11/01 08:58:41 adam
22 * Interface to isamc system now includes update and delete.
24 * Revision 1.19 1996/10/29 14:09:46 adam
25 * Use of cisam system - enabled if setting isamc is 1.
27 * Revision 1.18 1996/06/04 10:18:59 adam
28 * Minor changes - removed include of ctype.h.
30 * Revision 1.17 1996/05/14 15:47:07 adam
31 * Cleanup of various buffer size entities.
33 * Revision 1.16 1996/04/09 10:05:20 adam
34 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
36 * Revision 1.15 1996/03/21 14:50:09 adam
37 * File update uses modify-time instead of change-time.
39 * Revision 1.14 1996/02/07 14:06:37 adam
40 * Better progress report during register merge.
41 * New command: clean - removes temporary shadow files.
43 * Revision 1.13 1996/02/05 12:30:00 adam
44 * Logging reduced a bit.
45 * The remaining running time is estimated during register merge.
47 * Revision 1.12 1995/12/06 17:49:19 adam
48 * Uses dict_delete now.
50 * Revision 1.11 1995/12/06 16:06:43 adam
51 * Better diagnostics. Work on 'real' dictionary deletion.
53 * Revision 1.10 1995/12/06 12:41:22 adam
54 * New command 'stat' for the index program.
55 * Filenames can be read from stdin by specifying '-'.
56 * Bug fix/enhancement of the transformation from terms to regular
57 * expressons in the search engine.
59 * Revision 1.9 1995/10/10 12:24:39 adam
60 * Temporary sort files are compressed.
62 * Revision 1.8 1995/10/04 16:57:19 adam
63 * Key input and merge sort in one pass.
65 * Revision 1.7 1995/10/02 15:18:52 adam
66 * New member in recRetrieveCtrl: diagnostic.
68 * Revision 1.6 1995/09/29 15:51:56 adam
69 * First work on multi-way read.
71 * Revision 1.5 1995/09/29 14:01:43 adam
74 * Revision 1.4 1995/09/28 14:22:57 adam
75 * Sort uses smaller temporary files.
77 * Revision 1.3 1995/09/06 16:11:17 adam
78 * Option: only one word key per file.
80 * Revision 1.2 1995/09/04 12:33:42 adam
81 * Various cleanup. YAZ util used instead.
83 * Revision 1.1 1995/09/04 09:10:37 adam
84 * More work on index add/del/update.
85 * Merge sort implemented.
86 * Initial work on z39 server.
99 #define KEY_SIZE (1+sizeof(struct it_key))
100 #define INP_NAME_MAX 768
101 #define INP_BUF_START 60000
102 #define INP_BUF_ADD 400000
104 static int no_diffs = 0;
105 static int no_updates = 0;
106 static int no_deletions = 0;
107 static int no_insertions = 0;
108 static int no_iterations = 0;
111 int no; /* file no */
112 off_t offset; /* file offset */
113 unsigned char *buf; /* buffer block */
114 size_t buf_size; /* number of read bytes in block */
115 size_t chunk; /* number of bytes allocated */
116 size_t buf_ptr; /* current position in buffer */
117 char *prev_name; /* last word read */
118 int sysno; /* last sysno */
119 int seqno; /* last seqno */
120 off_t length; /* length of file */
121 /* handler invoked in each read */
122 void (*readHandler)(struct key_file *keyp, void *rinfo);
126 void getFnameTmp (char *fname, int no)
130 pre = res_get_def (common_resource, "keyTmpDir", ".");
131 sprintf (fname, "%s/key%d.tmp", pre, no);
134 void key_file_chunk_read (struct key_file *f)
138 getFnameTmp (fname, f->no);
139 fd = open (fname, O_BINARY|O_RDONLY);
142 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
147 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
149 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
153 if (lseek (fd, f->offset, SEEK_SET) == -1)
155 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
158 while (f->chunk - nr > 0)
160 r = read (fd, f->buf + nr, f->chunk - nr);
167 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
173 (*f->readHandler)(f, f->readInfo);
177 struct key_file *key_file_init (int no, int chunk)
181 f = xmalloc (sizeof(*f));
188 f->readHandler = NULL;
189 f->buf = xmalloc (f->chunk);
190 f->prev_name = xmalloc (INP_NAME_MAX);
191 *f->prev_name = '\0';
192 key_file_chunk_read (f);
196 int key_file_getc (struct key_file *f)
198 if (f->buf_ptr < f->buf_size)
199 return f->buf[(f->buf_ptr)++];
200 if (f->buf_size < f->chunk)
202 f->offset += f->buf_size;
203 key_file_chunk_read (f);
204 if (f->buf_ptr < f->buf_size)
205 return f->buf[(f->buf_ptr)++];
210 int key_file_decode (struct key_file *f)
214 c = key_file_getc (f);
221 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
224 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
225 d = (d << 8) + (key_file_getc (f) & 0xff);
228 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
229 d = (d << 8) + (key_file_getc (f) & 0xff);
230 d = (d << 8) + (key_file_getc (f) & 0xff);
236 int key_file_read (struct key_file *f, char *key)
241 c = key_file_getc (f);
244 strcpy (key, f->prev_name);
253 while ((key[i++] = key_file_getc (f)))
255 strcpy (f->prev_name, key);
258 d = key_file_decode (f);
261 itkey.sysno = d + f->sysno;
264 f->sysno = itkey.sysno;
267 d = key_file_decode (f);
268 itkey.seqno = d + f->seqno;
269 f->seqno = itkey.seqno;
270 memcpy (key + i, &itkey, sizeof(struct it_key));
271 return i + sizeof (struct it_key);
276 struct key_file **file;
281 int (*cmp)(const void *p1, const void *p2);
287 struct heap_info *key_heap_init (int nkeys,
288 int (*cmp)(const void *p1, const void *p2))
290 struct heap_info *hi;
293 hi = xmalloc (sizeof(*hi));
294 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
295 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
297 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
299 for (i = 0; i<= nkeys; i++)
302 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
307 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
312 hi->ptr[i1] = hi->ptr[i2];
317 static void key_heap_delete (struct heap_info *hi)
319 int cur = 1, child = 2;
321 assert (hi->heapnum > 0);
323 key_heap_swap (hi, 1, hi->heapnum);
325 while (child <= hi->heapnum) {
326 if (child < hi->heapnum &&
327 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
328 &hi->info.buf[hi->ptr[child+1]]) > 0)
330 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
331 &hi->info.buf[hi->ptr[child]]) > 0)
333 key_heap_swap (hi, cur, child);
342 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
347 cur = ++(hi->heapnum);
348 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
349 hi->info.file[hi->ptr[cur]] = kf;
352 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
353 &hi->info.buf[hi->ptr[cur]]) > 0)
355 key_heap_swap (hi, cur, parent);
361 static int heap_read_one (struct heap_info *hi, char *name, char *key)
364 char rbuf[INP_NAME_MAX];
370 strcpy (name, hi->info.buf[n]);
371 kf = hi->info.file[n];
373 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
374 key_heap_delete (hi);
375 if ((r = key_file_read (kf, rbuf)))
376 key_heap_insert (hi, rbuf, r, kf);
381 struct heap_cread_info {
382 char prev_name[INP_NAME_MAX];
383 char cur_name[INP_NAME_MAX];
385 struct heap_info *hi;
390 int heap_cread_item (void *vp, char **dst, int *insertMode)
392 struct heap_cread_info *p = vp;
393 struct heap_info *hi = p->hi;
397 *insertMode = p->key[0];
398 memcpy (*dst, p->key+1, sizeof(struct it_key));
399 (*dst) += sizeof(struct it_key);
403 strcpy (p->prev_name, p->cur_name);
404 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
406 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
411 *insertMode = p->key[0];
412 memcpy (*dst, p->key+1, sizeof(struct it_key));
413 (*dst) += sizeof(struct it_key);
417 int heap_inpc (struct heap_info *hi)
419 struct heap_cread_info hci;
420 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
422 hci.key = xmalloc (KEY_SIZE);
425 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
427 isamc_i->clientData = &hci;
428 isamc_i->read_item = heap_cread_item;
432 char this_name[INP_NAME_MAX];
433 ISAMC_P isamc_p, isamc_p2;
436 strcpy (this_name, hci.cur_name);
437 logf (LOG_DEBUG, "inserting %s", 1+hci.cur_name);
439 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
441 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
442 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
446 if (!dict_delete (hi->dict, this_name))
452 if (isamc_p2 != isamc_p)
453 dict_insert (hi->dict, this_name,
454 sizeof(ISAMC_P), &isamc_p2);
459 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
461 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
468 int heap_inp (struct heap_info *hi)
471 char next_name[INP_NAME_MAX];
472 char cur_name[INP_NAME_MAX];
473 int key_buf_size = INP_BUF_START;
479 next_key = xmalloc (KEY_SIZE);
480 key_buf = xmalloc (key_buf_size);
481 more = heap_read_one (hi, cur_name, key_buf);
482 while (more) /* EOF ? */
485 key_buf_ptr = KEY_SIZE;
488 if (!(more = heap_read_one (hi, next_name, next_key)))
490 if (*next_name && strcmp (next_name, cur_name))
492 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
493 key_buf_ptr += KEY_SIZE;
494 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
497 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
498 memcpy (new_key_buf, key_buf, key_buf_size);
499 key_buf_size += INP_BUF_ADD;
501 key_buf = new_key_buf;
505 nmemb = key_buf_ptr / KEY_SIZE;
506 assert (nmemb*KEY_SIZE == key_buf_ptr);
507 if ((info = dict_lookup (hi->dict, cur_name)))
509 ISAM_P isam_p, isam_p2;
510 logf (LOG_DEBUG, "updating %s", 1+cur_name);
511 memcpy (&isam_p, info+1, sizeof(ISAM_P));
512 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
516 if (!dict_delete (hi->dict, cur_name))
522 if (isam_p2 != isam_p)
523 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
529 logf (LOG_DEBUG, "inserting %s", 1+cur_name);
531 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
532 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
534 memcpy (key_buf, next_key, KEY_SIZE);
535 strcpy (cur_name, next_name);
540 struct progressInfo {
547 void progressFunc (struct key_file *keyp, void *info)
549 struct progressInfo *p = info;
550 time_t now, remaining;
552 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
556 if (now >= p->lastTime+10)
559 remaining = (now - p->startTime)*
560 ((double) p->totalBytes/p->totalOffset - 1.0);
561 if (remaining <= 130)
562 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
563 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
565 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
566 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
568 p->totalOffset += keyp->buf_size;
571 void key_input (int nkeys, int cache)
577 struct key_file **kf;
580 struct heap_info *hi;
581 struct progressInfo progressInfo;
589 getFnameTmp (fname, nkeys+1);
590 if (access (fname, R_OK) == -1)
597 dict = dict_open (FNAME_DICT, cache, 1);
600 logf (LOG_FATAL, "dict_open fail");
603 if (res_get_match (common_resource, "isam", "c", NULL))
605 isamc = isc_open (FNAME_ISAMC, 1, key_isamc_m ());
608 logf (LOG_FATAL, "isc_open fail");
614 isam = is_open (FNAME_ISAM, key_compare, 1, sizeof(struct it_key));
617 logf (LOG_FATAL, "is_open fail");
621 kf = xmalloc ((1+nkeys) * sizeof(*kf));
622 progressInfo.totalBytes = 0;
623 progressInfo.totalOffset = 0;
624 time (&progressInfo.startTime);
625 time (&progressInfo.lastTime);
626 for (i = 1; i<=nkeys; i++)
628 kf[i] = key_file_init (i, 32768);
629 kf[i]->readHandler = progressFunc;
630 kf[i]->readInfo = &progressInfo;
631 progressInfo.totalBytes += kf[i]->length;
632 progressInfo.totalOffset += kf[i]->buf_size;
634 hi = key_heap_init (nkeys, key_qsort_compare);
639 for (i = 1; i<=nkeys; i++)
640 if ((r = key_file_read (kf[i], rbuf)))
641 key_heap_insert (hi, rbuf, r, kf[i]);
652 for (i = 1; i<=nkeys; i++)
654 getFnameTmp (rbuf, i);
657 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
658 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
659 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
660 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
661 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);