2 * Copyright (C) 1994-1996, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.19 1996-10-29 14:09:46 adam
8 * Use of cisam system - enabled if setting isamc is 1.
10 * Revision 1.18 1996/06/04 10:18:59 adam
11 * Minor changes - removed include of ctype.h.
13 * Revision 1.17 1996/05/14 15:47:07 adam
14 * Cleanup of various buffer size entities.
16 * Revision 1.16 1996/04/09 10:05:20 adam
17 * Bug fix: prev_name buffer possibly too small; allocated in key_file_init.
19 * Revision 1.15 1996/03/21 14:50:09 adam
20 * File update uses modify-time instead of change-time.
22 * Revision 1.14 1996/02/07 14:06:37 adam
23 * Better progress report during register merge.
24 * New command: clean - removes temporary shadow files.
26 * Revision 1.13 1996/02/05 12:30:00 adam
27 * Logging reduced a bit.
28 * The remaining running time is estimated during register merge.
30 * Revision 1.12 1995/12/06 17:49:19 adam
31 * Uses dict_delete now.
33 * Revision 1.11 1995/12/06 16:06:43 adam
34 * Better diagnostics. Work on 'real' dictionary deletion.
36 * Revision 1.10 1995/12/06 12:41:22 adam
37 * New command 'stat' for the index program.
38 * Filenames can be read from stdin by specifying '-'.
39 * Bug fix/enhancement of the transformation from terms to regular
40 * expressons in the search engine.
42 * Revision 1.9 1995/10/10 12:24:39 adam
43 * Temporary sort files are compressed.
45 * Revision 1.8 1995/10/04 16:57:19 adam
46 * Key input and merge sort in one pass.
48 * Revision 1.7 1995/10/02 15:18:52 adam
49 * New member in recRetrieveCtrl: diagnostic.
51 * Revision 1.6 1995/09/29 15:51:56 adam
52 * First work on multi-way read.
54 * Revision 1.5 1995/09/29 14:01:43 adam
57 * Revision 1.4 1995/09/28 14:22:57 adam
58 * Sort uses smaller temporary files.
60 * Revision 1.3 1995/09/06 16:11:17 adam
61 * Option: only one word key per file.
63 * Revision 1.2 1995/09/04 12:33:42 adam
64 * Various cleanup. YAZ util used instead.
66 * Revision 1.1 1995/09/04 09:10:37 adam
67 * More work on index add/del/update.
68 * Merge sort implemented.
69 * Initial work on z39 server.
82 #define KEY_SIZE (1+sizeof(struct it_key))
83 #define INP_NAME_MAX 768
84 #define INP_BUF_START 60000
85 #define INP_BUF_ADD 400000
87 static int no_diffs = 0;
88 static int no_updates = 0;
89 static int no_deletions = 0;
90 static int no_insertions = 0;
91 static int no_iterations = 0;
95 off_t offset; /* file offset */
96 unsigned char *buf; /* buffer block */
97 size_t buf_size; /* number of read bytes in block */
98 size_t chunk; /* number of bytes allocated */
99 size_t buf_ptr; /* current position in buffer */
100 char *prev_name; /* last word read */
101 int sysno; /* last sysno */
102 int seqno; /* last seqno */
103 off_t length; /* length of file */
104 /* handler invoked in each read */
105 void (*readHandler)(struct key_file *keyp, void *rinfo);
109 void getFnameTmp (char *fname, int no)
113 pre = res_get_def (common_resource, "keyTmpDir", ".");
114 sprintf (fname, "%s/key%d.tmp", pre, no);
117 void key_file_chunk_read (struct key_file *f)
121 getFnameTmp (fname, f->no);
122 fd = open (fname, O_RDONLY);
125 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
130 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
132 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
136 if (lseek (fd, f->offset, SEEK_SET) == -1)
138 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
141 while (f->chunk - nr > 0)
143 r = read (fd, f->buf + nr, f->chunk - nr);
150 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
156 (*f->readHandler)(f, f->readInfo);
160 struct key_file *key_file_init (int no, int chunk)
164 f = xmalloc (sizeof(*f));
171 f->readHandler = NULL;
172 f->buf = xmalloc (f->chunk);
173 f->prev_name = xmalloc (INP_NAME_MAX);
174 *f->prev_name = '\0';
175 key_file_chunk_read (f);
179 int key_file_getc (struct key_file *f)
181 if (f->buf_ptr < f->buf_size)
182 return f->buf[(f->buf_ptr)++];
183 if (f->buf_size < f->chunk)
185 f->offset += f->buf_size;
186 key_file_chunk_read (f);
187 if (f->buf_ptr < f->buf_size)
188 return f->buf[(f->buf_ptr)++];
193 int key_file_decode (struct key_file *f)
197 c = key_file_getc (f);
204 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
207 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
208 d = (d << 8) + (key_file_getc (f) & 0xff);
211 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
212 d = (d << 8) + (key_file_getc (f) & 0xff);
213 d = (d << 8) + (key_file_getc (f) & 0xff);
219 int key_file_read (struct key_file *f, char *key)
224 c = key_file_getc (f);
227 strcpy (key, f->prev_name);
236 while ((key[i++] = key_file_getc (f)))
238 strcpy (f->prev_name, key);
241 d = key_file_decode (f);
244 itkey.sysno = d + f->sysno;
247 f->sysno = itkey.sysno;
250 d = key_file_decode (f);
251 itkey.seqno = d + f->seqno;
252 f->seqno = itkey.seqno;
253 memcpy (key + i, &itkey, sizeof(struct it_key));
254 return i + sizeof (struct it_key);
259 struct key_file **file;
264 int (*cmp)(const void *p1, const void *p2);
270 struct heap_info *key_heap_init (int nkeys,
271 int (*cmp)(const void *p1, const void *p2))
273 struct heap_info *hi;
276 hi = xmalloc (sizeof(*hi));
277 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
278 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
280 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
282 for (i = 0; i<= nkeys; i++)
285 hi->info.buf[i] = xmalloc (INP_NAME_MAX);
290 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
295 hi->ptr[i1] = hi->ptr[i2];
300 static void key_heap_delete (struct heap_info *hi)
302 int cur = 1, child = 2;
304 assert (hi->heapnum > 0);
306 key_heap_swap (hi, 1, hi->heapnum);
308 while (child <= hi->heapnum) {
309 if (child < hi->heapnum &&
310 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
311 &hi->info.buf[hi->ptr[child+1]]) > 0)
313 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
314 &hi->info.buf[hi->ptr[child]]) > 0)
316 key_heap_swap (hi, cur, child);
325 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
330 cur = ++(hi->heapnum);
331 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
332 hi->info.file[hi->ptr[cur]] = kf;
335 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
336 &hi->info.buf[hi->ptr[cur]]) > 0)
338 key_heap_swap (hi, cur, parent);
344 static int heap_read_one (struct heap_info *hi, char *name, char *key)
347 char rbuf[INP_NAME_MAX];
353 strcpy (name, hi->info.buf[n]);
354 kf = hi->info.file[n];
356 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
357 key_heap_delete (hi);
358 if ((r = key_file_read (kf, rbuf)))
359 key_heap_insert (hi, rbuf, r, kf);
364 struct heap_cread_info {
365 char prev_name[INP_NAME_MAX];
366 char cur_name[INP_NAME_MAX];
368 struct heap_info *hi;
373 int heap_cread_item (void *vp, char **dst, int *insertMode)
375 struct heap_cread_info *p = vp;
376 struct heap_info *hi = p->hi;
380 *insertMode = p->key[0];
381 memcpy (*dst, p->key+1, sizeof(struct it_key));
382 (*dst) += sizeof(struct it_key);
386 strcpy (p->prev_name, p->cur_name);
387 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
389 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
394 *insertMode = p->key[0];
395 memcpy (*dst, p->key+1, sizeof(struct it_key));
396 (*dst) += sizeof(struct it_key);
400 int heap_inpc (struct heap_info *hi)
402 struct heap_cread_info hci;
404 ISAMC_I isamc_i = xmalloc (sizeof(*isamc_i));
407 hci.key = xmalloc (KEY_SIZE);
410 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
412 isamc_i->clientData = &hci;
413 isamc_i->read_item = heap_cread_item;
417 logf (LOG_DEBUG, "inserting %s", 1+hci.cur_name);
418 if ((dict_info = dict_lookup (hi->dict, hci.cur_name)))
420 logf (LOG_FATAL, "Cannot merge really yet! %s", hci.cur_name);
426 strcpy (this_name, hci.cur_name);
427 isamc_p = isc_merge (hi->isamc, 0, isamc_i);
429 dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p);
436 int heap_inp (struct heap_info *hi)
439 char next_name[INP_NAME_MAX];
440 char cur_name[INP_NAME_MAX];
441 int key_buf_size = INP_BUF_START;
447 next_key = xmalloc (KEY_SIZE);
448 key_buf = xmalloc (key_buf_size);
449 more = heap_read_one (hi, cur_name, key_buf);
450 while (more) /* EOF ? */
453 key_buf_ptr = KEY_SIZE;
456 if (!(more = heap_read_one (hi, next_name, next_key)))
458 if (*next_name && strcmp (next_name, cur_name))
460 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
461 key_buf_ptr += KEY_SIZE;
462 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
465 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
466 memcpy (new_key_buf, key_buf, key_buf_size);
467 key_buf_size += INP_BUF_ADD;
469 key_buf = new_key_buf;
473 nmemb = key_buf_ptr / KEY_SIZE;
474 assert (nmemb*KEY_SIZE == key_buf_ptr);
475 if ((info = dict_lookup (hi->dict, cur_name)))
477 ISAM_P isam_p, isam_p2;
478 logf (LOG_DEBUG, "updating %s", 1+cur_name);
479 memcpy (&isam_p, info+1, sizeof(ISAM_P));
480 isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf);
484 if (!dict_delete (hi->dict, cur_name))
490 if (isam_p2 != isam_p)
491 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2);
497 logf (LOG_DEBUG, "inserting %s", 1+cur_name);
499 isam_p = is_merge (hi->isam, 0, nmemb, key_buf);
500 dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p);
502 memcpy (key_buf, next_key, KEY_SIZE);
503 strcpy (cur_name, next_name);
508 struct progressInfo {
515 void progressFunc (struct key_file *keyp, void *info)
517 struct progressInfo *p = info;
518 time_t now, remaining;
520 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
524 if (now >= p->lastTime+10)
527 remaining = (now - p->startTime)*
528 ((double) p->totalBytes/p->totalOffset - 1.0);
529 if (remaining <= 130)
530 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
531 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
533 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
534 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
536 p->totalOffset += keyp->buf_size;
539 void key_input (int nkeys, int cache)
545 struct key_file **kf;
548 struct heap_info *hi;
549 struct progressInfo progressInfo;
557 getFnameTmp (fname, nkeys+1);
558 if (access (fname, R_OK) == -1)
565 dict = dict_open (FNAME_DICT, cache, 1);
568 logf (LOG_FATAL, "dict_open fail");
571 if (res_get_match (common_resource, "isam", "c", NULL))
573 isamc = isc_open (FNAME_ISAMC, 1, key_isamc_m ());
576 logf (LOG_FATAL, "isc_open fail");
582 isam = is_open (FNAME_ISAM, key_compare, 1, sizeof(struct it_key));
585 logf (LOG_FATAL, "is_open fail");
589 kf = xmalloc ((1+nkeys) * sizeof(*kf));
590 progressInfo.totalBytes = 0;
591 progressInfo.totalOffset = 0;
592 time (&progressInfo.startTime);
593 time (&progressInfo.lastTime);
594 for (i = 1; i<=nkeys; i++)
596 kf[i] = key_file_init (i, 32768);
597 kf[i]->readHandler = progressFunc;
598 kf[i]->readInfo = &progressInfo;
599 progressInfo.totalBytes += kf[i]->length;
600 progressInfo.totalOffset += kf[i]->buf_size;
602 hi = key_heap_init (nkeys, key_qsort_compare);
607 for (i = 1; i<=nkeys; i++)
608 if ((r = key_file_read (kf[i], rbuf)))
609 key_heap_insert (hi, rbuf, r, kf[i]);
620 for (i = 1; i<=nkeys; i++)
622 getFnameTmp (rbuf, i);
625 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
626 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
627 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
628 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
629 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);