2 * Copyright (C) 1994-1998, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.30 1998-10-28 10:53:57 adam
8 * Added type cast to prevent warning.
10 * Revision 1.29 1998/06/11 15:41:39 adam
13 * Revision 1.28 1998/03/05 08:45:12 adam
14 * New result set model and modular ranking system. Moved towards
15 * descent server API. System information stored as "SGML" records.
17 * Revision 1.27 1998/02/17 10:32:52 adam
18 * Fixed bug: binary files weren't opened with flag b on NT.
20 * Revision 1.26 1998/01/29 13:39:13 adam
21 * Compress ISAM is default.
23 * Revision 1.25 1997/09/17 12:19:14 adam
24 * Zebra version corresponds to YAZ version 1.4.
25 * Changed Zebra server so that it doesn't depend on global common_resource.
27 * Revision 1.24 1997/09/09 13:38:07 adam
28 * Partial port to WIN95/NT.
30 * Revision 1.23 1997/09/04 13:57:39 adam
31 * Added O_BINARY for open calls.
33 * Revision 1.22 1997/02/12 20:39:45 adam
34 * Implemented options -f <n> that limits the log to the first <n>
36 * Changed some log messages also.
38 * Revision 1.21 1996/11/08 11:10:23 adam
39 * Buffers used during file match got bigger.
40 * Compressed ISAM support everywhere.
41 * Bug fixes regarding masking characters in queries.
42 * Redesigned Regexp-2 queries.
44 * Revision 1.20 1996/11/01 08:58:41 adam
45 * Interface to isamc system now includes update and delete.
47 * Revision 1.19 1996/10/29 14:09:46 adam
48 * Use of cisam system - enabled if setting isamc is 1.
50 * Revision 1.18 1996/06/04 10:18:59 adam
51 * Minor changes - removed include of ctype.h.
53 * Revision 1.17 1996/05/14 15:47:07 adam
54 * Cleanup of various buffer size entities.
56 * Revision 1.16 1996/04/09 10:05:20 adam
57 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
59 * Revision 1.15 1996/03/21 14:50:09 adam
60 * File update uses modify-time instead of change-time.
62 * Revision 1.14 1996/02/07 14:06:37 adam
63 * Better progress report during register merge.
64 * New command: clean - removes temporary shadow files.
66 * Revision 1.13 1996/02/05 12:30:00 adam
67 * Logging reduced a bit.
68 * The remaining running time is estimated during register merge.
70 * Revision 1.12 1995/12/06 17:49:19 adam
71 * Uses dict_delete now.
73 * Revision 1.11 1995/12/06 16:06:43 adam
74 * Better diagnostics. Work on 'real' dictionary deletion.
76 * Revision 1.10 1995/12/06 12:41:22 adam
77 * New command 'stat' for the index program.
78 * Filenames can be read from stdin by specifying '-'.
79 * Bug fix/enhancement of the transformation from terms to regular
80 * expressons in the search engine.
82 * Revision 1.9 1995/10/10 12:24:39 adam
83 * Temporary sort files are compressed.
85 * Revision 1.8 1995/10/04 16:57:19 adam
86 * Key input and merge sort in one pass.
88 * Revision 1.7 1995/10/02 15:18:52 adam
89 * New member in recRetrieveCtrl: diagnostic.
91 * Revision 1.6 1995/09/29 15:51:56 adam
92 * First work on multi-way read.
94 * Revision 1.5 1995/09/29 14:01:43 adam
97 * Revision 1.4 1995/09/28 14:22:57 adam
98 * Sort uses smaller temporary files.
100 * Revision 1.3 1995/09/06 16:11:17 adam
101 * Option: only one word key per file.
103 * Revision 1.2 1995/09/04 12:33:42 adam
104 * Various cleanup. YAZ util used instead.
106 * Revision 1.1 1995/09/04 09:10:37 adam
107 * More work on index add/del/update.
108 * Merge sort implemented.
109 * Initial work on z39 server.
126 #define KEY_SIZE (1+sizeof(struct it_key))
127 #define INP_NAME_MAX 768
128 #define INP_BUF_START 60000
129 #define INP_BUF_ADD 400000
131 static int no_diffs = 0;
132 static int no_updates = 0;
133 static int no_deletions = 0;
134 static int no_insertions = 0;
135 static int no_iterations = 0;
138 int no; /* file no */
139 off_t offset; /* file offset */
140 unsigned char *buf; /* buffer block */
141 size_t buf_size; /* number of read bytes in block */
142 size_t chunk; /* number of bytes allocated */
143 size_t buf_ptr; /* current position in buffer */
144 char *prev_name; /* last word read */
145 int sysno; /* last sysno */
146 int seqno; /* last seqno */
147 off_t length; /* length of file */
148 /* handler invoked in each read */
149 void (*readHandler)(struct key_file *keyp, void *rinfo);
153 void getFnameTmp (char *fname, int no)
157 pre = res_get_def (common_resource, "keyTmpDir", ".");
158 sprintf (fname, "%s/key%d.tmp", pre, no);
161 void key_file_chunk_read (struct key_file *f)
163 int nr = 0, r = 0, fd;
165 getFnameTmp (fname, f->no);
166 fd = open (fname, O_BINARY|O_RDONLY);
169 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
174 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
176 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
180 if (lseek (fd, f->offset, SEEK_SET) == -1)
182 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
185 while (f->chunk - nr > 0)
187 r = read (fd, f->buf + nr, f->chunk - nr);
194 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
200 (*f->readHandler)(f, f->readInfo);
204 struct key_file *key_file_init (int no, int chunk)
208 f = xmalloc (sizeof(*f));
215 f->readHandler = NULL;
216 f->buf = xmalloc (f->chunk);
217 f->prev_name = xmalloc (INP_NAME_MAX);
218 *f->prev_name = '\0';
219 key_file_chunk_read (f);
223 int key_file_getc (struct key_file *f)
225 if (f->buf_ptr < f->buf_size)
226 return f->buf[(f->buf_ptr)++];
227 if (f->buf_size < f->chunk)
229 f->offset += f->buf_size;
230 key_file_chunk_read (f);
231 if (f->buf_ptr < f->buf_size)
232 return f->buf[(f->buf_ptr)++];
237 int key_file_decode (struct key_file *f)
241 c = key_file_getc (f);
248 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
251 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
252 d = (d << 8) + (key_file_getc (f) & 0xff);
255 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
256 d = (d << 8) + (key_file_getc (f) & 0xff);
257 d = (d << 8) + (key_file_getc (f) & 0xff);
263 int key_file_read (struct key_file *f, char *key)
268 c = key_file_getc (f);
271 strcpy (key, f->prev_name);
280 while ((key[i++] = key_file_getc (f)))
282 strcpy (f->prev_name, key);
285 d = key_file_decode (f);
288 itkey.sysno = d + f->sysno;
291 f->sysno = itkey.sysno;
294 d = key_file_decode (f);
295 itkey.seqno = d + f->seqno;
296 f->seqno = itkey.seqno;
297 memcpy (key + i, &itkey, sizeof(struct it_key));
298 return i + sizeof (struct it_key);
303 struct key_file **file;
308 int (*cmp)(const void *p1, const void *p2);
314 struct heap_info *key_heap_init (int nkeys,
315 int (*cmp)(const void *p1, const void *p2))
317 struct heap_info *hi;
320 hi = xmalloc (sizeof(*hi));
321 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
322 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
324 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
326 for (i = 0; i<= nkeys; i++)
329 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
334 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
339 hi->ptr[i1] = hi->ptr[i2];
344 static void key_heap_delete (struct heap_info *hi)
346 int cur = 1, child = 2;
348 assert (hi->heapnum > 0);
350 key_heap_swap (hi, 1, hi->heapnum);
352 while (child <= hi->heapnum) {
353 if (child < hi->heapnum &&
354 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
355 &hi->info.buf[hi->ptr[child+1]]) > 0)
357 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
358 &hi->info.buf[hi->ptr[child]]) > 0)
360 key_heap_swap (hi, cur, child);
369 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
374 cur = ++(hi->heapnum);
375 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
376 hi->info.file[hi->ptr[cur]] = kf;
379 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
380 &hi->info.buf[hi->ptr[cur]]) > 0)
382 key_heap_swap (hi, cur, parent);
388 static int heap_read_one (struct heap_info *hi, char *name, char *key)
391 char rbuf[INP_NAME_MAX];
397 strcpy (name, hi->info.buf[n]);
398 kf = hi->info.file[n];
400 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
401 key_heap_delete (hi);
402 if ((r = key_file_read (kf, rbuf)))
403 key_heap_insert (hi, rbuf, r, kf);
408 struct heap_cread_info {
409 char prev_name[INP_NAME_MAX];
410 char cur_name[INP_NAME_MAX];
412 struct heap_info *hi;
417 int heap_cread_item (void *vp, char **dst, int *insertMode)
419 struct heap_cread_info *p = vp;
420 struct heap_info *hi = p->hi;
424 *insertMode = p->key[0];
425 memcpy (*dst, p->key+1, sizeof(struct it_key));
426 (*dst) += sizeof(struct it_key);
430 strcpy (p->prev_name, p->cur_name);
431 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
433 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
438 *insertMode = p->key[0];
439 memcpy (*dst, p->key+1, sizeof(struct it_key));
440 (*dst) += sizeof(struct it_key);
444 int heap_inpc (struct heap_info *hi)
446 struct heap_cread_info hci;
447 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
449 hci.key = xmalloc (KEY_SIZE);
452 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
454 isamc_i->clientData = &hci;
455 isamc_i->read_item = heap_cread_item;
459 char this_name[INP_NAME_MAX];
460 ISAMC_P isamc_p, isamc_p2;
463 strcpy (this_name, hci.cur_name);
464 assert (hci.cur_name[1]);
466 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
468 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
469 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
473 if (!dict_delete (hi->dict, this_name))
479 if (isamc_p2 != isamc_p)
480 dict_insert (hi->dict, this_name,
481 sizeof(ISAMC_P), &isamc_p2);
486 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
488 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
495 int heap_inp (struct heap_info *hi)
498 char next_name[INP_NAME_MAX];
499 char cur_name[INP_NAME_MAX];
500 int key_buf_size = INP_BUF_START;
506 next_key = xmalloc (KEY_SIZE);
507 key_buf = xmalloc (key_buf_size);
508 more = heap_read_one (hi, cur_name, key_buf);
509 while (more) /* EOF ? */
512 key_buf_ptr = KEY_SIZE;
515 if (!(more = heap_read_one (hi, next_name, next_key)))
517 if (*next_name && strcmp (next_name, cur_name))
519 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
520 key_buf_ptr += KEY_SIZE;
521 if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size)
524 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
525 memcpy (new_key_buf, key_buf, key_buf_size);
526 key_buf_size += INP_BUF_ADD;
528 key_buf = new_key_buf;
532 nmemb = key_buf_ptr / KEY_SIZE;
533 assert (nmemb * (int) KEY_SIZE == key_buf_ptr);
534 if ((info = dict_lookup (hi->dict, cur_name)))
536 ISAM_P isam_p, isam_p2;
537 memcpy (&isam_p, info+1, sizeof(ISAM_P));
538 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
542 if (!dict_delete (hi->dict, cur_name))
548 if (isam_p2 != isam_p)
549 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
556 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
557 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
559 memcpy (key_buf, next_key, KEY_SIZE);
560 strcpy (cur_name, next_name);
565 struct progressInfo {
572 void progressFunc (struct key_file *keyp, void *info)
574 struct progressInfo *p = info;
575 time_t now, remaining;
577 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
581 if (now >= p->lastTime+10)
584 remaining = (time_t) ((now - p->startTime)*
585 ((double) p->totalBytes/p->totalOffset - 1.0));
586 if (remaining <= 130)
587 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
588 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
590 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
591 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
593 p->totalOffset += keyp->buf_size;
600 void key_input (BFiles bfs, int nkeys, int cache)
606 struct key_file **kf;
609 struct heap_info *hi;
610 struct progressInfo progressInfo;
618 getFnameTmp (fname, nkeys+1);
619 if (access (fname, R_OK) == -1)
626 dict = dict_open (bfs, FNAME_DICT, cache, 1);
629 logf (LOG_FATAL, "dict_open fail");
632 if (!res_get_match (common_resource, "isam", "i", NULL))
634 isamc = isc_open (bfs,
635 FNAME_ISAMC, 1, key_isamc_m (common_resource));
638 logf (LOG_FATAL, "isc_open fail");
644 isam = is_open (bfs, FNAME_ISAM, key_compare, 1,
645 sizeof(struct it_key), common_resource);
648 logf (LOG_FATAL, "is_open fail");
652 kf = xmalloc ((1+nkeys) * sizeof(*kf));
653 progressInfo.totalBytes = 0;
654 progressInfo.totalOffset = 0;
655 time (&progressInfo.startTime);
656 time (&progressInfo.lastTime);
657 for (i = 1; i<=nkeys; i++)
659 kf[i] = key_file_init (i, 32768);
660 kf[i]->readHandler = progressFunc;
661 kf[i]->readInfo = &progressInfo;
662 progressInfo.totalBytes += kf[i]->length;
663 progressInfo.totalOffset += kf[i]->buf_size;
665 hi = key_heap_init (nkeys, key_qsort_compare);
670 for (i = 1; i<=nkeys; i++)
671 if ((r = key_file_read (kf[i], rbuf)))
672 key_heap_insert (hi, rbuf, r, kf[i]);
683 for (i = 1; i<=nkeys; i++)
685 getFnameTmp (rbuf, i);
688 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
689 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
690 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
691 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
692 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);