1 /* $Id: rsmultiandor.c,v 1.8 2004-10-26 15:32:11 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 /* Heap functions ***********************/
131 static void heap_dump_item( HEAP h, int i, int level) {
135 (void)rset_pos(h->heap[i]->rset,h->heap[i]->fd, &cur, &tot);
136 logf(LOG_LOG," %d %*s i=%p buf=%p %0.1f/%0.1f",i, level, "",
137 &(h->heap[i]), h->heap[i]->buf, cur,tot );
138 heap_dump_item(h, 2*i, level+1);
139 heap_dump_item(h, 2*i+1, level+1);
141 static void heap_dump( HEAP h,char *msg) {
142 logf(LOG_LOG, "heap dump: %s num=%d max=%d",msg, h->heapnum, h->heapmax);
143 heap_dump_item(h,1,1);
147 static void heap_swap (HEAP h, int x, int y)
149 struct heap_item *swap;
151 h->heap[x]=h->heap[y];
155 static int heap_cmp(HEAP h, int x, int y)
157 return (*h->kctrl->cmp)(h->heap[x]->buf,h->heap[y]->buf);
160 static int heap_empty(HEAP h)
162 return ( 0==h->heapnum );
165 static void heap_delete (HEAP h)
166 { /* deletes the first item in the heap, and balances the rest */
167 int cur = 1, child = 2;
168 h->heap[1]=0; /* been deleted */
169 heap_swap (h, 1, h->heapnum--);
170 while (child <= h->heapnum) {
171 if (child < h->heapnum && heap_cmp(h,child,1+child)>0 )
173 if (heap_cmp(h,cur,child) > 0)
175 heap_swap (h, cur, child);
184 static void heap_balance (HEAP h)
185 { /* The heap root element has changed value (to bigger) */
186 /* swap downwards until the heap is ordered again */
187 int cur = 1, child = 2;
188 while (child <= h->heapnum) {
189 if (child < h->heapnum && heap_cmp(h,child,1+child)>0 )
191 if (heap_cmp(h,cur,child) > 0)
193 heap_swap (h, cur, child);
203 static void heap_insert (HEAP h, struct heap_item *hi)
207 cur = ++(h->heapnum);
208 assert(cur <= h->heapmax);
211 while (parent && (heap_cmp(h,parent,cur) > 0))
214 heap_swap (h, cur, parent);
222 HEAP heap_create (NMEM nmem, int size, const struct key_control *kctrl)
224 HEAP h = (HEAP) nmem_malloc (nmem, sizeof(*h));
226 ++size; /* heap array starts at 1 */
230 h->heap = (struct heap_item**) nmem_malloc(nmem,size*sizeof(*h->heap));
231 h->heap[0]=0; /* not used */
235 static void heap_clear( HEAP h)
241 static void heap_destroy (HEAP h)
243 /* nothing to delete, all is nmem'd, and will go away in due time */
246 int compare_ands(const void *x, const void *y)
247 { /* used in qsort to get the multi-and args in optimal order */
248 /* that is, those with fewest occurrences first */
249 const struct heap_item *hx=x;
250 const struct heap_item *hy=y;
251 double cur, totx, toty;
252 rset_pos(hx->fd, &cur, &totx);
253 rset_pos(hy->fd, &cur, &toty);
254 if ( totx > toty +0.5 ) return 1;
255 if ( totx < toty -0.5 ) return -1;
256 return 0; /* return totx - toty, except for overflows and rounding */
259 /* Creating and deleting rsets ***********************/
261 static RSET rsmulti_andor_create( NMEM nmem, const struct key_control *kcontrol,
262 int scope, int no_rsets, RSET* rsets,
263 const struct rset_control *ctrl)
265 RSET rnew=rset_create_base(ctrl, nmem,kcontrol, scope,0);
266 struct rset_multiandor_info *info;
267 info = (struct rset_multiandor_info *) nmem_malloc(rnew->nmem,sizeof(*info));
268 info->no_rsets=no_rsets;
269 info->rsets=(RSET*)nmem_malloc(rnew->nmem, no_rsets*sizeof(*rsets));
270 memcpy(info->rsets,rsets,no_rsets*sizeof(*rsets));
275 RSET rsmultior_create( NMEM nmem, const struct key_control *kcontrol, int scope,
276 int no_rsets, RSET* rsets)
278 return rsmulti_andor_create(nmem, kcontrol, scope,
279 no_rsets, rsets, &control_or);
282 RSET rsmultiand_create( NMEM nmem, const struct key_control *kcontrol, int scope,
283 int no_rsets, RSET* rsets)
285 return rsmulti_andor_create(nmem, kcontrol, scope,
286 no_rsets, rsets, &control_and);
289 static void r_delete (RSET ct)
291 struct rset_multiandor_info *info = (struct rset_multiandor_info *) ct->priv;
293 for(i=0;i<info->no_rsets;i++)
294 rset_delete(info->rsets[i]);
298 /* Opening and closing fd's on them *********************/
300 static RSFD r_open_andor (RSET ct, int flag, int is_and)
303 struct rset_multiandor_rfd *p;
304 struct rset_multiandor_info *info = (struct rset_multiandor_info *) ct->priv;
305 const struct key_control *kctrl = ct->keycontrol;
308 if (flag & RSETF_WRITE)
310 logf (LOG_FATAL, "multiandor set type is read-only");
313 rfd=rfd_create_base(ct);
315 p=(struct rset_multiandor_rfd *)rfd->priv;
319 /* all other pointers shouls already be allocated, in right sizes! */
322 p = (struct rset_multiandor_rfd *) nmem_malloc (ct->nmem,sizeof(*p));
327 p->tailbits=nmem_malloc(ct->nmem, info->no_rsets*sizeof(char) );
329 p->h = heap_create( ct->nmem, info->no_rsets, kctrl);
330 p->items=(struct heap_item *) nmem_malloc(ct->nmem,
331 info->no_rsets*sizeof(*p->items));
332 for (i=0; i<info->no_rsets; i++){
333 p->items[i].rset=info->rsets[i];
334 p->items[i].buf=nmem_malloc(ct->nmem,kctrl->key_size);
342 { /* read the array and sort it */
343 for (i=0; i<info->no_rsets; i++){
344 p->items[i].fd=rset_open(info->rsets[i],RSETF_READ);
345 if (!rset_read(p->items[i].fd, p->items[i].buf, &p->items[i].term))
349 qsort(p->items, info->no_rsets, sizeof(p->items[0]), compare_ands);
351 { /* fill the heap for ORing */
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))
355 heap_insert(p->h, &(p->items[i]));
361 static RSFD r_open_or (RSET ct, int flag)
363 return r_open_andor(ct, flag, 0);
366 static RSFD r_open_and (RSET ct, int flag)
368 return r_open_andor(ct, flag, 1);
372 static void r_close (RSFD rfd)
374 struct rset_multiandor_info *info=
375 (struct rset_multiandor_info *)(rfd->rset->priv);
376 struct rset_multiandor_rfd *p=(struct rset_multiandor_rfd *)(rfd->priv);
381 for (i = 0; i<info->no_rsets; i++)
383 rset_close(p->items[i].fd);
384 rfd_delete_base(rfd);
389 static int r_forward_or(RSFD rfd, void *buf,
390 TERMID *term,const void *untilbuf)
391 { /* while heap head behind untilbuf, forward it and rebalance heap */
392 struct rset_multiandor_rfd *p=rfd->priv;
393 const struct key_control *kctrl=rfd->rset->keycontrol;
394 if (heap_empty(p->h))
396 while ( (*kctrl->cmp)(p->h->heap[1]->buf,untilbuf) < -rfd->rset->scope )
398 if (rset_forward(p->h->heap[1]->fd,p->h->heap[1]->buf,
399 &p->h->heap[1]->term, untilbuf))
404 if (heap_empty(p->h))
409 return r_read_or(rfd,buf,term);
413 static int r_read_or (RSFD rfd, void *buf, TERMID *term)
415 struct rset_multiandor_rfd *mrfd=rfd->priv;
416 const struct key_control *kctrl=rfd->rset->keycontrol;
417 struct heap_item *it;
419 if (heap_empty(mrfd->h))
421 it = mrfd->h->heap[1];
422 memcpy(buf,it->buf, kctrl->key_size);
426 rdres=rset_read(it->fd, it->buf, &it->term);
428 heap_balance(mrfd->h);
430 heap_delete(mrfd->h);
435 static int r_read_and (RSFD rfd, void *buf, TERMID *term)
436 { /* Has to return all hits where each item points to the */
437 /* same sysno (scope), in order. Keep an extra key (hitkey) */
438 /* as long as all records do not point to hitkey, forward */
439 /* them, and update hitkey to be the highest seen so far. */
440 /* (if any item eof's, mark eof, and return 0 thereafter) */
441 /* Once a hit has been found, scan all items for the smallest */
442 /* value. Mark all as being in the tail. Read next from that */
443 /* item, and if not in the same record, clear its tail bit */
444 struct rset_multiandor_rfd *p=rfd->priv;
445 const struct key_control *kctrl=rfd->rset->keycontrol;
446 struct rset_multiandor_info *info=rfd->rset->priv;
452 { /* we are tailing, find lowest tail and return it */
454 while ((mintail<info->no_rsets) && !p->tailbits[mintail])
455 mintail++; /* first tail */
456 for (i=mintail+1;i<info->no_rsets;i++)
460 cmp=(*kctrl->cmp)(p->items[i].buf,p->items[mintail].buf);
465 /* return the lowest tail */
466 memcpy(buf, p->items[mintail].buf, kctrl->key_size);
468 *term=p->items[mintail].term;
469 if (!rset_read(p->items[mintail].fd, p->items[mintail].buf,
470 &p->items[mintail].term))
472 p->eof=1; /* game over, once tails have been returned */
473 p->tailbits[mintail]=0;
478 cmp=(*kctrl->cmp)(p->items[mintail].buf,buf);
479 if (cmp >= rfd->rset->scope){
480 p->tailbits[mintail]=0;
485 /* not tailing, forward until all reocrds match, and set up */
486 /* as tails. the earlier 'if' will then return the hits */
488 return 0; /* nothing more to see */
489 i=1; /* assume items[0] is highest up */
490 while (i<info->no_rsets) {
491 cmp=(*kctrl->cmp)(p->items[0].buf,p->items[i].buf);
492 if (cmp<=-rfd->rset->scope) { /* [0] was behind, forward it */
493 if (!rset_forward(p->items[0].fd, p->items[0].buf,
494 &p->items[0].term, p->items[i].buf))
496 p->eof=1; /* game over */
499 i=0; /* start frowarding from scratch */
500 } else if (cmp>=rfd->rset->scope)
501 { /* [0] was ahead, forward i */
502 if (!rset_forward(p->items[i].fd, p->items[i].buf,
503 &p->items[i].term, p->items[0].buf))
505 p->eof=1; /* game over */
511 /* if we get this far, all rsets are now within +- scope of [0] */
512 /* ergo, we have a hit. Mark them all as tailing, and let the */
513 /* upper 'if' return the hits in right order */
514 for (i=0; i<info->no_rsets;i++)
516 p->tailcount=info->no_rsets;
521 static int r_forward_and(RSFD rfd, void *buf, TERMID *term,
522 const void *untilbuf)
524 struct rset_multiandor_rfd *p=rfd->priv;
525 const struct key_control *kctrl=rfd->rset->keycontrol;
526 struct rset_multiandor_info *info=rfd->rset->priv;
531 for (i=0; i<info->no_rsets;i++)
533 cmp=(*kctrl->cmp)(p->items[i].buf,untilbuf);
534 if ( cmp <= -rfd->rset->scope )
536 killtail=1; /* we are moving to a different hit */
537 if (!rset_forward(p->items[i].fd, p->items[i].buf,
538 &p->items[i].term, untilbuf))
540 p->eof=1; /* game over */
548 for (i=0; i<info->no_rsets;i++)
552 return r_read_and(rfd,buf,term);
555 static void r_pos (RSFD rfd, double *current, double *total)
557 struct rset_multiandor_info *info=
558 (struct rset_multiandor_info *)(rfd->rset->priv);
559 struct rset_multiandor_rfd *mrfd=(struct rset_multiandor_rfd *)(rfd->priv);
561 double scur=0.0, stot=0.0;
563 for (i=0; i<info->no_rsets; i++){
564 rset_pos(mrfd->items[i].fd, &cur, &tot);
565 /*logf(LOG_LOG, "r_pos: %d %0.1f %0.1f", i, cur,tot); */
569 if (stot <1.0) { /* nothing there */
572 /* logf(LOG_LOG, "r_pos: NULL %0.1f %0.1f", *current, *total);*/
576 *total=*current*stot/scur;
577 /*logf(LOG_LOG, "r_pos: = %0.1f %0.1f", *current, *total);*/
582 static int r_write (RSFD rfd, const void *buf)
584 logf (LOG_FATAL, "multior set type is read-only");
588 static void r_get_terms(RSET ct, TERMID *terms, int maxterms, int *curterm)
589 /* Special case: Some multi-ors have all terms pointing to the same */
590 /* term. We do not want to duplicate those. Other multiors (and ands) */
591 /* have different terms under them. Those we want. */
593 struct rset_multiandor_info *info =
594 (struct rset_multiandor_info *) ct->priv;
595 int firstterm= *curterm;
597 for (i=0;i<info->no_rsets;i++)
599 rset_getterms(info->rsets[i], terms, maxterms, curterm);
600 if ( ( (*curterm) > firstterm+1 ) &&
601 ( (*curterm) <= maxterms ) &&
602 ( terms[(*curterm)-1] == terms[firstterm] ) )
603 (*curterm)--; /* forget the term, seen that before */