2 * Copyright (C) 1994-1998, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.26 1998-01-29 13:39:13 adam
8 * Compress ISAM is default.
10 * Revision 1.25 1997/09/17 12:19:14 adam
11 * Zebra version corresponds to YAZ version 1.4.
12 * Changed Zebra server so that it doesn't depend on global common_resource.
14 * Revision 1.24 1997/09/09 13:38:07 adam
15 * Partial port to WIN95/NT.
17 * Revision 1.23 1997/09/04 13:57:39 adam
18 * Added O_BINARY for open calls.
20 * Revision 1.22 1997/02/12 20:39:45 adam
21 * Implemented options -f <n> that limits the log to the first <n>
23 * Changed some log messages also.
25 * Revision 1.21 1996/11/08 11:10:23 adam
26 * Buffers used during file match got bigger.
27 * Compressed ISAM support everywhere.
28 * Bug fixes regarding masking characters in queries.
29 * Redesigned Regexp-2 queries.
31 * Revision 1.20 1996/11/01 08:58:41 adam
32 * Interface to isamc system now includes update and delete.
34 * Revision 1.19 1996/10/29 14:09:46 adam
35 * Use of cisam system - enabled if setting isamc is 1.
37 * Revision 1.18 1996/06/04 10:18:59 adam
38 * Minor changes - removed include of ctype.h.
40 * Revision 1.17 1996/05/14 15:47:07 adam
41 * Cleanup of various buffer size entities.
43 * Revision 1.16 1996/04/09 10:05:20 adam
44 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
46 * Revision 1.15 1996/03/21 14:50:09 adam
47 * File update uses modify-time instead of change-time.
49 * Revision 1.14 1996/02/07 14:06:37 adam
50 * Better progress report during register merge.
51 * New command: clean - removes temporary shadow files.
53 * Revision 1.13 1996/02/05 12:30:00 adam
54 * Logging reduced a bit.
55 * The remaining running time is estimated during register merge.
57 * Revision 1.12 1995/12/06 17:49:19 adam
58 * Uses dict_delete now.
60 * Revision 1.11 1995/12/06 16:06:43 adam
61 * Better diagnostics. Work on 'real' dictionary deletion.
63 * Revision 1.10 1995/12/06 12:41:22 adam
64 * New command 'stat' for the index program.
65 * Filenames can be read from stdin by specifying '-'.
66 * Bug fix/enhancement of the transformation from terms to regular
67 * expressons in the search engine.
69 * Revision 1.9 1995/10/10 12:24:39 adam
70 * Temporary sort files are compressed.
72 * Revision 1.8 1995/10/04 16:57:19 adam
73 * Key input and merge sort in one pass.
75 * Revision 1.7 1995/10/02 15:18:52 adam
76 * New member in recRetrieveCtrl: diagnostic.
78 * Revision 1.6 1995/09/29 15:51:56 adam
79 * First work on multi-way read.
81 * Revision 1.5 1995/09/29 14:01:43 adam
84 * Revision 1.4 1995/09/28 14:22:57 adam
85 * Sort uses smaller temporary files.
87 * Revision 1.3 1995/09/06 16:11:17 adam
88 * Option: only one word key per file.
90 * Revision 1.2 1995/09/04 12:33:42 adam
91 * Various cleanup. YAZ util used instead.
93 * Revision 1.1 1995/09/04 09:10:37 adam
94 * More work on index add/del/update.
95 * Merge sort implemented.
96 * Initial work on z39 server.
113 #define KEY_SIZE (1+sizeof(struct it_key))
114 #define INP_NAME_MAX 768
115 #define INP_BUF_START 60000
116 #define INP_BUF_ADD 400000
118 static int no_diffs = 0;
119 static int no_updates = 0;
120 static int no_deletions = 0;
121 static int no_insertions = 0;
122 static int no_iterations = 0;
125 int no; /* file no */
126 off_t offset; /* file offset */
127 unsigned char *buf; /* buffer block */
128 size_t buf_size; /* number of read bytes in block */
129 size_t chunk; /* number of bytes allocated */
130 size_t buf_ptr; /* current position in buffer */
131 char *prev_name; /* last word read */
132 int sysno; /* last sysno */
133 int seqno; /* last seqno */
134 off_t length; /* length of file */
135 /* handler invoked in each read */
136 void (*readHandler)(struct key_file *keyp, void *rinfo);
140 void getFnameTmp (char *fname, int no)
144 pre = res_get_def (common_resource, "keyTmpDir", ".");
145 sprintf (fname, "%s/key%d.tmp", pre, no);
148 void key_file_chunk_read (struct key_file *f)
152 getFnameTmp (fname, f->no);
153 fd = open (fname, O_BINARY|O_RDONLY);
156 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
161 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
163 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
167 if (lseek (fd, f->offset, SEEK_SET) == -1)
169 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
172 while (f->chunk - nr > 0)
174 r = read (fd, f->buf + nr, f->chunk - nr);
181 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
187 (*f->readHandler)(f, f->readInfo);
191 struct key_file *key_file_init (int no, int chunk)
195 f = xmalloc (sizeof(*f));
202 f->readHandler = NULL;
203 f->buf = xmalloc (f->chunk);
204 f->prev_name = xmalloc (INP_NAME_MAX);
205 *f->prev_name = '\0';
206 key_file_chunk_read (f);
210 int key_file_getc (struct key_file *f)
212 if (f->buf_ptr < f->buf_size)
213 return f->buf[(f->buf_ptr)++];
214 if (f->buf_size < f->chunk)
216 f->offset += f->buf_size;
217 key_file_chunk_read (f);
218 if (f->buf_ptr < f->buf_size)
219 return f->buf[(f->buf_ptr)++];
224 int key_file_decode (struct key_file *f)
228 c = key_file_getc (f);
235 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
238 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
239 d = (d << 8) + (key_file_getc (f) & 0xff);
242 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
243 d = (d << 8) + (key_file_getc (f) & 0xff);
244 d = (d << 8) + (key_file_getc (f) & 0xff);
250 int key_file_read (struct key_file *f, char *key)
255 c = key_file_getc (f);
258 strcpy (key, f->prev_name);
267 while ((key[i++] = key_file_getc (f)))
269 strcpy (f->prev_name, key);
272 d = key_file_decode (f);
275 itkey.sysno = d + f->sysno;
278 f->sysno = itkey.sysno;
281 d = key_file_decode (f);
282 itkey.seqno = d + f->seqno;
283 f->seqno = itkey.seqno;
284 memcpy (key + i, &itkey, sizeof(struct it_key));
285 return i + sizeof (struct it_key);
290 struct key_file **file;
295 int (*cmp)(const void *p1, const void *p2);
301 struct heap_info *key_heap_init (int nkeys,
302 int (*cmp)(const void *p1, const void *p2))
304 struct heap_info *hi;
307 hi = xmalloc (sizeof(*hi));
308 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
309 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
311 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
313 for (i = 0; i<= nkeys; i++)
316 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
321 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
326 hi->ptr[i1] = hi->ptr[i2];
331 static void key_heap_delete (struct heap_info *hi)
333 int cur = 1, child = 2;
335 assert (hi->heapnum > 0);
337 key_heap_swap (hi, 1, hi->heapnum);
339 while (child <= hi->heapnum) {
340 if (child < hi->heapnum &&
341 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
342 &hi->info.buf[hi->ptr[child+1]]) > 0)
344 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
345 &hi->info.buf[hi->ptr[child]]) > 0)
347 key_heap_swap (hi, cur, child);
356 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
361 cur = ++(hi->heapnum);
362 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
363 hi->info.file[hi->ptr[cur]] = kf;
366 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
367 &hi->info.buf[hi->ptr[cur]]) > 0)
369 key_heap_swap (hi, cur, parent);
375 static int heap_read_one (struct heap_info *hi, char *name, char *key)
378 char rbuf[INP_NAME_MAX];
384 strcpy (name, hi->info.buf[n]);
385 kf = hi->info.file[n];
387 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
388 key_heap_delete (hi);
389 if ((r = key_file_read (kf, rbuf)))
390 key_heap_insert (hi, rbuf, r, kf);
395 struct heap_cread_info {
396 char prev_name[INP_NAME_MAX];
397 char cur_name[INP_NAME_MAX];
399 struct heap_info *hi;
404 int heap_cread_item (void *vp, char **dst, int *insertMode)
406 struct heap_cread_info *p = vp;
407 struct heap_info *hi = p->hi;
411 *insertMode = p->key[0];
412 memcpy (*dst, p->key+1, sizeof(struct it_key));
413 (*dst) += sizeof(struct it_key);
417 strcpy (p->prev_name, p->cur_name);
418 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
420 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
425 *insertMode = p->key[0];
426 memcpy (*dst, p->key+1, sizeof(struct it_key));
427 (*dst) += sizeof(struct it_key);
431 int heap_inpc (struct heap_info *hi)
433 struct heap_cread_info hci;
434 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
436 hci.key = xmalloc (KEY_SIZE);
439 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
441 isamc_i->clientData = &hci;
442 isamc_i->read_item = heap_cread_item;
446 char this_name[INP_NAME_MAX];
447 ISAMC_P isamc_p, isamc_p2;
450 strcpy (this_name, hci.cur_name);
451 logf (LOG_DEBUG, "inserting %s", 1+hci.cur_name);
453 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
455 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
456 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
460 if (!dict_delete (hi->dict, this_name))
466 if (isamc_p2 != isamc_p)
467 dict_insert (hi->dict, this_name,
468 sizeof(ISAMC_P), &isamc_p2);
473 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
475 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
482 int heap_inp (struct heap_info *hi)
485 char next_name[INP_NAME_MAX];
486 char cur_name[INP_NAME_MAX];
487 int key_buf_size = INP_BUF_START;
493 next_key = xmalloc (KEY_SIZE);
494 key_buf = xmalloc (key_buf_size);
495 more = heap_read_one (hi, cur_name, key_buf);
496 while (more) /* EOF ? */
499 key_buf_ptr = KEY_SIZE;
502 if (!(more = heap_read_one (hi, next_name, next_key)))
504 if (*next_name && strcmp (next_name, cur_name))
506 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
507 key_buf_ptr += KEY_SIZE;
508 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
511 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
512 memcpy (new_key_buf, key_buf, key_buf_size);
513 key_buf_size += INP_BUF_ADD;
515 key_buf = new_key_buf;
519 nmemb = key_buf_ptr / KEY_SIZE;
520 assert (nmemb*KEY_SIZE == key_buf_ptr);
521 if ((info = dict_lookup (hi->dict, cur_name)))
523 ISAM_P isam_p, isam_p2;
524 logf (LOG_DEBUG, "updating %s", 1+cur_name);
525 memcpy (&isam_p, info+1, sizeof(ISAM_P));
526 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
530 if (!dict_delete (hi->dict, cur_name))
536 if (isam_p2 != isam_p)
537 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
543 logf (LOG_DEBUG, "inserting %s", 1+cur_name);
545 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
546 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
548 memcpy (key_buf, next_key, KEY_SIZE);
549 strcpy (cur_name, next_name);
554 struct progressInfo {
561 void progressFunc (struct key_file *keyp, void *info)
563 struct progressInfo *p = info;
564 time_t now, remaining;
566 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
570 if (now >= p->lastTime+10)
573 remaining = (now - p->startTime)*
574 ((double) p->totalBytes/p->totalOffset - 1.0);
575 if (remaining <= 130)
576 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
577 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
579 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
580 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
582 p->totalOffset += keyp->buf_size;
589 void key_input (BFiles bfs, int nkeys, int cache)
595 struct key_file **kf;
598 struct heap_info *hi;
599 struct progressInfo progressInfo;
607 getFnameTmp (fname, nkeys+1);
608 if (access (fname, R_OK) == -1)
615 dict = dict_open (bfs, FNAME_DICT, cache, 1);
618 logf (LOG_FATAL, "dict_open fail");
621 if (!res_get_match (common_resource, "isam", "i", NULL))
623 isamc = isc_open (bfs,
624 FNAME_ISAMC, 1, key_isamc_m (common_resource));
627 logf (LOG_FATAL, "isc_open fail");
633 isam = is_open (bfs, FNAME_ISAM, key_compare, 1,
634 sizeof(struct it_key), common_resource);
637 logf (LOG_FATAL, "is_open fail");
641 kf = xmalloc ((1+nkeys) * sizeof(*kf));
642 progressInfo.totalBytes = 0;
643 progressInfo.totalOffset = 0;
644 time (&progressInfo.startTime);
645 time (&progressInfo.lastTime);
646 for (i = 1; i<=nkeys; i++)
648 kf[i] = key_file_init (i, 32768);
649 kf[i]->readHandler = progressFunc;
650 kf[i]->readInfo = &progressInfo;
651 progressInfo.totalBytes += kf[i]->length;
652 progressInfo.totalOffset += kf[i]->buf_size;
654 hi = key_heap_init (nkeys, key_qsort_compare);
659 for (i = 1; i<=nkeys; i++)
660 if ((r = key_file_read (kf[i], rbuf)))
661 key_heap_insert (hi, rbuf, r, kf[i]);
672 for (i = 1; i<=nkeys; i++)
674 getFnameTmp (rbuf, i);
677 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
678 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
679 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
680 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
681 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);