1 /* $Id: isamb.c,v 1.47.2.6 2006-08-14 17:14:08 adam Exp $
2 Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002,2003,2004
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 this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
25 #include <yaz/xmalloc.h>
42 #define ISAMB_DATA_OFFSET 3
44 /* maximum size of encoded buffer */
45 #define DST_ITEM_MAX 256
47 #define ISAMB_MAX_LEVEL 10
48 /* approx 2*max page + max size of item */
49 #define DST_BUF_SIZE 16840
51 #define ISAMB_CACHE_ENTRY_SIZE 4096
53 /* CAT_MAX: _must_ be power of 2 */
55 #define CAT_MASK (CAT_MAX-1)
56 /* CAT_NO: <= CAT_MAX */
59 /* ISAMB_PTR_CODEC=1 var, =0 fixed */
60 #define ISAMB_PTR_CODEC 0
62 struct ISAMB_cache_entry {
67 struct ISAMB_cache_entry *next;
73 struct ISAMB_head head;
74 struct ISAMB_cache_entry *cache_entries;
81 struct ISAMB_file *file;
83 int cache; /* 0=no cache, 1=use cache, -1=dummy isam (for testing only) */
84 int log_io; /* log level for bf_read/bf_write calls */
85 int log_freelist; /* log level for freelist handling */
86 int skipped_numbers; /* on a leaf node */
88 int skipped_nodes[ISAMB_MAX_LEVEL]; /* [0]=skipped leaves, 1=higher etc */
89 int accessed_nodes[ISAMB_MAX_LEVEL]; /* nodes we did not skip */
103 void *decodeClientData;
111 int maxlevel; /* total depth */
114 int skipped_numbers; /* on a leaf node */
115 int returned_numbers;
116 int skipped_nodes[ISAMB_MAX_LEVEL]; /* [0]=skipped leaves, 1=higher etc */
117 int accessed_nodes[ISAMB_MAX_LEVEL]; /* nodes we did not skip */
118 struct ISAMB_block **block;
123 static void encode_ptr (char **dst, unsigned pos)
125 unsigned char *bp = (unsigned char*) *dst;
129 *bp++ = 128 | (pos & 127);
136 static void encode_ptr (char **dst, unsigned pos)
138 memcpy(*dst, &pos, sizeof(pos));
139 (*dst) += sizeof(pos);
144 static void decode_ptr (char **src1, int *pos)
146 unsigned char **src = (unsigned char **) src1;
151 while (((c = *(*src)++) & 128))
153 d += ((c & 127) << r);
160 static void decode_ptr (char **src, int *pos)
162 memcpy (pos, *src, sizeof(*pos));
163 (*src) += sizeof(*pos);
167 ISAMB isamb_open (BFiles bfs, const char *name, int writeflag, ISAMC_M *method,
170 ISAMB isamb = xmalloc (sizeof(*isamb));
174 isamb->method = (ISAMC_M *) xmalloc (sizeof(*method));
175 memcpy (isamb->method, method, sizeof(*method));
176 isamb->no_cat = CAT_NO;
178 isamb->log_freelist = 0;
179 isamb->cache = cache;
180 isamb->skipped_numbers=0;
181 isamb->returned_numbers=0;
182 for (i=0;i<ISAMB_MAX_LEVEL;i++)
183 isamb->skipped_nodes[i]= isamb->accessed_nodes[i]=0;
186 isamb->file = xmalloc (sizeof(*isamb->file) * isamb->no_cat);
187 for (i = 0; i<isamb->no_cat; i++)
189 char fname[DST_BUF_SIZE];
190 isamb->file[i].cache_entries = 0;
191 isamb->file[i].head_dirty = 0;
192 sprintf (fname, "%s%c", name, i+'A');
194 isamb->file[i].bf = bf_open (bfs, fname, ISAMB_CACHE_ENTRY_SIZE,
197 isamb->file[i].bf = bf_open (bfs, fname, b_size, writeflag);
200 if (!bf_read (isamb->file[i].bf, 0, 0, sizeof(struct ISAMB_head),
201 &isamb->file[i].head))
203 isamb->file[i].head.first_block = ISAMB_CACHE_ENTRY_SIZE/b_size+1;
204 isamb->file[i].head.last_block = isamb->file[i].head.first_block;
205 isamb->file[i].head.block_size = b_size;
206 isamb->file[i].head.block_max = b_size - ISAMB_DATA_OFFSET;
207 isamb->file[i].head.free_list = 0;
209 assert (isamb->file[i].head.block_size >= ISAMB_DATA_OFFSET);
210 isamb->file[i].head_dirty = 0;
211 assert(isamb->file[i].head.block_size == b_size);
215 logf(LOG_WARN, "isamb debug enabled. Things will be slower than usual");
220 static void flush_blocks (ISAMB b, int cat)
222 while (b->file[cat].cache_entries)
224 struct ISAMB_cache_entry *ce_this = b->file[cat].cache_entries;
225 b->file[cat].cache_entries = ce_this->next;
229 yaz_log (b->log_io, "bf_write: flush_blocks");
230 bf_write (b->file[cat].bf, ce_this->pos, 0, 0, ce_this->buf);
232 xfree (ce_this->buf);
237 static int get_block (ISAMB b, ISAMC_P pos, char *userbuf, int wr)
239 int cat = pos&CAT_MASK;
240 int off = ((pos/CAT_MAX) &
241 (ISAMB_CACHE_ENTRY_SIZE / b->file[cat].head.block_size - 1))
242 * b->file[cat].head.block_size;
243 int norm = pos / (CAT_MASK*ISAMB_CACHE_ENTRY_SIZE / b->file[cat].head.block_size);
245 struct ISAMB_cache_entry **ce, *ce_this = 0, **ce_last = 0;
250 assert (ISAMB_CACHE_ENTRY_SIZE >= b->file[cat].head.block_size);
251 for (ce = &b->file[cat].cache_entries; *ce; ce = &(*ce)->next, no++)
254 if ((*ce)->pos == norm)
257 *ce = (*ce)->next; /* remove from list */
259 ce_this->next = b->file[cat].cache_entries; /* move to front */
260 b->file[cat].cache_entries = ce_this;
264 memcpy (ce_this->buf + off, userbuf,
265 b->file[cat].head.block_size);
269 memcpy (userbuf, ce_this->buf + off,
270 b->file[cat].head.block_size);
277 assert (ce_last && *ce_last);
279 *ce_last = 0; /* remove the last entry from list */
282 yaz_log (b->log_io, "bf_write: get_block");
283 bf_write (b->file[cat].bf, ce_this->pos, 0, 0, ce_this->buf);
285 xfree (ce_this->buf);
288 ce_this = xmalloc (sizeof(*ce_this));
289 ce_this->next = b->file[cat].cache_entries;
290 b->file[cat].cache_entries = ce_this;
291 ce_this->buf = xmalloc (ISAMB_CACHE_ENTRY_SIZE);
293 yaz_log (b->log_io, "bf_read: get_block");
294 if (!bf_read (b->file[cat].bf, norm, 0, 0, ce_this->buf))
295 memset (ce_this->buf, 0, ISAMB_CACHE_ENTRY_SIZE);
298 memcpy (ce_this->buf + off, userbuf, b->file[cat].head.block_size);
304 memcpy (userbuf, ce_this->buf + off, b->file[cat].head.block_size);
310 void isamb_close (ISAMB isamb)
313 for (i=0;isamb->accessed_nodes[i];i++)
314 logf(LOG_DEBUG,"isamb_close level leaf-%d: %d read, %d skipped",
315 i, isamb->accessed_nodes[i], isamb->skipped_nodes[i]);
316 logf(LOG_DEBUG,"isamb_close returned %d values, skipped %d",
317 isamb->skipped_numbers, isamb->returned_numbers);
318 for (i = 0; i<isamb->no_cat; i++)
320 flush_blocks (isamb, i);
321 if (isamb->file[i].head_dirty)
322 bf_write (isamb->file[i].bf, 0, 0,
323 sizeof(struct ISAMB_head), &isamb->file[i].head);
325 bf_close (isamb->file[i].bf);
328 xfree (isamb->method);
332 static struct ISAMB_block *open_block (ISAMB b, ISAMC_P pos)
334 int cat = pos&CAT_MASK;
335 struct ISAMB_block *p;
338 p = xmalloc (sizeof(*p));
340 p->cat = pos & CAT_MASK;
341 p->buf = xmalloc (b->file[cat].head.block_size);
344 if (!get_block (b, pos, p->buf, 0))
346 yaz_log (b->log_io, "bf_read: open_block");
347 if (!bf_read (b->file[cat].bf, pos/CAT_MAX, 0, 0, p->buf))
349 yaz_log (LOG_FATAL, "isamb: read fail for pos=%ld block=%ld",
350 (long) pos, (long) pos/CAT_MAX);
354 p->bytes = p->buf + ISAMB_DATA_OFFSET;
356 p->size = (p->buf[1] + 256 * p->buf[2]) - ISAMB_DATA_OFFSET;
359 yaz_log (LOG_FATAL, "Bad block size %d in pos=%d\n", p->size, pos);
361 assert (p->size >= 0);
365 p->decodeClientData = (*b->method->code_start)(ISAMC_DECODE);
366 yaz_log (LOG_DEBUG, "isamb_open_block: Opened block %d ofs=%d",pos, p->offset);
370 struct ISAMB_block *new_block (ISAMB b, int leaf, int cat)
372 struct ISAMB_block *p;
374 p = xmalloc (sizeof(*p));
375 p->buf = xmalloc (b->file[cat].head.block_size);
377 if (!b->file[cat].head.free_list)
380 block_no = b->file[cat].head.last_block++;
381 p->pos = block_no * CAT_MAX + cat;
385 p->pos = b->file[cat].head.free_list;
386 assert((p->pos & CAT_MASK) == cat);
387 if (!get_block (b, p->pos, p->buf, 0))
389 yaz_log (b->log_io, "bf_read: new_block");
390 if (!bf_read (b->file[cat].bf, p->pos/CAT_MAX, 0, 0, p->buf))
392 yaz_log (LOG_FATAL, "isamb: read fail for pos=%ld block=%ld",
393 (long) p->pos/CAT_MAX, (long) p->pos/CAT_MAX);
397 yaz_log (b->log_freelist, "got block %d from freelist %d:%d", p->pos,
398 cat, p->pos/CAT_MAX);
399 memcpy (&b->file[cat].head.free_list, p->buf, sizeof(int));
402 b->file[cat].head_dirty = 1;
403 memset (p->buf, 0, b->file[cat].head.block_size);
404 p->bytes = p->buf + ISAMB_DATA_OFFSET;
410 p->decodeClientData = (*b->method->code_start)(ISAMC_DECODE);
414 struct ISAMB_block *new_leaf (ISAMB b, int cat)
416 return new_block (b, 1, cat);
420 struct ISAMB_block *new_int (ISAMB b, int cat)
422 return new_block (b, 0, cat);
425 static void check_block (ISAMB b, struct ISAMB_block *p)
434 char *startp = p->bytes;
436 char *endp = p->bytes + p->size;
439 decode_ptr (&src, &pos);
440 assert ((pos&CAT_MASK) == p->cat);
444 decode_ptr (&src, &item_len);
445 assert (item_len > 0 && item_len < 30);
447 decode_ptr (&src, &pos);
448 assert ((pos&CAT_MASK) == p->cat);
453 void close_block (ISAMB b, struct ISAMB_block *p)
459 yaz_log (b->log_freelist, "release block %d from freelist %d:%d",
460 p->pos, p->cat, p->pos/CAT_MAX);
461 memcpy (p->buf, &b->file[p->cat].head.free_list, sizeof(int));
462 b->file[p->cat].head.free_list = p->pos;
463 if (!get_block (b, p->pos, p->buf, 1))
465 yaz_log (b->log_io, "bf_write: close_block (deleted)");
466 bf_write (b->file[p->cat].bf, p->pos/CAT_MAX, 0, 0, p->buf);
471 int size = p->size + ISAMB_DATA_OFFSET;
472 assert (p->size >= 0);
474 p->buf[1] = size & 255;
475 p->buf[2] = size >> 8;
477 if (!get_block (b, p->pos, p->buf, 1))
479 yaz_log (b->log_io, "bf_write: close_block");
480 bf_write (b->file[p->cat].bf, p->pos/CAT_MAX, 0, 0, p->buf);
483 (*b->method->code_stop)(ISAMC_DECODE, p->decodeClientData);
488 int insert_sub (ISAMB b, struct ISAMB_block **p,
489 void *new_item, int *mode,
491 struct ISAMB_block **sp,
492 void *sub_item, int *sub_size,
495 int insert_int (ISAMB b, struct ISAMB_block *p, void *lookahead_item,
497 ISAMC_I *stream, struct ISAMB_block **sp,
498 void *split_item, int *split_size, void *last_max_item)
500 char *startp = p->bytes;
502 char *endp = p->bytes + p->size;
504 struct ISAMB_block *sub_p1 = 0, *sub_p2 = 0;
505 char sub_item[DST_ITEM_MAX];
511 assert(p->size >= 0);
512 decode_ptr (&src, &pos);
518 decode_ptr (&src, &item_len);
519 d = (*b->method->compare_item)(src, lookahead_item);
522 sub_p1 = open_block (b, pos);
524 more = insert_sub (b, &sub_p1, lookahead_item, mode,
526 sub_item, &sub_size, src);
531 decode_ptr (&src, &pos);
535 sub_p1 = open_block (b, pos);
537 more = insert_sub (b, &sub_p1, lookahead_item, mode, stream, &sub_p2,
538 sub_item, &sub_size, last_max_item);
542 /* there was a split - must insert pointer in this one */
543 char dst_buf[DST_BUF_SIZE];
546 assert (sub_size < 30 && sub_size > 1);
548 memcpy (dst, startp, src - startp);
552 encode_ptr (&dst, sub_size); /* sub length and item */
553 memcpy (dst, sub_item, sub_size);
556 encode_ptr (&dst, sub_p2->pos); /* pos */
558 if (endp - src) /* remaining data */
560 memcpy (dst, src, endp - src);
563 p->size = dst - dst_buf;
564 assert (p->size >= 0);
565 if (p->size <= b->file[p->cat].head.block_max)
567 memcpy (startp, dst_buf, dst - dst_buf);
569 close_block(b, sub_p2);
579 close_block(b, sub_p2);
581 half = src + b->file[p->cat].head.block_size/2;
582 decode_ptr (&src, &pos);
585 decode_ptr (&src, split_size);
587 decode_ptr (&src, &pos);
589 p_new_size = src - dst_buf;
590 memcpy (p->bytes, dst_buf, p_new_size);
592 decode_ptr (&src, split_size);
593 memcpy (split_item, src, *split_size);
596 *sp = new_int (b, p->cat);
597 (*sp)->size = endp - src;
598 memcpy ((*sp)->bytes, src, (*sp)->size);
600 p->size = p_new_size;
604 close_block (b, sub_p1);
609 int insert_leaf (ISAMB b, struct ISAMB_block **sp1, void *lookahead_item,
610 int *lookahead_mode, ISAMC_I *stream,
611 struct ISAMB_block **sp2,
612 void *sub_item, int *sub_size,
615 struct ISAMB_block *p = *sp1;
616 char *src = 0, *endp = 0;
617 char dst_buf[DST_BUF_SIZE], *dst = dst_buf;
619 void *c1 = (*b->method->code_start)(ISAMC_DECODE);
620 void *c2 = (*b->method->code_start)(ISAMC_ENCODE);
622 int quater = b->file[b->no_cat-1].head.block_max / CAT_MAX;
623 char *cut = dst_buf + quater * 2;
624 char *maxp = dst_buf + b->file[b->no_cat-1].head.block_max;
627 char cut_item_buf[DST_ITEM_MAX];
628 int cut_item_size = 0;
632 char file_item_buf[DST_ITEM_MAX];
633 char *file_item = file_item_buf;
636 endp = p->bytes + p->size;
637 (*b->method->code_item)(ISAMC_DECODE, c1, &file_item, &src);
641 char *lookahead_next;
645 d = (*b->method->compare_item)(file_item_buf, lookahead_item);
649 dst_item = lookahead_item;
650 if (!*lookahead_mode)
652 yaz_log (LOG_WARN, "isamb: Inconsistent register (1)");
653 assert (*lookahead_mode);
657 dst_item = file_item_buf;
658 if (!*lookahead_mode && d == 0)
662 else if (!half1 && dst > cut)
664 char *dst_item_0 = dst_item;
665 half1 = dst; /* candidate for splitting */
667 (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &dst_item);
669 cut_item_size = dst_item - dst_item_0;
670 memcpy (cut_item_buf, dst_item_0, cut_item_size);
675 (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &dst_item);
682 lookahead_next = lookahead_item;
683 if (!(*stream->read_item)(stream->clientData,
690 if (lookahead_item && max_item &&
691 (*b->method->compare_item)(max_item, lookahead_item) <= 0)
702 lookahead_next = lookahead_item;
703 if (!(*stream->read_item)(stream->clientData,
704 &lookahead_next, lookahead_mode))
711 file_item = file_item_buf;
712 (*b->method->code_item)(ISAMC_DECODE, c1, &file_item, &src);
718 file_item = file_item_buf;
719 (*b->method->code_item)(ISAMC_DECODE, c1, &file_item, &src);
723 maxp = dst_buf + b->file[b->no_cat-1].head.block_max + quater;
724 while (lookahead_item)
726 char *dst_item = lookahead_item;
730 (*b->method->compare_item)(max_item, lookahead_item) <= 0)
735 if (!*lookahead_mode)
737 yaz_log (LOG_WARN, "isamb: Inconsistent register (2)");
740 else if (!half1 && dst > cut)
742 char *dst_item_0 = dst_item;
743 half1 = dst; /* candidate for splitting */
745 (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &dst_item);
747 cut_item_size = dst_item - dst_item_0;
748 memcpy (cut_item_buf, dst_item_0, cut_item_size);
753 (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &dst_item);
762 dst_item = lookahead_item;
763 if (!(*stream->read_item)(stream->clientData, &dst_item,
770 new_size = dst - dst_buf;
771 if (p && p->cat != b->no_cat-1 &&
772 new_size > b->file[p->cat].head.block_max)
774 /* non-btree block will be removed */
777 /* delete it too!! */
778 p = 0; /* make a new one anyway */
781 { /* must create a new one */
783 for (i = 0; i < b->no_cat; i++)
784 if (new_size <= b->file[i].head.block_max)
790 if (new_size > b->file[p->cat].head.block_max)
793 char *cut_item = cut_item_buf;
799 p->size = half1 - dst_buf;
800 memcpy (p->bytes, dst_buf, half1 - dst_buf);
803 *sp2 = new_leaf (b, p->cat);
805 (*b->method->code_reset)(c2);
807 first_dst = (*sp2)->bytes;
809 (*b->method->code_item)(ISAMC_ENCODE, c2, &first_dst, &cut_item);
811 memcpy (first_dst, half2, dst - half2);
813 (*sp2)->size = (first_dst - (*sp2)->bytes) + (dst - half2);
816 memcpy (sub_item, cut_item_buf, cut_item_size);
817 *sub_size = cut_item_size;
821 memcpy (p->bytes, dst_buf, dst - dst_buf);
824 (*b->method->code_stop)(ISAMC_DECODE, c1);
825 (*b->method->code_stop)(ISAMC_ENCODE, c2);
830 int insert_sub (ISAMB b, struct ISAMB_block **p, void *new_item,
833 struct ISAMB_block **sp,
834 void *sub_item, int *sub_size,
837 if (!*p || (*p)->leaf)
838 return insert_leaf (b, p, new_item, mode, stream, sp, sub_item,
841 return insert_int (b, *p, new_item, mode, stream, sp, sub_item,
845 int isamb_unlink (ISAMB b, ISAMC_P pos)
847 struct ISAMB_block *p1;
851 p1 = open_block(b, pos);
857 char *src = p1->bytes + p1->offset;
859 decode_ptr(&src, &sub_p);
860 isamb_unlink(b, sub_p);
862 while (src != p1->bytes + p1->size)
864 decode_ptr(&src, &item_len);
866 decode_ptr(&src, &sub_p);
867 isamb_unlink(b, sub_p);
874 int isamb_merge (ISAMB b, ISAMC_P pos, ISAMC_I *stream)
876 char item_buf[DST_ITEM_MAX];
888 (*stream->read_item)(stream->clientData, &item_ptr, &i_mode);
893 more = (*stream->read_item)(stream->clientData, &item_ptr, &i_mode);
896 struct ISAMB_block *p = 0, *sp = 0;
897 char sub_item[DST_ITEM_MAX];
901 p = open_block (b, pos);
902 more = insert_sub (b, &p, item_buf, &i_mode, stream, &sp,
903 sub_item, &sub_size, 0);
905 { /* increase level of tree by one */
906 struct ISAMB_block *p2 = new_int (b, p->cat);
907 char *dst = p2->bytes + p2->size;
909 encode_ptr (&dst, p->pos);
910 assert (sub_size < 20);
911 encode_ptr (&dst, sub_size);
912 memcpy (dst, sub_item, sub_size);
914 encode_ptr (&dst, sp->pos);
916 p2->size = dst - p2->bytes;
917 pos = p2->pos; /* return new super page */
922 pos = p->pos; /* return current one (again) */
928 ISAMB_PP isamb_pp_open_x (ISAMB isamb, ISAMB_P pos, int *level)
930 ISAMB_PP pp = xmalloc (sizeof(*pp));
934 pp->block = xmalloc (ISAMB_MAX_LEVEL * sizeof(*pp->block));
941 pp->skipped_numbers=0;
942 pp->returned_numbers=0;
943 for (i = 0; i<ISAMB_MAX_LEVEL; i++)
944 pp->skipped_nodes[i] = pp->accessed_nodes[i] = 0;
950 struct ISAMB_block *p = open_block (isamb, pos);
951 char *src = p->bytes + p->offset;
952 pp->block[pp->level] = p;
954 pp->total_size += p->size;
959 decode_ptr (&src, &pos);
960 p->offset = src - p->bytes;
962 pp->accessed_nodes[pp->level]++;
965 pp->block[pp->level+1] = 0;
966 pp->maxlevel = pp->level;
972 ISAMB_PP isamb_pp_open (ISAMB isamb, ISAMB_P pos)
974 return isamb_pp_open_x (isamb, pos, 0);
977 void isamb_pp_close_x (ISAMB_PP pp, int *size, int *blocks)
982 logf(LOG_DEBUG,"isamb_pp_close lev=%d returned %d values, skipped %d",
983 pp->maxlevel, pp->skipped_numbers, pp->returned_numbers);
984 for (i=pp->maxlevel;i>=0;i--)
985 if ( pp->skipped_nodes[i] || pp->accessed_nodes[i])
986 logf(LOG_DEBUG,"isamb_pp_close level leaf-%d: %d read, %d skipped", i,
987 pp->accessed_nodes[i], pp->skipped_nodes[i]);
988 pp->isamb->skipped_numbers += pp->skipped_numbers;
989 pp->isamb->returned_numbers += pp->returned_numbers;
990 for (i=pp->maxlevel;i>=0;i--)
992 pp->isamb->accessed_nodes[i] += pp->accessed_nodes[i];
993 pp->isamb->skipped_nodes[i] += pp->skipped_nodes[i];
996 *size = pp->total_size;
998 *blocks = pp->no_blocks;
999 for (i = 0; i <= pp->level; i++)
1000 close_block (pp->isamb, pp->block[i]);
1005 int isamb_block_info (ISAMB isamb, int cat)
1007 if (cat >= 0 && cat < isamb->no_cat)
1008 return isamb->file[cat].head.block_size;
1012 void isamb_pp_close (ISAMB_PP pp)
1014 isamb_pp_close_x (pp, 0, 0);
1017 /* simple recursive dumper .. */
1018 static void isamb_dump_r (ISAMB b, ISAMB_P pos, void (*pr)(const char *str),
1022 char prefix_str[1024];
1025 struct ISAMB_block *p = open_block (b, pos);
1026 sprintf(prefix_str, "%*s %d cat=%d size=%d max=%d", level*2, "",
1027 pos, p->cat, p->size, b->file[p->cat].head.block_max);
1029 sprintf(prefix_str, "%*s %d", level*2, "", pos);
1032 while (p->offset < p->size)
1034 char *src = p->bytes + p->offset;
1036 (*b->method->code_item)(ISAMC_DECODE, p->decodeClientData,
1038 (*b->method->log_item)(LOG_DEBUG, buf, prefix_str);
1039 p->offset = src - (char*) p->bytes;
1041 assert(p->offset == p->size);
1045 char *src = p->bytes + p->offset;
1049 decode_ptr (&src, &sub);
1050 p->offset = src - (char*) p->bytes;
1052 isamb_dump_r(b, sub, pr, level+1);
1054 while (p->offset < p->size)
1056 decode_ptr (&src, &item_len);
1057 (*b->method->log_item)(LOG_DEBUG, src, prefix_str);
1059 decode_ptr (&src, &sub);
1061 p->offset = src - (char*) p->bytes;
1063 isamb_dump_r(b, sub, pr, level+1);
1070 void isamb_dump (ISAMB b, ISAMB_P pos, void (*pr)(const char *str))
1072 isamb_dump_r(b, pos, pr, 0);
1076 /* Old isamb_pp_read that Adam wrote, kept as a reference in case we need to
1077 debug the more complex pp_read that also forwards. May be deleted near end
1078 of 2004, if it has not shown to be useful */
1081 int isamb_pp_read (ISAMB_PP pp, void *buf)
1085 struct ISAMB_block *p = pp->block[pp->level];
1089 while (p->offset == p->size)
1092 while (p->offset == p->size)
1096 close_block (pp->isamb, pp->block[pp->level]);
1097 pp->block[pp->level] = 0;
1099 p = pp->block[pp->level];
1102 src = p->bytes + p->offset;
1104 decode_ptr (&src, &item_len);
1106 decode_ptr (&src, &pos);
1108 p->offset = src - (char*) p->bytes;
1114 pp->block[pp->level] = p = open_block (pp->isamb, pos);
1116 pp->total_size += p->size;
1123 src = p->bytes + p->offset;
1124 decode_ptr (&src, &pos);
1125 p->offset = src - (char*) p->bytes;
1129 assert (p->offset < p->size);
1131 src = p->bytes + p->offset;
1132 (*pp->isamb->method->code_item)(ISAMC_DECODE, p->decodeClientData,
1134 p->offset = src - (char*) p->bytes;
1135 /* key_logdump_txt(LOG_DEBUG,buf, "isamb_pp_read returning 1"); */
1140 int isamb_pp_read (ISAMB_PP pp, void *buf)
1142 return isamb_pp_forward(pp,buf,0);
1146 #define NEW_FORWARD 1
1148 #if NEW_FORWARD == 1
1152 #define ISAMB_DEBUG 1
1154 static int isamb_pp_on_right_node(ISAMB_PP pp, int level, const void *untilbuf)
1155 { /* looks one node higher to see if we should be on this node at all */
1156 /* useful in backing off quickly, and in avoiding tail descends */
1157 /* call with pp->level to begin with */
1158 struct ISAMB_block *p;
1165 logf(LOG_DEBUG,"isamb_pp_on_right returning true for root");
1167 return 1; /* we can never skip the root node */
1171 assert(p->offset <= p->size);
1172 if (p->offset < p->size )
1174 assert(p->offset>0);
1175 src=p->bytes + p->offset;
1176 decode_ptr(&src,&item_len);
1178 (*pp->isamb->method->log_item)(LOG_DEBUG,untilbuf,"on_leaf: until");
1179 (*pp->isamb->method->log_item)(LOG_DEBUG,src,"on_leaf: value");
1181 cmp=(*pp->isamb->method->compare_item)(untilbuf,src);
1184 logf(LOG_DEBUG,"isamb_pp_on_right returning true "
1185 "cmp=%d lev=%d ofs=%d",cmp,level,p->offset);
1191 logf(LOG_DEBUG,"isamb_pp_on_right returning false "
1192 "cmp=%d lev=%d ofs=%d",cmp,level,p->offset);
1199 logf(LOG_DEBUG,"isamb_pp_on_right at tail, looking higher "
1202 return isamb_pp_on_right_node(pp, level, untilbuf);
1204 } /* isamb_pp_on_right_node */
1206 static int isamb_pp_read_on_leaf(ISAMB_PP pp, void *buf)
1207 { /* reads the next item on the current leaf, returns 0 if end of leaf*/
1208 struct ISAMB_block *p = pp->block[pp->level];
1213 if (!p || p->offset == p->size) {
1215 logf(LOG_DEBUG,"isamb_pp_read_on_leaf returning 0 on node %d",p->pos);
1217 return 0; /* at end of leaf */
1219 src=p->bytes + p->offset;
1221 (*pp->isamb->method->code_item)
1222 (ISAMC_DECODE, p->decodeClientData,&dst, &src);
1223 p->offset = src - (char*) p->bytes;
1226 (*pp->isamb->method->log_item)(LOG_DEBUG, buf, "read_on_leaf returning 1");
1230 } /* read_on_leaf */
1232 static int isamb_pp_forward_on_leaf(ISAMB_PP pp, void *buf, const void *untilbuf)
1233 { /* forwards on the current leaf, returns 0 if not found */
1237 if (!isamb_pp_read_on_leaf(pp,buf))
1239 /* FIXME - this is an extra function call, inline the read? */
1240 cmp=(*pp->isamb->method->compare_item)(untilbuf,buf);
1241 if (cmp <2){ /* found a good one */
1244 logf(LOG_DEBUG, "isam_pp_fwd_on_leaf skipped %d items",skips);
1246 pp->returned_numbers++;
1250 if (!isamb_pp_on_right_node(pp, pp->level, untilbuf))
1251 return 0; /* never mind the rest of this leaf */
1252 pp->skipped_numbers++;
1255 } /* forward_on_leaf */
1257 static int isamb_pp_climb_level(ISAMB_PP pp, int *pos)
1258 { /* climbs higher in the tree, until finds a level with data left */
1259 /* returns the node to (consider to) descend to in *pos) */
1260 struct ISAMB_block *p = pp->block[pp->level];
1264 logf(LOG_DEBUG,"isamb_pp_climb_level starting "
1265 "at level %d node %d ofs=%d sz=%d",
1266 pp->level, p->pos, p->offset, p->size);
1268 assert(pp->level >= 0);
1271 assert(p->offset <= p->size);
1275 logf(LOG_DEBUG,"isamb_pp_climb_level returning 0 at root");
1279 assert(pp->level>0);
1280 close_block(pp->isamb, pp->block[pp->level]);
1281 pp->block[pp->level]=0;
1283 p=pp->block[pp->level];
1285 logf(LOG_DEBUG,"isamb_pp_climb_level climbed to level %d node %d ofs=%d",
1286 pp->level, p->pos, p->offset);
1289 assert(p->offset <= p->size);
1290 if (p->offset == p->size ) {
1291 /* we came from the last pointer, climb on */
1292 if (!isamb_pp_climb_level(pp,pos))
1294 p=pp->block[pp->level];
1298 /* skip the child we just came from */
1300 logf(LOG_DEBUG,"isam_pp_climb_level: skipping lev=%d ofs=%d sz=%d",
1301 pp->level, p->offset, p->size);
1303 assert (p->offset < p->size );
1304 src=p->bytes + p->offset;
1305 decode_ptr(&src, &item_len);
1307 decode_ptr(&src, pos);
1308 p->offset=src - (char *)p->bytes;
1315 static int isamb_pp_forward_unode(ISAMB_PP pp, int pos, const void *untilbuf)
1316 { /* scans a upper node until it finds a child <= untilbuf */
1317 /* pp points to the key value, as always. pos is the child read from */
1319 /* if all values are too small, returns the last child in the node */
1320 /* FIXME - this can be detected, and avoided by looking at the */
1321 /* parent node, but that gets messy. Presumably the cost is */
1322 /* pretty low anyway */
1323 struct ISAMB_block *p = pp->block[pp->level];
1324 char *src = p->bytes + p->offset;
1331 logf(LOG_DEBUG,"isamb_pp_forward_unode starting "
1332 "at level %d node %d ofs=%di sz=%d",
1333 pp->level, p->pos, p->offset, p->size);
1336 assert(p->offset <= p->size);
1337 if (p->offset == p->size) {
1339 logf(LOG_DEBUG,"isamb_pp_forward_unode returning at end "
1340 "at level %d node %d ofs=%di sz=%d",
1341 pp->level, p->pos, p->offset, p->size);
1343 return pos; /* already at the end of it */
1345 while(p->offset < p->size)
1347 decode_ptr(&src,&item_len);
1348 cmp=(*pp->isamb->method->compare_item)(untilbuf,src);
1350 decode_ptr(&src,&nxtpos);
1354 logf(LOG_DEBUG,"isamb_pp_forward_unode returning a hit "
1355 "at level %d node %d ofs=%d sz=%d",
1356 pp->level, p->pos, p->offset, p->size);
1361 p->offset=src-(char*)p->bytes;
1362 (pp->skipped_nodes[pp->maxlevel - pp->level -1])++;
1368 logf(LOG_DEBUG,"isamb_pp_forward_unode returning at tail "
1369 "at level %d node %d ofs=%d sz=%d skips=%d",
1370 pp->level, p->pos, p->offset, p->size, skips);
1372 return pos; /* that's the last one in the line */
1374 } /* forward_unode */
1376 static void isamb_pp_descend_to_leaf(ISAMB_PP pp, int pos, const void *untilbuf)
1377 { /* climbs down the tree, from pos, to the leftmost leaf */
1378 struct ISAMB_block *p = pp->block[pp->level];
1382 logf(LOG_DEBUG,"isamb_pp_descend_to_leaf "
1383 "starting at lev %d node %d ofs=%d lf=%d u=%p",
1384 pp->level, p->pos, p->offset, p->leaf, untilbuf);
1387 pos=isamb_pp_forward_unode(pp,pos,untilbuf);
1390 p = open_block(pp->isamb, pos);
1391 pp->block[pp->level] = p;
1392 ++(pp->accessed_nodes[pp->maxlevel-pp->level]);
1395 logf(LOG_DEBUG,"isamb_pp_descend_to_leaf "
1396 "got lev %d node %d lf=%d",
1397 pp->level, p->pos, p->leaf);
1401 assert (p->offset==0);
1402 src = p->bytes + p->offset;
1403 decode_ptr(&src, &pos);
1404 p->offset = src-(char*)p->bytes;
1405 isamb_pp_descend_to_leaf(pp,pos,untilbuf);
1407 logf(LOG_DEBUG,"isamb_pp_descend_to_leaf "
1408 "returning at lev %d node %d ofs=%d lf=%d",
1409 pp->level, p->pos, p->offset, p->leaf);
1411 } /* descend_to_leaf */
1413 static int isamb_pp_find_next_leaf(ISAMB_PP pp)
1414 { /* finds the next leaf by climbing up and down */
1416 if (!isamb_pp_climb_level(pp,&pos))
1418 isamb_pp_descend_to_leaf(pp, pos,0);
1422 static int isamb_pp_climb_desc(ISAMB_PP pp, void *buf, const void *untilbuf)
1423 { /* climbs up and descends to a leaf where values >= *untilbuf are found */
1426 struct ISAMB_block *p = pp->block[pp->level];
1427 logf(LOG_DEBUG,"isamb_pp_climb_desc starting "
1428 "at level %d node %d ofs=%d sz=%d",
1429 pp->level, p->pos, p->offset, p->size);
1431 if (!isamb_pp_climb_level(pp,&pos))
1433 /* see if it would pay to climb one higher */
1434 if (!isamb_pp_on_right_node(pp, pp->level, untilbuf))
1435 if (!isamb_pp_climb_level(pp,&pos))
1437 isamb_pp_descend_to_leaf(pp, pos,untilbuf);
1439 p = pp->block[pp->level];
1440 logf(LOG_DEBUG,"isamb_pp_climb_desc done "
1441 "at level %d node %d ofs=%d sz=%d",
1442 pp->level, p->pos, p->offset, p->size);
1447 int isamb_pp_forward (ISAMB_PP pp, void *buf, const void *untilbuf)
1450 struct ISAMB_block *p = pp->block[pp->level];
1454 logf(LOG_DEBUG,"isamb_pp_forward starting "
1455 "at level %d node %d ofs=%d sz=%d u=%p",
1456 pp->level, p->pos, p->offset, p->size,untilbuf);
1460 if (isamb_pp_forward_on_leaf( pp, buf, untilbuf)) {
1463 logf(LOG_DEBUG,"isamb_pp_forward (f) returning (A) "
1464 "at level %d node %d ofs=%d sz=%d",
1465 pp->level, p->pos, p->offset, p->size);
1469 if (! isamb_pp_climb_desc( pp, buf, untilbuf)) {
1472 logf(LOG_DEBUG,"isamb_pp_forward (f) returning notfound (B) "
1473 "at level %d node %d ofs=%d sz=%d",
1474 pp->level, p->pos, p->offset, p->size);
1476 return 0; /* could not find a leaf */
1479 if (isamb_pp_forward_on_leaf( pp, buf, untilbuf)) {
1482 logf(LOG_DEBUG,"isamb_pp_forward (f) returning (C) "
1483 "at level %d node %d ofs=%d sz=%d",
1484 pp->level, p->pos, p->offset, p->size);
1488 } while (isamb_pp_find_next_leaf(pp));
1489 return 0; /* could not find at all */
1491 else { /* no untilbuf, a straight read */
1492 /* FIXME - this should be moved
1493 * directly into the pp_read */
1494 /* keeping here now, to keep same
1495 * interface as the old fwd */
1496 if (isamb_pp_read_on_leaf( pp, buf)) {
1499 logf(LOG_DEBUG,"isamb_pp_forward (read) returning (D) "
1500 "at level %d node %d ofs=%d sz=%d",
1501 pp->level, p->pos, p->offset, p->size);
1505 if (isamb_pp_find_next_leaf(pp)) {
1508 logf(LOG_DEBUG,"isamb_pp_forward (read) returning (E) "
1509 "at level %d node %d ofs=%d sz=%d",
1510 pp->level, p->pos, p->offset, p->size);
1512 return isamb_pp_read_on_leaf(pp, buf);
1517 } /* isam_pp_forward (new version) */
1519 #elif NEW_FORWARD == 0
1521 int isamb_pp_forward (ISAMB_PP pp, void *buf, const void *untilbuf)
1525 * while at end of node
1526 * climb higher. If out, return 0
1527 * while not on a leaf (and not at its end)
1536 * The upper nodes consist of a sequence of nodenumbers and keys
1537 * When opening a block, the first node number is read in, and
1538 * offset points to the first key, which is the upper limit of keys
1539 * in the node just read.
1543 struct ISAMB_block *p = pp->block[pp->level];
1548 int descending=0; /* used to prevent a border condition error */
1552 logf(LOG_DEBUG,"isamb_pp_forward starting [%p] p=%d",pp,p->pos);
1554 (*pp->isamb->method->log_item)(LOG_DEBUG, untilbuf, "until");
1555 (*pp->isamb->method->log_item)(LOG_DEBUG, buf, "buf");
1560 while ( (p->offset == p->size) && !descending )
1561 { /* end of this block - climb higher */
1562 assert (p->offset <= p->size);
1564 logf(LOG_DEBUG,"isamb_pp_forward climbing from l=%d",
1570 logf(LOG_DEBUG,"isamb_pp_forward returning 0 at root");
1572 return 0; /* at end of the root, nothing left */
1574 close_block(pp->isamb, pp->block[pp->level]);
1575 pp->block[pp->level]=0;
1577 p=pp->block[pp->level];
1579 logf(LOG_DEBUG,"isamb_pp_forward climbed to node %d off=%d",
1583 assert(p->offset <= p->size);
1584 /* skip the child we have handled */
1585 if (p->offset != p->size)
1587 src = p->bytes + p->offset;
1588 decode_ptr(&src, &item_len);
1590 (*pp->isamb->method->log_item)(LOG_DEBUG, src,
1591 " isamb_pp_forward "
1592 "climb skipping old key");
1595 decode_ptr(&src,&pos);
1596 p->offset = src - (char*) p->bytes;
1597 break; /* even if this puts us at the end of the block, we
1598 need to descend to the last pos. UGLY coding,
1599 clean up some day */
1604 src = p->bytes + p->offset;
1605 if (p->offset == p->size)
1606 cmp=-2 ; /* descend to the last node, as we have
1610 decode_ptr(&src, &item_len);
1612 logf(LOG_DEBUG,"isamb_pp_forward (B) on a high node. "
1613 "ofs=%d sz=%d nxtpos=%d ",
1614 p->offset,p->size,pos);
1615 (*pp->isamb->method->log_item)(LOG_DEBUG, src, "");
1618 cmp=(*pp->isamb->method->compare_item)(untilbuf,src);
1622 decode_ptr(&src,&nxtpos);
1627 logf(LOG_DEBUG,"isambb_pp_forward descending l=%d p=%d ",
1630 descending=1; /* prevent climbing for a while */
1632 p = open_block(pp->isamb,pos);
1633 pp->block[pp->level] = p ;
1634 pp->total_size += p->size;
1635 (pp->accessed_nodes[pp->maxlevel - pp->level])++;
1638 { /* block starts with a pos */
1639 src = p->bytes + p->offset;
1640 decode_ptr(&src,&pos);
1641 p->offset=src-(char*) p->bytes;
1643 logf(LOG_DEBUG,"isamb_pp_forward: block %d starts with %d",
1647 } /* descend to the node */
1649 { /* skip the node */
1650 p->offset = src - (char*) p->bytes;
1652 (pp->skipped_nodes[pp->maxlevel - pp->level -1])++;
1655 "isamb_pp_forward: skipping block on level %d, noting "
1657 pp->level, pp->maxlevel - pp->level-1 ,
1658 pp->skipped_nodes[pp->maxlevel - pp->level-1 ]);
1660 /* 0 is always leafs, 1 is one level above leafs etc, no
1661 * matter how high tree */
1663 } /* not on a leaf */
1666 if (p->offset == p->size) {
1671 assert (p->offset < p->size);
1672 src = p->bytes + p->offset;
1674 (*pp->isamb->method->code_item)(ISAMC_DECODE, p->decodeClientData,
1676 p->offset = src - (char*) p->bytes;
1678 cmp=(*pp->isamb->method->compare_item)(untilbuf,buf);
1682 logf(LOG_DEBUG,"isamb_pp_forward on a leaf. cmp=%d",
1684 (*pp->isamb->method->log_item)(LOG_DEBUG, buf, "");
1691 (*pp->isamb->method->log_item)(
1692 LOG_DEBUG, buf, "isamb_pp_forward returning 1");
1696 (*pp->isamb->method->log_item)(
1697 LOG_DEBUG, buf, "isamb_pp_read returning 1 (fwd)");
1700 pp->returned_numbers++;
1704 pp->skipped_numbers++;
1710 #elif NEW_FORWARD == 2
1712 int isamb_pp_forward (ISAMB_PP pp, void *buf, const void *untilb)
1716 struct ISAMB_block *p = pp->block[pp->level];
1721 while (p->offset == p->size)
1724 while (p->offset == p->size)
1728 close_block (pp->isamb, pp->block[pp->level]);
1729 pp->block[pp->level] = 0;
1731 p = pp->block[pp->level];
1736 src = p->bytes + p->offset;
1738 decode_ptr (&src, &item_len);
1740 decode_ptr (&src, &pos);
1742 p->offset = src - (char*) p->bytes;
1744 src = p->bytes + p->offset;
1748 if (!untilb || p->offset == p->size)
1750 assert(p->offset < p->size);
1751 decode_ptr (&src, &item_len);
1752 if ((*pp->isamb->method->compare_item)(untilb, src) <= 1)
1755 decode_ptr (&src, &pos);
1756 p->offset = src - (char*) p->bytes;
1763 pp->block[pp->level] = p = open_block (pp->isamb, pos);
1765 pp->total_size += p->size;
1773 src = p->bytes + p->offset;
1776 decode_ptr (&src, &pos);
1777 p->offset = src - (char*) p->bytes;
1779 if (!untilb || p->offset == p->size)
1781 assert(p->offset < p->size);
1782 decode_ptr (&src, &item_len);
1783 if ((*pp->isamb->method->compare_item)(untilb, src) <= 1)
1790 assert (p->offset < p->size);
1795 src = p->bytes + p->offset;
1796 (*pp->isamb->method->code_item)(ISAMC_DECODE, p->decodeClientData,
1798 p->offset = src - (char*) p->bytes;
1799 if (!untilb || (*pp->isamb->method->compare_item)(untilb, dst0) <= 1)
1802 if (p->offset == p->size) goto again;
1804 /* key_logdump_txt(LOG_DEBUG,buf, "isamb_pp_read returning 1"); */
1810 int isamb_pp_num (ISAMB_PP pp)
1815 static void isamb_pp_leaf_pos( ISAMB_PP pp,
1816 int *current, int *total, void *dummybuf )
1818 struct ISAMB_block *p = pp->block[pp->level];
1820 char *end=p->bytes+p->size;
1821 char *cur=p->bytes+p->offset;
1823 assert(p->offset <= p->size);
1831 (*pp->isamb->method->code_item)
1832 (ISAMC_DECODE, p->decodeClientData,&dst, &src);
1833 assert(dst<(char*) dummybuf+100); /*FIXME */
1838 logf(LOG_DEBUG, "isamb_pp_leaf_pos: cur=%d tot=%d ofs=%d sz=%d lev=%d",
1839 *current, *total, p->offset, p->size, pp->level);
1843 static void isamb_pp_upper_pos( ISAMB_PP pp, int *current, int *total,
1844 int size, int level )
1845 { /* estimates total/current occurrences from here up, excl leaf */
1846 struct ISAMB_block *p = pp->block[level];
1848 char *end=p->bytes+p->size;
1849 char *cur=p->bytes+p->offset;
1854 logf(LOG_DEBUG,"isamb_pp_upper_pos at beginning l=%d "
1855 "cur=%d tot=%d ofs=%d sz=%d pos=%d",
1856 level, *current, *total, p->offset, p->size, p->pos);
1857 assert (p->offset <= p->size);
1858 decode_ptr (&src, &child ); /* first child */
1865 decode_ptr (&src, &item_size );
1866 assert(src+item_size<=end);
1868 decode_ptr (&src, &child );
1871 isamb_pp_upper_pos(pp, current, total, *total, level-1);
1874 void isamb_pp_pos( ISAMB_PP pp, int *current, int *total )
1875 { /* return an estimate of the current position and of the total number of */
1876 /* occureences in the isam tree, based on the current leaf */
1877 struct ISAMB_block *p = pp->block[pp->level];
1878 char dummy[100]; /* 100 bytes/entry must be enough */
1882 isamb_pp_leaf_pos(pp,current, total, dummy);
1884 isamb_pp_upper_pos(pp, current, total, *total, pp->level-1);
1886 logf(LOG_DEBUG,"isamb_pp_pos: C=%d T=%d =%6.2f%%",
1887 *current, *total, 100.0*(*current)/(*total));