2 * Copyright (C) 1994-1999, Index Data
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.31 1999-02-02 14:50:56 adam
8 * Updated WIN32 code specific sections. Changed header.
10 * Revision 1.30 1998/10/28 10:53:57 adam
11 * Added type cast to prevent warning.
13 * Revision 1.29 1998/06/11 15:41:39 adam
16 * Revision 1.28 1998/03/05 08:45:12 adam
17 * New result set model and modular ranking system. Moved towards
18 * descent server API. System information stored as "SGML" records.
20 * Revision 1.27 1998/02/17 10:32:52 adam
21 * Fixed bug: binary files weren't opened with flag b on NT.
23 * Revision 1.26 1998/01/29 13:39:13 adam
24 * Compress ISAM is default.
26 * Revision 1.25 1997/09/17 12:19:14 adam
27 * Zebra version corresponds to YAZ version 1.4.
28 * Changed Zebra server so that it doesn't depend on global common_resource.
30 * Revision 1.24 1997/09/09 13:38:07 adam
31 * Partial port to WIN95/NT.
33 * Revision 1.23 1997/09/04 13:57:39 adam
34 * Added O_BINARY for open calls.
36 * Revision 1.22 1997/02/12 20:39:45 adam
37 * Implemented options -f <n> that limits the log to the first <n>
39 * Changed some log messages also.
41 * Revision 1.21 1996/11/08 11:10:23 adam
42 * Buffers used during file match got bigger.
43 * Compressed ISAM support everywhere.
44 * Bug fixes regarding masking characters in queries.
45 * Redesigned Regexp-2 queries.
47 * Revision 1.20 1996/11/01 08:58:41 adam
48 * Interface to isamc system now includes update and delete.
50 * Revision 1.19 1996/10/29 14:09:46 adam
51 * Use of cisam system - enabled if setting isamc is 1.
53 * Revision 1.18 1996/06/04 10:18:59 adam
54 * Minor changes - removed include of ctype.h.
56 * Revision 1.17 1996/05/14 15:47:07 adam
57 * Cleanup of various buffer size entities.
59 * Revision 1.16 1996/04/09 10:05:20 adam
60 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
62 * Revision 1.15 1996/03/21 14:50:09 adam
63 * File update uses modify-time instead of change-time.
65 * Revision 1.14 1996/02/07 14:06:37 adam
66 * Better progress report during register merge.
67 * New command: clean - removes temporary shadow files.
69 * Revision 1.13 1996/02/05 12:30:00 adam
70 * Logging reduced a bit.
71 * The remaining running time is estimated during register merge.
73 * Revision 1.12 1995/12/06 17:49:19 adam
74 * Uses dict_delete now.
76 * Revision 1.11 1995/12/06 16:06:43 adam
77 * Better diagnostics. Work on 'real' dictionary deletion.
79 * Revision 1.10 1995/12/06 12:41:22 adam
80 * New command 'stat' for the index program.
81 * Filenames can be read from stdin by specifying '-'.
82 * Bug fix/enhancement of the transformation from terms to regular
83 * expressons in the search engine.
85 * Revision 1.9 1995/10/10 12:24:39 adam
86 * Temporary sort files are compressed.
88 * Revision 1.8 1995/10/04 16:57:19 adam
89 * Key input and merge sort in one pass.
91 * Revision 1.7 1995/10/02 15:18:52 adam
92 * New member in recRetrieveCtrl: diagnostic.
94 * Revision 1.6 1995/09/29 15:51:56 adam
95 * First work on multi-way read.
97 * Revision 1.5 1995/09/29 14:01:43 adam
100 * Revision 1.4 1995/09/28 14:22:57 adam
101 * Sort uses smaller temporary files.
103 * Revision 1.3 1995/09/06 16:11:17 adam
104 * Option: only one word key per file.
106 * Revision 1.2 1995/09/04 12:33:42 adam
107 * Various cleanup. YAZ util used instead.
109 * Revision 1.1 1995/09/04 09:10:37 adam
110 * More work on index add/del/update.
111 * Merge sort implemented.
112 * Initial work on z39 server.
129 #define KEY_SIZE (1+sizeof(struct it_key))
130 #define INP_NAME_MAX 768
131 #define INP_BUF_START 60000
132 #define INP_BUF_ADD 400000
134 static int no_diffs = 0;
135 static int no_updates = 0;
136 static int no_deletions = 0;
137 static int no_insertions = 0;
138 static int no_iterations = 0;
141 int no; /* file no */
142 off_t offset; /* file offset */
143 unsigned char *buf; /* buffer block */
144 size_t buf_size; /* number of read bytes in block */
145 size_t chunk; /* number of bytes allocated */
146 size_t buf_ptr; /* current position in buffer */
147 char *prev_name; /* last word read */
148 int sysno; /* last sysno */
149 int seqno; /* last seqno */
150 off_t length; /* length of file */
151 /* handler invoked in each read */
152 void (*readHandler)(struct key_file *keyp, void *rinfo);
156 void getFnameTmp (char *fname, int no)
160 pre = res_get_def (common_resource, "keyTmpDir", ".");
161 sprintf (fname, "%s/key%d.tmp", pre, no);
164 void key_file_chunk_read (struct key_file *f)
166 int nr = 0, r = 0, fd;
168 getFnameTmp (fname, f->no);
169 fd = open (fname, O_BINARY|O_RDONLY);
172 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
177 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
179 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
183 if (lseek (fd, f->offset, SEEK_SET) == -1)
185 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
188 while (f->chunk - nr > 0)
190 r = read (fd, f->buf + nr, f->chunk - nr);
197 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
203 (*f->readHandler)(f, f->readInfo);
207 struct key_file *key_file_init (int no, int chunk)
211 f = xmalloc (sizeof(*f));
218 f->readHandler = NULL;
219 f->buf = xmalloc (f->chunk);
220 f->prev_name = xmalloc (INP_NAME_MAX);
221 *f->prev_name = '\0';
222 key_file_chunk_read (f);
226 int key_file_getc (struct key_file *f)
228 if (f->buf_ptr < f->buf_size)
229 return f->buf[(f->buf_ptr)++];
230 if (f->buf_size < f->chunk)
232 f->offset += f->buf_size;
233 key_file_chunk_read (f);
234 if (f->buf_ptr < f->buf_size)
235 return f->buf[(f->buf_ptr)++];
240 int key_file_decode (struct key_file *f)
244 c = key_file_getc (f);
251 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
254 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
255 d = (d << 8) + (key_file_getc (f) & 0xff);
258 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
259 d = (d << 8) + (key_file_getc (f) & 0xff);
260 d = (d << 8) + (key_file_getc (f) & 0xff);
266 int key_file_read (struct key_file *f, char *key)
271 c = key_file_getc (f);
274 strcpy (key, f->prev_name);
283 while ((key[i++] = key_file_getc (f)))
285 strcpy (f->prev_name, key);
288 d = key_file_decode (f);
291 itkey.sysno = d + f->sysno;
294 f->sysno = itkey.sysno;
297 d = key_file_decode (f);
298 itkey.seqno = d + f->seqno;
299 f->seqno = itkey.seqno;
300 memcpy (key + i, &itkey, sizeof(struct it_key));
301 return i + sizeof (struct it_key);
306 struct key_file **file;
311 int (*cmp)(const void *p1, const void *p2);
317 struct heap_info *key_heap_init (int nkeys,
318 int (*cmp)(const void *p1, const void *p2))
320 struct heap_info *hi;
323 hi = xmalloc (sizeof(*hi));
324 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
325 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
327 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
329 for (i = 0; i<= nkeys; i++)
332 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
337 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
342 hi->ptr[i1] = hi->ptr[i2];
347 static void key_heap_delete (struct heap_info *hi)
349 int cur = 1, child = 2;
351 assert (hi->heapnum > 0);
353 key_heap_swap (hi, 1, hi->heapnum);
355 while (child <= hi->heapnum) {
356 if (child < hi->heapnum &&
357 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
358 &hi->info.buf[hi->ptr[child+1]]) > 0)
360 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
361 &hi->info.buf[hi->ptr[child]]) > 0)
363 key_heap_swap (hi, cur, child);
372 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
377 cur = ++(hi->heapnum);
378 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
379 hi->info.file[hi->ptr[cur]] = kf;
382 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
383 &hi->info.buf[hi->ptr[cur]]) > 0)
385 key_heap_swap (hi, cur, parent);
391 static int heap_read_one (struct heap_info *hi, char *name, char *key)
394 char rbuf[INP_NAME_MAX];
400 strcpy (name, hi->info.buf[n]);
401 kf = hi->info.file[n];
403 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
404 key_heap_delete (hi);
405 if ((r = key_file_read (kf, rbuf)))
406 key_heap_insert (hi, rbuf, r, kf);
411 struct heap_cread_info {
412 char prev_name[INP_NAME_MAX];
413 char cur_name[INP_NAME_MAX];
415 struct heap_info *hi;
420 int heap_cread_item (void *vp, char **dst, int *insertMode)
422 struct heap_cread_info *p = vp;
423 struct heap_info *hi = p->hi;
427 *insertMode = p->key[0];
428 memcpy (*dst, p->key+1, sizeof(struct it_key));
429 (*dst) += sizeof(struct it_key);
433 strcpy (p->prev_name, p->cur_name);
434 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
436 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
441 *insertMode = p->key[0];
442 memcpy (*dst, p->key+1, sizeof(struct it_key));
443 (*dst) += sizeof(struct it_key);
447 int heap_inpc (struct heap_info *hi)
449 struct heap_cread_info hci;
450 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
452 hci.key = xmalloc (KEY_SIZE);
455 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
457 isamc_i->clientData = &hci;
458 isamc_i->read_item = heap_cread_item;
462 char this_name[INP_NAME_MAX];
463 ISAMC_P isamc_p, isamc_p2;
466 strcpy (this_name, hci.cur_name);
467 assert (hci.cur_name[1]);
469 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
471 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
472 isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i);
476 if (!dict_delete (hi->dict, this_name))
482 if (isamc_p2 != isamc_p)
483 dict_insert (hi->dict, this_name,
484 sizeof(ISAMC_P), &isamc_p2);
489 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
491 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
498 int heap_inp (struct heap_info *hi)
501 char next_name[INP_NAME_MAX];
502 char cur_name[INP_NAME_MAX];
503 int key_buf_size = INP_BUF_START;
509 next_key = xmalloc (KEY_SIZE);
510 key_buf = xmalloc (key_buf_size);
511 more = heap_read_one (hi, cur_name, key_buf);
512 while (more) /* EOF ? */
515 key_buf_ptr = KEY_SIZE;
518 if (!(more = heap_read_one (hi, next_name, next_key)))
520 if (*next_name && strcmp (next_name, cur_name))
522 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
523 key_buf_ptr += KEY_SIZE;
524 if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size)
527 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
528 memcpy (new_key_buf, key_buf, key_buf_size);
529 key_buf_size += INP_BUF_ADD;
531 key_buf = new_key_buf;
535 nmemb = key_buf_ptr / KEY_SIZE;
536 assert (nmemb * (int) KEY_SIZE == key_buf_ptr);
537 if ((info = dict_lookup (hi->dict, cur_name)))
539 ISAM_P isam_p, isam_p2;
540 memcpy (&isam_p, info+1, sizeof(ISAM_P));
541 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
545 if (!dict_delete (hi->dict, cur_name))
551 if (isam_p2 != isam_p)
552 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
559 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
560 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
562 memcpy (key_buf, next_key, KEY_SIZE);
563 strcpy (cur_name, next_name);
568 struct progressInfo {
575 void progressFunc (struct key_file *keyp, void *info)
577 struct progressInfo *p = info;
578 time_t now, remaining;
580 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
584 if (now >= p->lastTime+10)
587 remaining = (time_t) ((now - p->startTime)*
588 ((double) p->totalBytes/p->totalOffset - 1.0));
589 if (remaining <= 130)
590 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
591 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
593 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
594 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
596 p->totalOffset += keyp->buf_size;
603 void key_input (BFiles bfs, int nkeys, int cache)
609 struct key_file **kf;
612 struct heap_info *hi;
613 struct progressInfo progressInfo;
621 getFnameTmp (fname, nkeys+1);
622 if (access (fname, R_OK) == -1)
629 dict = dict_open (bfs, FNAME_DICT, cache, 1);
632 logf (LOG_FATAL, "dict_open fail");
635 if (!res_get_match (common_resource, "isam", "i", NULL))
637 isamc = isc_open (bfs,
638 FNAME_ISAMC, 1, key_isamc_m (common_resource));
641 logf (LOG_FATAL, "isc_open fail");
647 isam = is_open (bfs, FNAME_ISAM, key_compare, 1,
648 sizeof(struct it_key), common_resource);
651 logf (LOG_FATAL, "is_open fail");
655 kf = xmalloc ((1+nkeys) * sizeof(*kf));
656 progressInfo.totalBytes = 0;
657 progressInfo.totalOffset = 0;
658 time (&progressInfo.startTime);
659 time (&progressInfo.lastTime);
660 for (i = 1; i<=nkeys; i++)
662 kf[i] = key_file_init (i, 32768);
663 kf[i]->readHandler = progressFunc;
664 kf[i]->readInfo = &progressInfo;
665 progressInfo.totalBytes += kf[i]->length;
666 progressInfo.totalOffset += kf[i]->buf_size;
668 hi = key_heap_init (nkeys, key_qsort_compare);
673 for (i = 1; i<=nkeys; i++)
674 if ((r = key_file_read (kf[i], rbuf)))
675 key_heap_insert (hi, rbuf, r, kf[i]);
686 for (i = 1; i<=nkeys; i++)
688 getFnameTmp (rbuf, i);
691 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
692 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
693 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
694 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
695 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);