2 * Copyright (C) 1994-1995, Index Data I/S
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.13 1996-02-05 12:30:00 adam
8 * Logging reduced a bit.
9 * The remaining running time is estimated during register merge.
11 * Revision 1.12 1995/12/06 17:49:19 adam
12 * Uses dict_delete now.
14 * Revision 1.11 1995/12/06 16:06:43 adam
15 * Better diagnostics. Work on 'real' dictionary deletion.
17 * Revision 1.10 1995/12/06 12:41:22 adam
18 * New command 'stat' for the index program.
19 * Filenames can be read from stdin by specifying '-'.
20 * Bug fix/enhancement of the transformation from terms to regular
21 * expressons in the search engine.
23 * Revision 1.9 1995/10/10 12:24:39 adam
24 * Temporary sort files are compressed.
26 * Revision 1.8 1995/10/04 16:57:19 adam
27 * Key input and merge sort in one pass.
29 * Revision 1.7 1995/10/02 15:18:52 adam
30 * New member in recRetrieveCtrl: diagnostic.
32 * Revision 1.6 1995/09/29 15:51:56 adam
33 * First work on multi-way read.
35 * Revision 1.5 1995/09/29 14:01:43 adam
38 * Revision 1.4 1995/09/28 14:22:57 adam
39 * Sort uses smaller temporary files.
41 * Revision 1.3 1995/09/06 16:11:17 adam
42 * Option: only one word key per file.
44 * Revision 1.2 1995/09/04 12:33:42 adam
45 * Various cleanup. YAZ util used instead.
47 * Revision 1.1 1995/09/04 09:10:37 adam
48 * More work on index add/del/update.
49 * Merge sort implemented.
50 * Initial work on z39 server.
65 #define KEY_SIZE (1+sizeof(struct it_key))
66 #define INP_NAME_MAX 8192
67 #define INP_BUF_START 60000
68 #define INP_BUF_ADD 400000
70 static int no_diffs = 0;
71 static int no_updates = 0;
72 static int no_deletions = 0;
73 static int no_insertions = 0;
74 static int no_iterations = 0;
78 off_t offset; /* file offset */
79 unsigned char *buf; /* buffer block */
80 size_t buf_size; /* number of read bytes in block */
81 size_t chunk; /* number of bytes allocated */
82 size_t buf_ptr; /* current position in buffer */
83 char *prev_name; /* last word read */
84 int sysno; /* last sysno */
85 int seqno; /* last seqno */
86 off_t length; /* length of file */
87 /* handler invoked in each read */
88 void (*readHandler)(struct key_file *keyp, void *rinfo);
92 void getFnameTmp (char *fname, int no)
94 sprintf (fname, TEMP_FNAME, no);
97 void key_file_chunk_read (struct key_file *f)
101 getFnameTmp (fname, f->no);
102 fd = open (fname, O_RDONLY);
105 logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname);
110 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
112 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
116 if (lseek (fd, f->offset, SEEK_SET) == -1)
118 logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname);
121 while (f->chunk - nr > 0)
123 r = read (fd, f->buf + nr, f->chunk - nr);
130 logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname);
136 (*f->readHandler)(f, f->readInfo);
140 struct key_file *key_file_init (int no, int chunk)
144 f = xmalloc (sizeof(*f));
151 f->readHandler = NULL;
152 f->buf = xmalloc (f->chunk);
153 f->prev_name = xmalloc (256);
154 *f->prev_name = '\0';
155 key_file_chunk_read (f);
159 int key_file_getc (struct key_file *f)
161 if (f->buf_ptr < f->buf_size)
162 return f->buf[(f->buf_ptr)++];
163 if (f->buf_size < f->chunk)
165 f->offset += f->buf_size;
166 key_file_chunk_read (f);
167 if (f->buf_ptr < f->buf_size)
168 return f->buf[(f->buf_ptr)++];
173 int key_file_decode (struct key_file *f)
177 c = key_file_getc (f);
184 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
187 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
188 d = (d << 8) + (key_file_getc (f) & 0xff);
191 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
192 d = (d << 8) + (key_file_getc (f) & 0xff);
193 d = (d << 8) + (key_file_getc (f) & 0xff);
199 int key_file_read (struct key_file *f, char *key)
204 c = key_file_getc (f);
207 strcpy (key, f->prev_name);
216 while ((key[i++] = key_file_getc (f)))
218 strcpy (f->prev_name, key);
221 d = key_file_decode (f);
224 itkey.sysno = d + f->sysno;
227 f->sysno = itkey.sysno;
230 d = key_file_decode (f);
231 itkey.seqno = d + f->seqno;
232 f->seqno = itkey.seqno;
233 memcpy (key + i, &itkey, sizeof(struct it_key));
234 return i + sizeof (struct it_key);
239 struct key_file **file;
244 int (*cmp)(const void *p1, const void *p2);
247 struct heap_info *key_heap_init (int nkeys,
248 int (*cmp)(const void *p1, const void *p2))
250 struct heap_info *hi;
253 hi = xmalloc (sizeof(*hi));
254 hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys));
255 hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
257 hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys));
259 for (i = 0; i<= nkeys; i++)
262 hi->info.buf[i] = xmalloc (512);
267 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
272 hi->ptr[i1] = hi->ptr[i2];
277 static void key_heap_delete (struct heap_info *hi)
279 int cur = 1, child = 2;
281 assert (hi->heapnum > 0);
283 key_heap_swap (hi, 1, hi->heapnum);
285 while (child <= hi->heapnum) {
286 if (child < hi->heapnum &&
287 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
288 &hi->info.buf[hi->ptr[child+1]]) > 0)
290 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
291 &hi->info.buf[hi->ptr[child]]) > 0)
293 key_heap_swap (hi, cur, child);
302 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
307 cur = ++(hi->heapnum);
308 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
309 hi->info.file[hi->ptr[cur]] = kf;
312 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
313 &hi->info.buf[hi->ptr[cur]]) > 0)
315 key_heap_swap (hi, cur, parent);
321 static int heap_read_one (struct heap_info *hi, char *name, char *key)
330 strcpy (name, hi->info.buf[n]);
331 kf = hi->info.file[n];
333 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
334 key_heap_delete (hi);
335 if ((r = key_file_read (kf, rbuf)))
336 key_heap_insert (hi, rbuf, r, kf);
341 int heap_inp (Dict dict, ISAM isam, struct heap_info *hi)
344 char next_name[INP_NAME_MAX+1];
345 char cur_name[INP_NAME_MAX+1];
346 int key_buf_size = INP_BUF_START;
352 next_key = xmalloc (KEY_SIZE);
353 key_buf = xmalloc (key_buf_size * (KEY_SIZE));
354 more = heap_read_one (hi, cur_name, key_buf);
355 while (more) /* EOF ? */
358 key_buf_ptr = KEY_SIZE;
361 if (!(more = heap_read_one (hi, next_name, next_key)))
363 if (*next_name && strcmp (next_name, cur_name))
365 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
366 key_buf_ptr += KEY_SIZE;
367 if (key_buf_ptr+KEY_SIZE >= key_buf_size)
370 new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD);
371 memcpy (new_key_buf, key_buf, key_buf_size);
372 key_buf_size += INP_BUF_ADD;
374 key_buf = new_key_buf;
378 nmemb = key_buf_ptr / KEY_SIZE;
379 assert (nmemb*KEY_SIZE == key_buf_ptr);
380 if ((info = dict_lookup (dict, cur_name)))
382 ISAM_P isam_p, isam_p2;
383 logf (LOG_DEBUG, "updating %s", cur_name);
384 memcpy (&isam_p, info+1, sizeof(ISAM_P));
385 isam_p2 = is_merge (isam, isam_p, nmemb, key_buf);
389 if (!dict_delete (dict, cur_name))
395 if (isam_p2 != isam_p)
396 dict_insert (dict, cur_name, sizeof(ISAM_P), &isam_p2);
402 logf (LOG_DEBUG, "inserting %s", cur_name);
404 isam_p = is_merge (isam, 0, nmemb, key_buf);
405 dict_insert (dict, cur_name, sizeof(ISAM_P), &isam_p);
407 memcpy (key_buf, next_key, KEY_SIZE);
408 strcpy (cur_name, next_name);
413 struct progressInfo {
420 void progressFunc (struct key_file *keyp, void *info)
422 struct progressInfo *p = info;
423 time_t now, remaining;
425 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
427 p->totalOffset += keyp->buf_size;
430 if (now < p->lastTime+10)
433 remaining = (now - p->startTime)*
434 ((double) p->totalBytes/p->totalOffset - 1.0);
435 if (remaining <= 130)
436 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
437 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
439 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
440 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
443 void key_input (const char *dict_fname, const char *isam_fname,
444 int nkeys, int cache)
449 struct key_file **kf;
452 struct heap_info *hi;
453 struct progressInfo progressInfo;
455 dict = dict_open (dict_fname, cache, 1);
458 logf (LOG_FATAL, "dict_open fail of `%s'", dict_fname);
461 isam = is_open (isam_fname, key_compare, 1, sizeof(struct it_key));
464 logf (LOG_FATAL, "is_open fail of `%s'", isam_fname);
468 kf = xmalloc ((1+nkeys) * sizeof(*kf));
469 progressInfo.totalBytes = 0;
470 progressInfo.totalOffset = 0;
471 time (&progressInfo.startTime);
472 time (&progressInfo.lastTime);
473 for (i = 1; i<=nkeys; i++)
475 kf[i] = key_file_init (i, 32768);
476 kf[i]->readHandler = progressFunc;
477 kf[i]->readInfo = &progressInfo;
478 progressInfo.totalBytes += kf[i]->length;
479 progressInfo.totalOffset += kf[i]->buf_size;
481 hi = key_heap_init (nkeys, key_qsort_compare);
482 for (i = 1; i<=nkeys; i++)
483 if ((r = key_file_read (kf[i], rbuf)))
484 key_heap_insert (hi, rbuf, r, kf[i]);
485 heap_inp (dict, isam, hi);
490 for (i = 1; i<=nkeys; i++)
492 getFnameTmp (rbuf, i);
495 logf (LOG_LOG, "Iterations . . .%7d", no_iterations);
496 logf (LOG_LOG, "Distinct words .%7d", no_diffs);
497 logf (LOG_LOG, "Updates. . . . .%7d", no_updates);
498 logf (LOG_LOG, "Deletions. . . .%7d", no_deletions);
499 logf (LOG_LOG, "Insertions . . .%7d", no_insertions);