1 /* $Id: rsmultiandor.c,v 1.10 2004-11-19 10:27:14 heikki Exp $
2 Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002
5 This file is part of the Zebra server.
7 Zebra is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 2, or (at your option) any later
12 Zebra is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 You should have received a copy of the GNU General Public License
18 along with Zebra; see the file LICENSE.zebra. If not, write to the
19 Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
25 * This module implements the rsmultior and rsmultiand result sets
27 * rsmultior is based on a heap, from which we find the next hit.
29 * rsmultiand is based on a simple array of rsets, and a linear
30 * search to find the record that exists in all of those rsets.
31 * To speed things up, the array is sorted so that the smallest
32 * rsets come first, they are most likely to have the hits furthest
33 * away, and thus forwarding to them makes the most sense.
46 static RSFD r_open_and (RSET ct, int flag);
47 static RSFD r_open_or (RSET ct, int flag);
48 static void r_close (RSFD rfd);
49 static void r_delete (RSET ct);
50 static int r_read_and (RSFD rfd, void *buf, TERMID *term);
51 static int r_read_or (RSFD rfd, void *buf, TERMID *term);
52 static int r_write (RSFD rfd, const void *buf);
53 static int r_forward_and(RSFD rfd, void *buf, TERMID *term,
54 const void *untilbuf);
55 static int r_forward_or(RSFD rfd, void *buf, TERMID *term,
56 const void *untilbuf);
57 static void r_pos (RSFD rfd, double *current, double *total);
58 static void r_get_terms(RSET ct, TERMID *terms, int maxterms, int *curterm);
60 static const struct rset_control control_or =
72 static const struct rset_control control_and =
85 const struct rset_control *rset_kind_multior = &control_or;
86 const struct rset_control *rset_kind_multiand = &control_and;
88 /* The heap structure:
89 * The rset contains a list or rsets we are ORing together
90 * The rfd contains a heap of heap-items, which contain
91 * a rfd opened to those rsets, and a buffer for one key.
92 * They also contain a ptr to the rset list in the rset
93 * itself, for practical reasons.
106 const struct key_control *kctrl;
107 struct heap_item **heap; /* ptrs to the rfd */
109 typedef struct heap *HEAP;
112 struct rset_multiandor_info {
118 struct rset_multiandor_rfd {
120 struct heap_item *items; /* we alloc and free them here */
121 HEAP h; /* and move around here */
122 zint hits; /* returned so far */
123 int eof; /* seen the end of it */
124 int tailcount; /* how many items are tailing */
128 static int log_level=0;
129 static int log_level_initialized=0;
132 /* Heap functions ***********************/
135 static void heap_dump_item( HEAP h, int i, int level) {
139 (void)rset_pos(h->heap[i]->rset,h->heap[i]->fd, &cur, &tot);
140 yaz_log(log_level," %d %*s i=%p buf=%p %0.1f/%0.1f",i, level, "",
141 &(h->heap[i]), h->heap[i]->buf, cur,tot );
142 heap_dump_item(h, 2*i, level+1);
143 heap_dump_item(h, 2*i+1, level+1);
145 static void heap_dump( HEAP h,char *msg) {
146 yaz_log(log_level, "heap dump: %s num=%d max=%d",msg, h->heapnum, h->heapmax);
147 heap_dump_item(h,1,1);
151 static void heap_swap (HEAP h, int x, int y)
153 struct heap_item *swap;
155 h->heap[x]=h->heap[y];
159 static int heap_cmp(HEAP h, int x, int y)
161 return (*h->kctrl->cmp)(h->heap[x]->buf,h->heap[y]->buf);
164 static int heap_empty(HEAP h)
166 return ( 0==h->heapnum );
169 static void heap_delete (HEAP h)
170 { /* deletes the first item in the heap, and balances the rest */
171 int cur = 1, child = 2;
172 h->heap[1]=0; /* been deleted */
173 heap_swap (h, 1, h->heapnum--);
174 while (child <= h->heapnum) {
175 if (child < h->heapnum && heap_cmp(h,child,1+child)>0 )
177 if (heap_cmp(h,cur,child) > 0)
179 heap_swap (h, cur, child);
188 static void heap_balance (HEAP h)
189 { /* The heap root element has changed value (to bigger) */
190 /* swap downwards until the heap is ordered again */
191 int cur = 1, child = 2;
192 while (child <= h->heapnum) {
193 if (child < h->heapnum && heap_cmp(h,child,1+child)>0 )
195 if (heap_cmp(h,cur,child) > 0)
197 heap_swap (h, cur, child);
207 static void heap_insert (HEAP h, struct heap_item *hi)
211 cur = ++(h->heapnum);
212 assert(cur <= h->heapmax);
215 while (parent && (heap_cmp(h,parent,cur) > 0))
218 heap_swap (h, cur, parent);
226 HEAP heap_create (NMEM nmem, int size, const struct key_control *kctrl)
228 HEAP h = (HEAP) nmem_malloc (nmem, sizeof(*h));
230 ++size; /* heap array starts at 1 */
234 h->heap = (struct heap_item**) nmem_malloc(nmem,size*sizeof(*h->heap));
235 h->heap[0]=0; /* not used */
239 static void heap_clear( HEAP h)
245 static void heap_destroy (HEAP h)
247 /* nothing to delete, all is nmem'd, and will go away in due time */
250 int compare_ands(const void *x, const void *y)
251 { /* used in qsort to get the multi-and args in optimal order */
252 /* that is, those with fewest occurrences first */
253 const struct heap_item *hx=x;
254 const struct heap_item *hy=y;
255 double cur, totx, toty;
256 rset_pos(hx->fd, &cur, &totx);
257 rset_pos(hy->fd, &cur, &toty);
258 if ( totx > toty +0.5 ) return 1;
259 if ( totx < toty -0.5 ) return -1;
260 return 0; /* return totx - toty, except for overflows and rounding */
263 /* Creating and deleting rsets ***********************/
265 static RSET rsmulti_andor_create( NMEM nmem, const struct key_control *kcontrol,
266 int scope, int no_rsets, RSET* rsets,
267 const struct rset_control *ctrl)
269 RSET rnew=rset_create_base(ctrl, nmem,kcontrol, scope,0);
270 struct rset_multiandor_info *info;
271 if (!log_level_initialized)
273 log_level=yaz_log_module_level("rsmultiandor");
274 log_level_initialized=1;
276 info = (struct rset_multiandor_info *) nmem_malloc(rnew->nmem,sizeof(*info));
277 info->no_rsets=no_rsets;
278 info->rsets=(RSET*)nmem_malloc(rnew->nmem, no_rsets*sizeof(*rsets));
279 memcpy(info->rsets,rsets,no_rsets*sizeof(*rsets));
284 RSET rsmultior_create( NMEM nmem, const struct key_control *kcontrol, int scope,
285 int no_rsets, RSET* rsets)
287 return rsmulti_andor_create(nmem, kcontrol, scope,
288 no_rsets, rsets, &control_or);
291 RSET rsmultiand_create( NMEM nmem, const struct key_control *kcontrol, int scope,
292 int no_rsets, RSET* rsets)
294 return rsmulti_andor_create(nmem, kcontrol, scope,
295 no_rsets, rsets, &control_and);
298 static void r_delete (RSET ct)
300 struct rset_multiandor_info *info = (struct rset_multiandor_info *) ct->priv;
302 for(i=0;i<info->no_rsets;i++)
303 rset_delete(info->rsets[i]);
307 /* Opening and closing fd's on them *********************/
309 static RSFD r_open_andor (RSET ct, int flag, int is_and)
312 struct rset_multiandor_rfd *p;
313 struct rset_multiandor_info *info = (struct rset_multiandor_info *) ct->priv;
314 const struct key_control *kctrl = ct->keycontrol;
317 if (flag & RSETF_WRITE)
319 yaz_log (YLOG_FATAL, "multiandor set type is read-only");
322 rfd=rfd_create_base(ct);
324 p=(struct rset_multiandor_rfd *)rfd->priv;
328 /* all other pointers shouls already be allocated, in right sizes! */
331 p = (struct rset_multiandor_rfd *) nmem_malloc (ct->nmem,sizeof(*p));
336 p->tailbits=nmem_malloc(ct->nmem, info->no_rsets*sizeof(char) );
338 p->h = heap_create( ct->nmem, info->no_rsets, kctrl);
339 p->items=(struct heap_item *) nmem_malloc(ct->nmem,
340 info->no_rsets*sizeof(*p->items));
341 for (i=0; i<info->no_rsets; i++){
342 p->items[i].rset=info->rsets[i];
343 p->items[i].buf=nmem_malloc(ct->nmem,kctrl->key_size);
351 { /* read the array and sort it */
352 for (i=0; i<info->no_rsets; i++){
353 p->items[i].fd=rset_open(info->rsets[i],RSETF_READ);
354 if (!rset_read(p->items[i].fd, p->items[i].buf, &p->items[i].term))
358 qsort(p->items, info->no_rsets, sizeof(p->items[0]), compare_ands);
360 { /* fill the heap for ORing */
361 for (i=0; i<info->no_rsets; i++){
362 p->items[i].fd=rset_open(info->rsets[i],RSETF_READ);
363 if ( rset_read(p->items[i].fd, p->items[i].buf, &p->items[i].term))
364 heap_insert(p->h, &(p->items[i]));
370 static RSFD r_open_or (RSET ct, int flag)
372 return r_open_andor(ct, flag, 0);
375 static RSFD r_open_and (RSET ct, int flag)
377 return r_open_andor(ct, flag, 1);
381 static void r_close (RSFD rfd)
383 struct rset_multiandor_info *info=
384 (struct rset_multiandor_info *)(rfd->rset->priv);
385 struct rset_multiandor_rfd *p=(struct rset_multiandor_rfd *)(rfd->priv);
390 for (i = 0; i<info->no_rsets; i++)
392 rset_close(p->items[i].fd);
393 rfd_delete_base(rfd);
398 static int r_forward_or(RSFD rfd, void *buf,
399 TERMID *term,const void *untilbuf)
400 { /* while heap head behind untilbuf, forward it and rebalance heap */
401 struct rset_multiandor_rfd *p=rfd->priv;
402 const struct key_control *kctrl=rfd->rset->keycontrol;
403 if (heap_empty(p->h))
405 while ( (*kctrl->cmp)(p->h->heap[1]->buf,untilbuf) < -rfd->rset->scope )
407 if (rset_forward(p->h->heap[1]->fd,p->h->heap[1]->buf,
408 &p->h->heap[1]->term, untilbuf))
413 if (heap_empty(p->h))
418 return r_read_or(rfd,buf,term);
422 static int r_read_or (RSFD rfd, void *buf, TERMID *term)
424 struct rset_multiandor_rfd *mrfd=rfd->priv;
425 const struct key_control *kctrl=rfd->rset->keycontrol;
426 struct heap_item *it;
428 if (heap_empty(mrfd->h))
430 it = mrfd->h->heap[1];
431 memcpy(buf,it->buf, kctrl->key_size);
435 rdres=rset_read(it->fd, it->buf, &it->term);
437 heap_balance(mrfd->h);
439 heap_delete(mrfd->h);
444 static int r_read_and (RSFD rfd, void *buf, TERMID *term)
445 { /* Has to return all hits where each item points to the */
446 /* same sysno (scope), in order. Keep an extra key (hitkey) */
447 /* as long as all records do not point to hitkey, forward */
448 /* them, and update hitkey to be the highest seen so far. */
449 /* (if any item eof's, mark eof, and return 0 thereafter) */
450 /* Once a hit has been found, scan all items for the smallest */
451 /* value. Mark all as being in the tail. Read next from that */
452 /* item, and if not in the same record, clear its tail bit */
453 struct rset_multiandor_rfd *p=rfd->priv;
454 const struct key_control *kctrl=rfd->rset->keycontrol;
455 struct rset_multiandor_info *info=rfd->rset->priv;
461 { /* we are tailing, find lowest tail and return it */
463 while ((mintail<info->no_rsets) && !p->tailbits[mintail])
464 mintail++; /* first tail */
465 for (i=mintail+1;i<info->no_rsets;i++)
469 cmp=(*kctrl->cmp)(p->items[i].buf,p->items[mintail].buf);
474 /* return the lowest tail */
475 memcpy(buf, p->items[mintail].buf, kctrl->key_size);
477 *term=p->items[mintail].term;
478 if (!rset_read(p->items[mintail].fd, p->items[mintail].buf,
479 &p->items[mintail].term))
481 p->eof=1; /* game over, once tails have been returned */
482 p->tailbits[mintail]=0;
487 cmp=(*kctrl->cmp)(p->items[mintail].buf,buf);
488 if (cmp >= rfd->rset->scope){
489 p->tailbits[mintail]=0;
494 /* not tailing, forward until all reocrds match, and set up */
495 /* as tails. the earlier 'if' will then return the hits */
497 return 0; /* nothing more to see */
498 i=1; /* assume items[0] is highest up */
499 while (i<info->no_rsets) {
500 cmp=(*kctrl->cmp)(p->items[0].buf,p->items[i].buf);
501 if (cmp<=-rfd->rset->scope) { /* [0] was behind, forward it */
502 if (!rset_forward(p->items[0].fd, p->items[0].buf,
503 &p->items[0].term, p->items[i].buf))
505 p->eof=1; /* game over */
508 i=0; /* start frowarding from scratch */
509 } else if (cmp>=rfd->rset->scope)
510 { /* [0] was ahead, forward i */
511 if (!rset_forward(p->items[i].fd, p->items[i].buf,
512 &p->items[i].term, p->items[0].buf))
514 p->eof=1; /* game over */
520 /* if we get this far, all rsets are now within +- scope of [0] */
521 /* ergo, we have a hit. Mark them all as tailing, and let the */
522 /* upper 'if' return the hits in right order */
523 for (i=0; i<info->no_rsets;i++)
525 p->tailcount=info->no_rsets;
530 static int r_forward_and(RSFD rfd, void *buf, TERMID *term,
531 const void *untilbuf)
533 struct rset_multiandor_rfd *p=rfd->priv;
534 const struct key_control *kctrl=rfd->rset->keycontrol;
535 struct rset_multiandor_info *info=rfd->rset->priv;
540 for (i=0; i<info->no_rsets;i++)
542 cmp=(*kctrl->cmp)(p->items[i].buf,untilbuf);
543 if ( cmp <= -rfd->rset->scope )
545 killtail=1; /* we are moving to a different hit */
546 if (!rset_forward(p->items[i].fd, p->items[i].buf,
547 &p->items[i].term, untilbuf))
549 p->eof=1; /* game over */
557 for (i=0; i<info->no_rsets;i++)
561 return r_read_and(rfd,buf,term);
564 static void r_pos (RSFD rfd, double *current, double *total)
566 struct rset_multiandor_info *info=
567 (struct rset_multiandor_info *)(rfd->rset->priv);
568 struct rset_multiandor_rfd *mrfd=(struct rset_multiandor_rfd *)(rfd->priv);
570 double scur=0.0, stot=0.0;
572 for (i=0; i<info->no_rsets; i++){
573 rset_pos(mrfd->items[i].fd, &cur, &tot);
574 yaz_log(log_level, "r_pos: %d %0.1f %0.1f", i, cur,tot);
578 if (stot <1.0) { /* nothing there */
581 yaz_log(log_level, "r_pos: NULL %0.1f %0.1f", *current, *total);
585 *total=*current*stot/scur;
586 yaz_log(log_level, "r_pos: = %0.1f %0.1f", *current, *total);
591 static int r_write (RSFD rfd, const void *buf)
593 yaz_log (YLOG_FATAL, "multior set type is read-only");
597 static void r_get_terms(RSET ct, TERMID *terms, int maxterms, int *curterm)
598 /* Special case: Some multi-ors have all terms pointing to the same */
599 /* term. We do not want to duplicate those. Other multiors (and ands) */
600 /* have different terms under them. Those we want. */
602 struct rset_multiandor_info *info =
603 (struct rset_multiandor_info *) ct->priv;
604 int firstterm= *curterm;
606 for (i=0;i<info->no_rsets;i++)
608 rset_getterms(info->rsets[i], terms, maxterms, curterm);
609 if ( ( (*curterm) > firstterm+1 ) &&
610 ( (*curterm) <= maxterms ) &&
611 ( terms[(*curterm)-1] == terms[firstterm] ) )
612 (*curterm)--; /* forget the term, seen that before */