+/* $Id: isamb.c,v 1.61 2004-12-08 14:02:36 adam Exp $
+ Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002,2003,2004
+ Index Data Aps
+This file is part of the Zebra server.
+
+Zebra is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2, or (at your option) any later
+version.
+
+Zebra is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with Zebra; see the file LICENSE.zebra. If not, write to the
+Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
+02111-1307, USA.
+*/
+
+#include <string.h>
#include <yaz/xmalloc.h>
-#include <yaz/log.h>
-#include <isamb.h>
+#include <yaz/ylog.h>
+#include <idzebra/isamb.h>
#include <assert.h>
+#ifndef ISAMB_DEBUG
+#define ISAMB_DEBUG 0
+#endif
+
+#define ISAMB_MAJOR_VERSION 2
+#define ISAMB_MINOR_VERSION 0
+
struct ISAMB_head {
- int first_block;
- int last_block;
+ zint first_block;
+ zint last_block;
+ zint free_list;
+ zint no_items;
int block_size;
+ int block_max;
+ int block_offset;
};
-#define ISAMB_DATA_OFFSET 3
+/* maximum size of encoded buffer */
+#define DST_ITEM_MAX 256
-struct ISAMB_s {
- BFiles bfs;
+#define ISAMB_MAX_LEVEL 10
+/* approx 2*max page + max size of item */
+#define DST_BUF_SIZE 16840
+
+#define ISAMB_CACHE_ENTRY_SIZE 4096
+
+/* CAT_MAX: _must_ be power of 2 */
+#define CAT_MAX 4
+#define CAT_MASK (CAT_MAX-1)
+/* CAT_NO: <= CAT_MAX */
+#define CAT_NO 4
+
+/* ISAMB_PTR_CODEC=1 var, =0 fixed */
+#define ISAMB_PTR_CODEC 1
+
+struct ISAMB_cache_entry {
+ ISAMB_P pos;
+ unsigned char *buf;
+ int dirty;
+ int hits;
+ struct ISAMB_cache_entry *next;
+};
+
+struct ISAMB_file {
BFile bf;
- ISAMC_M method;
int head_dirty;
-
struct ISAMB_head head;
+ struct ISAMB_cache_entry *cache_entries;
+};
+
+struct ISAMB_s {
+ BFiles bfs;
+ ISAMC_M *method;
+
+ struct ISAMB_file *file;
+ int no_cat;
+ int cache; /* 0=no cache, 1=use cache, -1=dummy isam (for testing only) */
+ int log_io; /* log level for bf_read/bf_write calls */
+ int log_freelist; /* log level for freelist handling */
+ zint skipped_numbers; /* on a leaf node */
+ zint returned_numbers;
+ zint skipped_nodes[ISAMB_MAX_LEVEL]; /* [0]=skipped leaves, 1=higher etc */
+ zint accessed_nodes[ISAMB_MAX_LEVEL]; /* nodes we did not skip */
};
struct ISAMB_block {
- int pos;
+ ISAMB_P pos;
+ int cat;
int size;
int leaf;
int dirty;
+ int deleted;
int offset;
- unsigned char *bytes;
+ zint no_items; /* number of nodes in this + children */
+ char *bytes;
+ char *cbuf;
+ unsigned char *buf;
void *decodeClientData;
+ int log_rw;
};
struct ISAMB_PP_s {
ISAMB isamb;
+ ISAMB_P pos;
int level;
+ int maxlevel; /* total depth */
+ zint total_size;
+ zint no_blocks;
+ zint skipped_numbers; /* on a leaf node */
+ zint returned_numbers;
+ zint skipped_nodes[ISAMB_MAX_LEVEL]; /* [0]=skipped leaves, 1=higher etc */
+ zint accessed_nodes[ISAMB_MAX_LEVEL]; /* nodes we did not skip */
struct ISAMB_block **block;
+ int scope; /* on what level we forward */
};
-void encode_ptr (char **dst, int pos)
+
+#if ISAMB_PTR_CODEC
+static void encode_ptr (char **dst, zint pos)
+{
+ unsigned char *bp = (unsigned char*) *dst;
+
+ while (pos > 127)
+ {
+ *bp++ = 128 | (pos & 127);
+ pos = pos >> 7;
+ }
+ *bp++ = pos;
+ *dst = (char *) bp;
+}
+#else
+static void encode_ptr (char **dst, zint pos)
{
- memcpy (*dst, &pos, sizeof(pos));
+ memcpy(*dst, &pos, sizeof(pos));
(*dst) += sizeof(pos);
}
+#endif
-void decode_ptr (char **src, int *pos)
+#if ISAMB_PTR_CODEC
+static void decode_ptr (const char **src1, zint *pos)
+{
+ const unsigned char **src = (const unsigned char **) src1;
+ zint d = 0;
+ unsigned char c;
+ unsigned r = 0;
+
+ while (((c = *(*src)++) & 128))
+ {
+ d += ((zint) (c & 127) << r);
+ r += 7;
+ }
+ d += ((zint) c << r);
+ *pos = d;
+}
+#else
+static void decode_ptr (const char **src, zint *pos)
{
- memcpy (pos, *src, sizeof(*pos));
- (*src) += sizeof(*pos);
+ memcpy (pos, *src, sizeof(*pos));
+ (*src) += sizeof(*pos);
}
+#endif
-ISAMB isamb_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method)
+ISAMB isamb_open (BFiles bfs, const char *name, int writeflag, ISAMC_M *method,
+ int cache)
{
ISAMB isamb = xmalloc (sizeof(*isamb));
+ int i, b_size = 32;
isamb->bfs = bfs;
- isamb->method = (ISAMC_M) xmalloc (sizeof(*method));
+ isamb->method = (ISAMC_M *) xmalloc (sizeof(*method));
memcpy (isamb->method, method, sizeof(*method));
+ isamb->no_cat = CAT_NO;
+ isamb->log_io = 0;
+ isamb->log_freelist = 0;
+ isamb->cache = cache;
+ isamb->skipped_numbers=0;
+ isamb->returned_numbers=0;
+ for (i=0;i<ISAMB_MAX_LEVEL;i++)
+ isamb->skipped_nodes[i]= isamb->accessed_nodes[i]=0;
- isamb->head.first_block = 1;
- isamb->head.last_block = 1;
- isamb->head.block_size = 1024;
- isamb->head_dirty = 0;
+ assert (cache == 0);
+ isamb->file = xmalloc (sizeof(*isamb->file) * isamb->no_cat);
+ for (i = 0; i < isamb->no_cat; i++)
+ {
+ char fname[DST_BUF_SIZE];
+ char hbuf[DST_BUF_SIZE];
+ isamb->file[i].cache_entries = 0;
+ isamb->file[i].head_dirty = 0;
+ sprintf (fname, "%s%c", name, i+'A');
+ if (cache)
+ isamb->file[i].bf = bf_open (bfs, fname, ISAMB_CACHE_ENTRY_SIZE,
+ writeflag);
+ else
+ isamb->file[i].bf = bf_open (bfs, fname, b_size, writeflag);
- isamb->bf = bf_open (bfs, name, isamb->head.block_size, writeflag);
-
- bf_read (isamb->bf, 0, 0, sizeof(struct ISAMB_head),
- &isamb->head);
+ /* fill-in default values (for empty isamb) */
+ isamb->file[i].head.first_block = ISAMB_CACHE_ENTRY_SIZE/b_size+1;
+ isamb->file[i].head.last_block = isamb->file[i].head.first_block;
+ isamb->file[i].head.block_size = b_size;
+ if (i == isamb->no_cat-1 || b_size > 128)
+ isamb->file[i].head.block_offset = 8;
+ else
+ isamb->file[i].head.block_offset = 4;
+ isamb->file[i].head.block_max =
+ b_size - isamb->file[i].head.block_offset;
+ isamb->file[i].head.free_list = 0;
+ if (bf_read (isamb->file[i].bf, 0, 0, 0, hbuf))
+ {
+ /* got header assume "isamb"major minor len can fit in 16 bytes */
+ zint zint_tmp;
+ int major, minor, len, pos = 0;
+ int left;
+ const char *src = 0;
+ if (memcmp(hbuf, "isamb", 5))
+ {
+ yaz_log(YLOG_WARN, "bad isamb header for file %s", fname);
+ return 0;
+ }
+ if (sscanf(hbuf+5, "%d %d %d", &major, &minor, &len) != 3)
+ {
+ yaz_log(YLOG_WARN, "bad isamb header for file %s", fname);
+ return 0;
+ }
+ if (major != ISAMB_MAJOR_VERSION)
+ {
+ yaz_log(YLOG_WARN, "bad major version for file %s %d, must be %d",
+ fname, major, ISAMB_MAJOR_VERSION);
+ return 0;
+ }
+ for (left = len - b_size; left > 0; left = left - b_size)
+ {
+ pos++;
+ if (!bf_read (isamb->file[i].bf, pos, 0, 0, hbuf + pos*b_size))
+ {
+ yaz_log(YLOG_WARN, "truncated isamb header for "
+ "file=%s len=%d pos=%d",
+ fname, len, pos);
+ return 0;
+ }
+ }
+ src = hbuf + 16;
+ decode_ptr(&src, &isamb->file[i].head.first_block);
+ decode_ptr(&src, &isamb->file[i].head.last_block);
+ decode_ptr(&src, &zint_tmp);
+ isamb->file[i].head.block_size = zint_tmp;
+ decode_ptr(&src, &zint_tmp);
+ isamb->file[i].head.block_max = zint_tmp;
+ decode_ptr(&src, &isamb->file[i].head.free_list);
+ }
+ assert (isamb->file[i].head.block_size >= isamb->file[i].head.block_offset);
+ isamb->file[i].head_dirty = 0;
+ assert(isamb->file[i].head.block_size == b_size);
+ b_size = b_size * 4;
+ }
+#if ISAMB_DEBUG
+ yaz_log(YLOG_WARN, "isamb debug enabled. Things will be slower than usual");
+#endif
return isamb;
}
+static void flush_blocks (ISAMB b, int cat)
+{
+ while (b->file[cat].cache_entries)
+ {
+ struct ISAMB_cache_entry *ce_this = b->file[cat].cache_entries;
+ b->file[cat].cache_entries = ce_this->next;
+
+ if (ce_this->dirty)
+ {
+ yaz_log (b->log_io, "bf_write: flush_blocks");
+ bf_write (b->file[cat].bf, ce_this->pos, 0, 0, ce_this->buf);
+ }
+ xfree (ce_this->buf);
+ xfree (ce_this);
+ }
+}
+
+static int get_block (ISAMB b, ISAMC_P pos, char *userbuf, int wr)
+{
+ int cat = (int) (pos&CAT_MASK);
+ int off = (int) (((pos/CAT_MAX) &
+ (ISAMB_CACHE_ENTRY_SIZE / b->file[cat].head.block_size - 1))
+ * b->file[cat].head.block_size);
+ zint norm = pos / (CAT_MASK*ISAMB_CACHE_ENTRY_SIZE / b->file[cat].head.block_size);
+ int no = 0;
+ struct ISAMB_cache_entry **ce, *ce_this = 0, **ce_last = 0;
+
+ if (!b->cache)
+ return 0;
+
+ assert (ISAMB_CACHE_ENTRY_SIZE >= b->file[cat].head.block_size);
+ for (ce = &b->file[cat].cache_entries; *ce; ce = &(*ce)->next, no++)
+ {
+ ce_last = ce;
+ if ((*ce)->pos == norm)
+ {
+ ce_this = *ce;
+ *ce = (*ce)->next; /* remove from list */
+
+ ce_this->next = b->file[cat].cache_entries; /* move to front */
+ b->file[cat].cache_entries = ce_this;
+
+ if (wr)
+ {
+ memcpy (ce_this->buf + off, userbuf,
+ b->file[cat].head.block_size);
+ ce_this->dirty = 1;
+ }
+ else
+ memcpy (userbuf, ce_this->buf + off,
+ b->file[cat].head.block_size);
+ return 1;
+ }
+ }
+ if (no >= 40)
+ {
+ assert (no == 40);
+ assert (ce_last && *ce_last);
+ ce_this = *ce_last;
+ *ce_last = 0; /* remove the last entry from list */
+ if (ce_this->dirty)
+ {
+ yaz_log (b->log_io, "bf_write: get_block");
+ bf_write (b->file[cat].bf, ce_this->pos, 0, 0, ce_this->buf);
+ }
+ xfree (ce_this->buf);
+ xfree (ce_this);
+ }
+ ce_this = xmalloc (sizeof(*ce_this));
+ ce_this->next = b->file[cat].cache_entries;
+ b->file[cat].cache_entries = ce_this;
+ ce_this->buf = xmalloc (ISAMB_CACHE_ENTRY_SIZE);
+ ce_this->pos = norm;
+ yaz_log (b->log_io, "bf_read: get_block");
+ if (!bf_read (b->file[cat].bf, norm, 0, 0, ce_this->buf))
+ memset (ce_this->buf, 0, ISAMB_CACHE_ENTRY_SIZE);
+ if (wr)
+ {
+ memcpy (ce_this->buf + off, userbuf, b->file[cat].head.block_size);
+ ce_this->dirty = 1;
+ }
+ else
+ {
+ ce_this->dirty = 0;
+ memcpy (userbuf, ce_this->buf + off, b->file[cat].head.block_size);
+ }
+ return 1;
+}
+
+
void isamb_close (ISAMB isamb)
{
- if (isamb->head_dirty)
- bf_write (isamb->bf, 0, 0, sizeof(struct ISAMB_head), &isamb->head);
+ int i;
+ for (i=0;isamb->accessed_nodes[i];i++)
+ yaz_log(YLOG_DEBUG,"isamb_close level leaf-%d: "ZINT_FORMAT" read, "
+ ZINT_FORMAT" skipped",
+ i, isamb->accessed_nodes[i], isamb->skipped_nodes[i]);
+ yaz_log(YLOG_DEBUG,"isamb_close returned "ZINT_FORMAT" values, "
+ "skipped "ZINT_FORMAT,
+ isamb->skipped_numbers, isamb->returned_numbers);
+ for (i = 0; i<isamb->no_cat; i++)
+ {
+ flush_blocks (isamb, i);
+ if (isamb->file[i].head_dirty)
+ {
+ char hbuf[DST_BUF_SIZE];
+ int major = ISAMB_MAJOR_VERSION;
+ int minor = ISAMB_MINOR_VERSION;
+ int len = 16;
+ char *dst = hbuf + 16;
+ int pos = 0, left;
+ int b_size = isamb->file[i].head.block_size;
+
+ encode_ptr(&dst, isamb->file[i].head.first_block);
+ encode_ptr(&dst, isamb->file[i].head.last_block);
+ encode_ptr(&dst, isamb->file[i].head.block_size);
+ encode_ptr(&dst, isamb->file[i].head.block_max);
+ encode_ptr(&dst, isamb->file[i].head.free_list);
+ memset(dst, '\0', b_size); /* ensure no random bytes are written */
+
+ len = dst - hbuf;
+
+ /* print exactly 16 bytes (including trailing 0) */
+ sprintf(hbuf, "isamb%02d %02d %02d\r\n", major, minor, len);
+
+ bf_write (isamb->file[i].bf, pos, 0, 0, hbuf);
+
+ for (left = len - b_size; left > 0; left = left - b_size)
+ {
+ pos++;
+ bf_write (isamb->file[i].bf, pos, 0, 0, hbuf + pos*b_size);
+ }
+ }
+ bf_close (isamb->file[i].bf);
+ }
+ xfree (isamb->file);
xfree (isamb->method);
xfree (isamb);
}
-struct ISAMB_block *open_block (ISAMB b, ISAMC_P pos)
+/* open_block: read one block at pos.
+ Decode leading sys bytes .. consisting of
+ Offset:Meaning
+ 0: leader byte, != 0 leaf, == 0, non-leaf
+ 1-2: used size of block
+ 3-7*: number of items and all children
+
+ * Reserve 5 bytes for large block sizes. 1 for small ones .. Number
+ of items. We can thus have at most 2^40 nodes.
+*/
+static struct ISAMB_block *open_block (ISAMB b, ISAMC_P pos)
{
+ int cat = (int) (pos&CAT_MASK);
+ const char *src;
+ int offset = b->file[cat].head.block_offset;
struct ISAMB_block *p;
if (!pos)
return 0;
p = xmalloc (sizeof(*p));
p->pos = pos;
- p->bytes = xmalloc (b->head.block_size);
- bf_read (b->bf, pos, 0, 0, p->bytes);
- p->leaf = p->bytes[0];
- p->size = p->bytes[1] + 256 * p->bytes[2];
- p->offset = ISAMB_DATA_OFFSET;
+ p->cat = (int) (pos & CAT_MASK);
+ p->buf = xmalloc (b->file[cat].head.block_size);
+ p->cbuf = 0;
+
+ if (!get_block (b, pos, p->buf, 0))
+ {
+ yaz_log (b->log_io, "bf_read: open_block");
+ if (!bf_read (b->file[cat].bf, pos/CAT_MAX, 0, 0, p->buf))
+ {
+ yaz_log (YLOG_FATAL, "isamb: read fail for pos=%ld block=%ld",
+ (long) pos, (long) pos/CAT_MAX);
+ abort();
+ }
+ }
+ p->bytes = p->buf + offset;
+ p->leaf = p->buf[0];
+ p->size = (p->buf[1] + 256 * p->buf[2]) - offset;
+ if (p->size < 0)
+ {
+ yaz_log (YLOG_FATAL, "Bad block size %d in pos=" ZINT_FORMAT "\n",
+ p->size, pos);
+ }
+ assert (p->size >= 0);
+ src = p->buf + 3;
+ decode_ptr(&src, &p->no_items);
+
+ p->offset = 0;
p->dirty = 0;
- p->decodeClientData = (*b->method->code_start)(ISAMC_DECODE);
+ p->deleted = 0;
+ p->decodeClientData = (*b->method->codec.start)();
return p;
}
-struct ISAMB_block *new_block (ISAMB b, int leaf)
+struct ISAMB_block *new_block (ISAMB b, int leaf, int cat)
{
struct ISAMB_block *p;
-
+
p = xmalloc (sizeof(*p));
- p->pos = b->head.last_block++;
- b->head_dirty = 1;
- p->bytes = xmalloc (b->head.block_size);
- memset (p->bytes, 0, b->head.block_size);
+ p->buf = xmalloc (b->file[cat].head.block_size);
+
+ if (!b->file[cat].head.free_list)
+ {
+ zint block_no;
+ block_no = b->file[cat].head.last_block++;
+ p->pos = block_no * CAT_MAX + cat;
+ }
+ else
+ {
+ p->pos = b->file[cat].head.free_list;
+ assert((p->pos & CAT_MASK) == cat);
+ if (!get_block (b, p->pos, p->buf, 0))
+ {
+ yaz_log (b->log_io, "bf_read: new_block");
+ if (!bf_read (b->file[cat].bf, p->pos/CAT_MAX, 0, 0, p->buf))
+ {
+ yaz_log (YLOG_FATAL, "isamb: read fail for pos=%ld block=%ld",
+ (long) p->pos/CAT_MAX, (long) p->pos/CAT_MAX);
+ abort ();
+ }
+ }
+ yaz_log (b->log_freelist, "got block " ZINT_FORMAT " from freelist %d:" ZINT_FORMAT, p->pos,
+ cat, p->pos/CAT_MAX);
+ memcpy (&b->file[cat].head.free_list, p->buf, sizeof(int));
+ }
+ p->cat = cat;
+ b->file[cat].head_dirty = 1;
+ memset (p->buf, 0, b->file[cat].head.block_size);
+ p->bytes = p->buf + b->file[cat].head.block_offset;
p->leaf = leaf;
- p->size = ISAMB_DATA_OFFSET;
+ p->size = 0;
p->dirty = 1;
- p->offset = ISAMB_DATA_OFFSET;
- p->decodeClientData = (*b->method->code_start)(ISAMC_DECODE);
+ p->deleted = 0;
+ p->offset = 0;
+ p->no_items = 0;
+ p->decodeClientData = (*b->method->codec.start)();
return p;
}
-void close_block (ISAMB b, struct ISAMB_block *p)
+struct ISAMB_block *new_leaf (ISAMB b, int cat)
{
- if (!p)
- return;
- if (p->dirty)
- {
- p->bytes[0] = p->leaf;
- p->bytes[1] = p->size & 255;
- p->bytes[2] = p->size >> 8;
- bf_write (b->bf, p->pos, 0, 0, p->bytes);
- }
- (*b->method->code_stop)(ISAMC_DECODE, p->decodeClientData);
- xfree (p->bytes);
- xfree (p);
+ return new_block (b, 1, cat);
}
-void insert_sub (ISAMB b, struct ISAMB_block *p, const void *new_item,
- struct ISAMB_block **sp,
- void *sub_item, int *sub_size);
-void insert_leaf (ISAMB b, struct ISAMB_block *p, const void *new_item,
- struct ISAMB_block **sp,
- void *sub_item, int *sub_size)
+struct ISAMB_block *new_int (ISAMB b, int cat)
{
- char dst_buf[2048];
- char *dst = dst_buf;
- char *src = p->bytes + ISAMB_DATA_OFFSET;
- char *endp = p->bytes + p->size;
- void *c1 = (*b->method->code_start)(ISAMC_DECODE);
- void *c2 = (*b->method->code_start)(ISAMC_ENCODE);
- char *half1 = 0;
- char *half2 = 0;
- char *cut = dst_buf + p->size / 2;
- char cut_item_buf[256];
- int cut_item_size = 0;
+ return new_block (b, 0, cat);
+}
- while (src != endp)
+static void check_block (ISAMB b, struct ISAMB_block *p)
+{
+ assert(b); /* mostly to make the compiler shut up about unused b */
+ if (p->leaf)
{
- char file_item_buf[256];
- char *file_item = file_item_buf;
-
- (*b->method->code_item)(ISAMC_DECODE, c1, &file_item, &src);
- if (new_item)
+ ;
+ }
+ else
+ {
+ /* sanity check */
+ char *startp = p->bytes;
+ const char *src = startp;
+ char *endp = p->bytes + p->size;
+ ISAMB_P pos;
+
+ decode_ptr (&src, &pos);
+ assert ((pos&CAT_MASK) == p->cat);
+ while (src != endp)
{
- int d = (*b->method->compare_item)(file_item_buf, new_item);
- if (d > 0)
- {
- char *item_ptr = (char*) new_item;
- (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &item_ptr);
- new_item = 0;
- p->dirty = 1;
- }
- else if (d == 0)
- {
- new_item = 0;
- }
+ zint item_len;
+ decode_ptr (&src, &item_len);
+ assert (item_len > 0 && item_len < 80);
+ src += item_len;
+ decode_ptr (&src, &pos);
+ if ((pos&CAT_MASK) != p->cat)
+ {
+ assert ((pos&CAT_MASK) == p->cat);
+ }
}
-
- if (!half1 && dst > cut)
- {
- half1 = dst; /* candidate for splitting */
-
- file_item = file_item_buf;
- (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &file_item);
-
- cut_item_size = file_item - file_item_buf;
- memcpy (cut_item_buf, file_item_buf, cut_item_size);
+ }
+}
- half2 = dst;
- }
- else
+void close_block (ISAMB b, struct ISAMB_block *p)
+{
+ if (!p)
+ return;
+ if (p->deleted)
+ {
+ yaz_log (b->log_freelist, "release block " ZINT_FORMAT " from freelist %d:" ZINT_FORMAT,
+ p->pos, p->cat, p->pos/CAT_MAX);
+ memcpy (p->buf, &b->file[p->cat].head.free_list, sizeof(int));
+ b->file[p->cat].head.free_list = p->pos;
+ if (!get_block (b, p->pos, p->buf, 1))
{
- file_item = file_item_buf;
- (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &file_item);
+ yaz_log (b->log_io, "bf_write: close_block (deleted)");
+ bf_write (b->file[p->cat].bf, p->pos/CAT_MAX, 0, 0, p->buf);
}
}
- if (new_item)
- {
- char *item_ptr = (char*) new_item;
- (*b->method->code_item)(ISAMC_ENCODE, c2, &dst, &item_ptr);
- new_item = 0;
- p->dirty = 1;
- }
- p->size = dst - dst_buf + ISAMB_DATA_OFFSET;
- if (p->size > b->head.block_size)
- {
- char *first_dst;
- char *cut_item = cut_item_buf;
-
- /* first half */
- p->size = half1 - dst_buf + ISAMB_DATA_OFFSET;
- memcpy (p->bytes+ISAMB_DATA_OFFSET, dst_buf, half1 - dst_buf);
-
- /* second half */
- *sp = new_block (b, 1);
-
- (*b->method->code_reset)(c2);
-
- first_dst = (*sp)->bytes + ISAMB_DATA_OFFSET;
-
- (*b->method->code_item)(ISAMC_ENCODE, c2, &first_dst, &cut_item);
-
- memcpy (first_dst, half2, dst - half2);
-
- (*sp)->size = (first_dst - (char*) (*sp)->bytes) + (dst - half2);
- (*sp)->dirty = 1;
- p->dirty = 1;
- memcpy (sub_item, cut_item_buf, cut_item_size);
- *sub_size = cut_item_size;
-
- yaz_log (LOG_LOG, "l split %d / %d", p->size, (*sp)->size);
-
- }
- else
+ else if (p->dirty)
{
- assert (p->size > ISAMB_DATA_OFFSET);
- assert (p->size <= b->head.block_size);
- memcpy (p->bytes+ISAMB_DATA_OFFSET, dst_buf, dst - dst_buf);
- *sp = 0;
+ int offset = b->file[p->cat].head.block_offset;
+ int size = p->size + offset;
+ char *dst = p->buf + 3;
+ assert (p->size >= 0);
+
+ /* memset becuase encode_ptr usually does not write all bytes */
+ memset(p->buf, 0, b->file[p->cat].head.block_offset);
+ p->buf[0] = p->leaf;
+ p->buf[1] = size & 255;
+ p->buf[2] = size >> 8;
+ encode_ptr(&dst, p->no_items);
+ check_block(b, p);
+ if (!get_block (b, p->pos, p->buf, 1))
+ {
+ yaz_log (b->log_io, "bf_write: close_block");
+ bf_write (b->file[p->cat].bf, p->pos/CAT_MAX, 0, 0, p->buf);
+ }
}
- (*b->method->code_stop)(ISAMC_DECODE, c1);
- (*b->method->code_stop)(ISAMC_ENCODE, c2);
+ (*b->method->codec.stop)(p->decodeClientData);
+ xfree (p->buf);
+ xfree (p);
}
-void insert_int (ISAMB b, struct ISAMB_block *p, const void *new_item,
- struct ISAMB_block **sp,
- void *split_item, int *split_size)
+int insert_sub (ISAMB b, struct ISAMB_block **p,
+ void *new_item, int *mode,
+ ISAMC_I *stream,
+ struct ISAMB_block **sp,
+ void *sub_item, int *sub_size,
+ const void *max_item);
+
+int insert_int (ISAMB b, struct ISAMB_block *p, void *lookahead_item,
+ int *mode,
+ ISAMC_I *stream, struct ISAMB_block **sp,
+ void *split_item, int *split_size, const void *last_max_item)
{
- char *startp = p->bytes + ISAMB_DATA_OFFSET;
- char *src = startp;
+ char *startp = p->bytes;
+ const char *src = startp;
char *endp = p->bytes + p->size;
- int pos;
+ ISAMB_P pos;
struct ISAMB_block *sub_p1 = 0, *sub_p2 = 0;
- char sub_item[256];
+ char sub_item[DST_ITEM_MAX];
int sub_size;
+ int more = 0;
+ zint diff_terms = 0;
*sp = 0;
+ assert(p->size >= 0);
decode_ptr (&src, &pos);
while (src != endp)
{
- int item_len;
+ zint item_len;
int d;
+ const char *src0 = src;
decode_ptr (&src, &item_len);
- d = (*b->method->compare_item)(src, new_item);
+ d = (*b->method->compare_item)(src, lookahead_item);
if (d > 0)
{
sub_p1 = open_block (b, pos);
assert (sub_p1);
- insert_sub (b, sub_p1, new_item, &sub_p2,
- sub_item, &sub_size);
+ diff_terms -= sub_p1->no_items;
+ more = insert_sub (b, &sub_p1, lookahead_item, mode,
+ stream, &sub_p2,
+ sub_item, &sub_size, src);
+ diff_terms += sub_p1->no_items;
+ src = src0;
break;
}
src += item_len;
{
sub_p1 = open_block (b, pos);
assert (sub_p1);
- insert_sub (b, sub_p1, new_item, &sub_p2,
- sub_item, &sub_size);
+ diff_terms -= sub_p1->no_items;
+ more = insert_sub (b, &sub_p1, lookahead_item, mode, stream, &sub_p2,
+ sub_item, &sub_size, last_max_item);
+ diff_terms += sub_p1->no_items;
}
if (sub_p2)
+ diff_terms += sub_p2->no_items;
+ if (diff_terms)
{
- char dst_buf[2048];
+ p->dirty = 1;
+ p->no_items += diff_terms;
+ }
+ if (sub_p2)
+ {
+ /* there was a split - must insert pointer in this one */
+ char dst_buf[DST_BUF_SIZE];
char *dst = dst_buf;
- assert (sub_size < 20);
+ assert (sub_size < 80 && sub_size > 1);
memcpy (dst, startp, src - startp);
memcpy (dst, src, endp - src);
dst += endp - src;
}
- p->size = dst - dst_buf + ISAMB_DATA_OFFSET;
- if (p->size <= b->head.block_size)
+ p->size = dst - dst_buf;
+ assert (p->size >= 0);
+
+
+ if (p->size <= b->file[p->cat].head.block_max)
{
memcpy (startp, dst_buf, dst - dst_buf);
}
else
{
+ struct ISAMB_block *sub_p3;
+ zint split_size_tmp;
+ zint no_items_first_half = 0;
int p_new_size;
- char *half;
+ const char *half;
src = dst_buf;
endp = dst;
- half = src + b->head.block_size/2;
+ half = src + b->file[p->cat].head.block_size/2;
decode_ptr (&src, &pos);
+
+ /* read sub block so we can get no_items for it */
+ sub_p3 = open_block(b, pos);
+ no_items_first_half += sub_p3->no_items;
+ close_block(b, sub_p3);
+
while (src <= half)
{
- decode_ptr (&src, split_size);
+ decode_ptr (&src, &split_size_tmp);
+ *split_size = (int) split_size_tmp;
+
src += *split_size;
decode_ptr (&src, &pos);
+
+ /* read sub block so we can get no_items for it */
+ sub_p3 = open_block(b, pos);
+ no_items_first_half += sub_p3->no_items;
+ close_block(b, sub_p3);
}
+ /* p is first half */
p_new_size = src - dst_buf;
- memcpy (p->bytes + ISAMB_DATA_OFFSET, dst_buf, p_new_size);
- p_new_size += ISAMB_DATA_OFFSET;
+ memcpy (p->bytes, dst_buf, p_new_size);
- decode_ptr (&src, split_size);
+ decode_ptr (&src, &split_size_tmp);
+ *split_size = (int) split_size_tmp;
memcpy (split_item, src, *split_size);
src += *split_size;
- *sp = new_block (b, 0);
+ /* *sp is second half */
+ *sp = new_int (b, p->cat);
(*sp)->size = endp - src;
- memcpy ((*sp)->bytes+ISAMB_DATA_OFFSET, src, (*sp)->size);
- (*sp)->size += ISAMB_DATA_OFFSET;
+ memcpy ((*sp)->bytes, src, (*sp)->size);
- yaz_log (LOG_LOG, "i split %d -> %d %d",
- p->size, p_new_size, (*sp)->size);
p->size = p_new_size;
+
+ /* adjust no_items in first&second half */
+ (*sp)->no_items = p->no_items - no_items_first_half;
+ p->no_items = no_items_first_half;
}
p->dirty = 1;
close_block (b, sub_p2);
}
close_block (b, sub_p1);
+ return more;
}
-void insert_sub (ISAMB b, struct ISAMB_block *p, const void *new_item,
+int insert_leaf (ISAMB b, struct ISAMB_block **sp1, void *lookahead_item,
+ int *lookahead_mode, ISAMC_I *stream,
+ struct ISAMB_block **sp2,
+ void *sub_item, int *sub_size,
+ const void *max_item)
+{
+ struct ISAMB_block *p = *sp1;
+ char *endp = 0;
+ const char *src = 0;
+ char dst_buf[DST_BUF_SIZE], *dst = dst_buf;
+ int new_size;
+ void *c1 = (*b->method->codec.start)();
+ void *c2 = (*b->method->codec.start)();
+ int more = 1;
+ int quater = b->file[b->no_cat-1].head.block_max / 4;
+ char *mid_cut = dst_buf + quater * 2;
+ char *tail_cut = dst_buf + quater * 3;
+ char *maxp = dst_buf + b->file[b->no_cat-1].head.block_max;
+ char *half1 = 0;
+ char *half2 = 0;
+ char cut_item_buf[DST_ITEM_MAX];
+ int cut_item_size = 0;
+ int no_items = 0; /* number of items (total) */
+ int no_items_1 = 0; /* number of items (first half) */
+
+ if (p && p->size)
+ {
+ char file_item_buf[DST_ITEM_MAX];
+ char *file_item = file_item_buf;
+
+ src = p->bytes;
+ endp = p->bytes + p->size;
+ (*b->method->codec.decode)(c1, &file_item, &src);
+ while (1)
+ {
+ const char *dst_item = 0; /* resulting item to be inserted */
+ char *lookahead_next;
+ int d = -1;
+
+ if (lookahead_item)
+ d = (*b->method->compare_item)(file_item_buf, lookahead_item);
+
+ /* d now holds comparison between existing file item and
+ lookahead item
+ d = 0: equal
+ d > 0: lookahead before file
+ d < 0: lookahead after file
+ */
+ if (d > 0)
+ {
+ /* lookahead must be inserted */
+ dst_item = lookahead_item;
+ /* if this is not an insertion, it's really bad .. */
+ if (!*lookahead_mode)
+ {
+ yaz_log (YLOG_WARN, "isamb: Inconsistent register (1)");
+ assert (*lookahead_mode);
+ }
+ }
+ else
+ dst_item = file_item_buf;
+
+
+ if (!*lookahead_mode && d == 0)
+ {
+ /* it's a deletion and they match so there is nothing to be
+ inserted anyway .. But mark the thing bad (file item
+ was part of input.. The item will not be part of output */
+ p->dirty = 1;
+ }
+ else if (!half1 && dst > mid_cut)
+ {
+ /* we have reached the splitting point for the first time */
+ const char *dst_item_0 = dst_item;
+ half1 = dst; /* candidate for splitting */
+
+ /* encode the resulting item */
+ (*b->method->codec.encode)(c2, &dst, &dst_item);
+
+ cut_item_size = dst_item - dst_item_0;
+ assert(cut_item_size > 0);
+ memcpy (cut_item_buf, dst_item_0, cut_item_size);
+
+ half2 = dst;
+ no_items_1 = no_items;
+ no_items++;
+ }
+ else
+ {
+ /* encode the resulting item */
+ (*b->method->codec.encode)(c2, &dst, &dst_item);
+ no_items++;
+ }
+
+ /* now move "pointers" .. result has been encoded .. */
+ if (d > 0)
+ {
+ /* we must move the lookahead pointer */
+
+ if (dst > maxp)
+ /* no more room. Mark lookahead as "gone".. */
+ lookahead_item = 0;
+ else
+ {
+ /* move it really.. */
+ lookahead_next = lookahead_item;
+ if (!(*stream->read_item)(stream->clientData,
+ &lookahead_next,
+ lookahead_mode))
+ {
+ /* end of stream reached: no "more" and no lookahead */
+ lookahead_item = 0;
+ more = 0;
+ }
+ if (lookahead_item && max_item &&
+ (*b->method->compare_item)(max_item, lookahead_item) <= 0)
+ {
+ /* the lookahead goes beyond what we allow in this
+ leaf. Mark it as "gone" */
+ lookahead_item = 0;
+ }
+
+ p->dirty = 1;
+ }
+ }
+ else if (d == 0)
+ {
+ /* exact match .. move both pointers */
+
+ lookahead_next = lookahead_item;
+ if (!(*stream->read_item)(stream->clientData,
+ &lookahead_next, lookahead_mode))
+ {
+ lookahead_item = 0;
+ more = 0;
+ }
+ if (src == endp)
+ break; /* end of file stream reached .. */
+ file_item = file_item_buf; /* move file pointer */
+ (*b->method->codec.decode)(c1, &file_item, &src);
+ }
+ else
+ {
+ /* file pointer must be moved */
+ if (src == endp)
+ break;
+ file_item = file_item_buf;
+ (*b->method->codec.decode)(c1, &file_item, &src);
+ }
+ }
+ }
+ maxp = dst_buf + b->file[b->no_cat-1].head.block_max + quater;
+ /* this loop runs when we are "appending" to a leaf page. That is
+ either it's empty (new) or all file items have been read in
+ previous loop */
+ while (lookahead_item)
+ {
+ char *dst_item;
+ const char *src = lookahead_item;
+ char *dst_0 = dst;
+
+ /* compare lookahead with max item */
+ if (max_item &&
+ (*b->method->compare_item)(max_item, lookahead_item) <= 0)
+ {
+ /* stop if we have reached the value of max item */
+ break;
+ }
+ if (!*lookahead_mode)
+ {
+ /* this is append. So a delete is bad */
+ yaz_log (YLOG_WARN, "isamb: Inconsistent register (2)");
+ abort();
+ }
+ else if (!half1 && dst > tail_cut)
+ {
+ const char *src_0 = src;
+ half1 = dst; /* candidate for splitting */
+
+ (*b->method->codec.encode)(c2, &dst, &src);
+
+ cut_item_size = src - src_0;
+ assert(cut_item_size > 0);
+ memcpy (cut_item_buf, src_0, cut_item_size);
+
+ no_items_1 = no_items;
+ half2 = dst;
+ }
+ else
+ (*b->method->codec.encode)(c2, &dst, &src);
+
+ if (dst > maxp)
+ {
+ dst = dst_0;
+ break;
+ }
+ no_items++;
+ if (p)
+ p->dirty = 1;
+ dst_item = lookahead_item;
+ if (!(*stream->read_item)(stream->clientData, &dst_item,
+ lookahead_mode))
+ {
+ lookahead_item = 0;
+ more = 0;
+ }
+ }
+ new_size = dst - dst_buf;
+ if (p && p->cat != b->no_cat-1 &&
+ new_size > b->file[p->cat].head.block_max)
+ {
+ /* non-btree block will be removed */
+ p->deleted = 1;
+ close_block (b, p);
+ /* delete it too!! */
+ p = 0; /* make a new one anyway */
+ }
+ if (!p)
+ { /* must create a new one */
+ int i;
+ for (i = 0; i < b->no_cat; i++)
+ if (new_size <= b->file[i].head.block_max)
+ break;
+ if (i == b->no_cat)
+ i = b->no_cat - 1;
+ p = new_leaf (b, i);
+ }
+ if (new_size > b->file[p->cat].head.block_max)
+ {
+ char *first_dst;
+ const char *cut_item = cut_item_buf;
+
+ assert (half1);
+ assert (half2);
+
+ assert(cut_item_size > 0);
+
+ /* first half */
+ p->size = half1 - dst_buf;
+ memcpy (p->bytes, dst_buf, half1 - dst_buf);
+ p->no_items = no_items_1;
+
+ /* second half */
+ *sp2 = new_leaf (b, p->cat);
+
+ (*b->method->codec.reset)(c2);
+
+ first_dst = (*sp2)->bytes;
+
+ (*b->method->codec.encode)(c2, &first_dst, &cut_item);
+
+ memcpy (first_dst, half2, dst - half2);
+
+ (*sp2)->size = (first_dst - (*sp2)->bytes) + (dst - half2);
+ (*sp2)->no_items = no_items - no_items_1;
+ (*sp2)->dirty = 1;
+ p->dirty = 1;
+ memcpy (sub_item, cut_item_buf, cut_item_size);
+ *sub_size = cut_item_size;
+ }
+ else
+ {
+ memcpy (p->bytes, dst_buf, dst - dst_buf);
+ p->size = new_size;
+ p->no_items = no_items;
+ }
+ (*b->method->codec.stop)(c1);
+ (*b->method->codec.stop)(c2);
+ *sp1 = p;
+ return more;
+}
+
+int insert_sub (ISAMB b, struct ISAMB_block **p, void *new_item,
+ int *mode,
+ ISAMC_I *stream,
struct ISAMB_block **sp,
- void *sub_item, int *sub_size)
+ void *sub_item, int *sub_size,
+ const void *max_item)
{
- if (p->leaf)
- insert_leaf (b, p, new_item, sp, sub_item, sub_size);
+ if (!*p || (*p)->leaf)
+ return insert_leaf (b, p, new_item, mode, stream, sp, sub_item,
+ sub_size, max_item);
else
- insert_int (b, p, new_item, sp, sub_item, sub_size);
+ return insert_int (b, *p, new_item, mode, stream, sp, sub_item,
+ sub_size, max_item);
}
-int isamb_insert_one (ISAMB b, const void *item, ISAMC_P pos)
+int isamb_unlink (ISAMB b, ISAMC_P pos)
{
- struct ISAMB_block *p, *sp = 0;
- char sub_item[256];
- int sub_size;
+ struct ISAMB_block *p1;
if (!pos)
- p = new_block (b, 1);
- else
- p = open_block (b, pos);
- if (!p)
- return -1;
-
- insert_sub (b, p, item, &sp, sub_item, &sub_size);
- if (sp)
- { /* increase level of tree by one */
- struct ISAMB_block *p2 = new_block (b, 0);
- char *dst = p2->bytes + p2->size;
-
- encode_ptr (&dst, p->pos);
- assert (sub_size < 20);
- encode_ptr (&dst, sub_size);
- memcpy (dst, sub_item, sub_size);
- dst += sub_size;
- encode_ptr (&dst, sp->pos);
+ return 0;
+ p1 = open_block(b, pos);
+ p1->deleted = 1;
+ if (!p1->leaf)
+ {
+ zint sub_p;
+ zint item_len;
+ const char *src = p1->bytes + p1->offset;
- p2->size = dst - (char*) p2->bytes;
- pos = p2->pos; /* return new super page */
- close_block (b, sp);
- close_block (b, p2);
+ decode_ptr(&src, &sub_p);
+ isamb_unlink(b, sub_p);
+
+ while (src != p1->bytes + p1->size)
+ {
+ decode_ptr(&src, &item_len);
+ src += item_len;
+ decode_ptr(&src, &sub_p);
+ isamb_unlink(b, sub_p);
+ }
}
- else
- pos = p->pos; /* return current one (again) */
- close_block (b, p);
- return pos;
+ close_block(b, p1);
+ return 0;
}
-ISAMB_P isamb_merge (ISAMB b, ISAMB_P pos, ISAMC_I data)
+ISAMB_P isamb_merge (ISAMB b, ISAMC_P pos, ISAMC_I *stream)
{
+ char item_buf[DST_ITEM_MAX];
+ char *item_ptr;
int i_mode;
- char item_buf[256];
- char *item_ptr = item_buf;
- while ((*data->read_item)(data->clientData, &item_ptr, &i_mode))
+ int more;
+ int must_delete = 0;
+
+ if (b->cache < 0)
+ {
+ int more = 1;
+ while (more)
+ {
+ item_ptr = item_buf;
+ more =
+ (*stream->read_item)(stream->clientData, &item_ptr, &i_mode);
+ }
+ return 1;
+ }
+ item_ptr = item_buf;
+ more = (*stream->read_item)(stream->clientData, &item_ptr, &i_mode);
+ while (more)
+ {
+ struct ISAMB_block *p = 0, *sp = 0;
+ char sub_item[DST_ITEM_MAX];
+ int sub_size;
+
+ if (pos)
+ p = open_block (b, pos);
+ more = insert_sub (b, &p, item_buf, &i_mode, stream, &sp,
+ sub_item, &sub_size, 0);
+ if (sp)
+ { /* increase level of tree by one */
+ struct ISAMB_block *p2 = new_int (b, p->cat);
+ char *dst = p2->bytes + p2->size;
+
+ encode_ptr (&dst, p->pos);
+ assert (sub_size < 40);
+ encode_ptr (&dst, sub_size);
+ memcpy (dst, sub_item, sub_size);
+ dst += sub_size;
+ encode_ptr (&dst, sp->pos);
+
+ p2->size = dst - p2->bytes;
+ p2->no_items = p->no_items + sp->no_items;
+ pos = p2->pos; /* return new super page */
+ close_block (b, sp);
+ close_block (b, p2);
+ }
+ else
+ {
+ pos = p->pos; /* return current one (again) */
+ }
+ if (p->no_items == 0)
+ must_delete = 1;
+ else
+ must_delete = 0;
+ close_block (b, p);
+ }
+ if (must_delete)
{
- item_ptr = item_buf;
- pos = isamb_insert_one (b, item_buf, pos);
+ isamb_unlink(b, pos);
+ return 0;
}
return pos;
}
-ISAMB_PP isamb_pp_open (ISAMB isamb, ISAMB_P pos)
+ISAMB_PP isamb_pp_open_x (ISAMB isamb, ISAMB_P pos, int *level, int scope)
{
ISAMB_PP pp = xmalloc (sizeof(*pp));
+ int i;
pp->isamb = isamb;
- pp->block = xmalloc (10 * sizeof(*pp->block));
+ pp->block = xmalloc (ISAMB_MAX_LEVEL * sizeof(*pp->block));
+ pp->pos = pos;
pp->level = 0;
+ pp->maxlevel=0;
+ pp->total_size = 0;
+ pp->no_blocks = 0;
+ pp->skipped_numbers=0;
+ pp->returned_numbers=0;
+ pp->scope=scope;
+ for (i=0;i<ISAMB_MAX_LEVEL;i++)
+ pp->skipped_nodes[i] = pp->accessed_nodes[i]=0;
while (1)
{
struct ISAMB_block *p = open_block (isamb, pos);
- char *src = p->bytes + p->offset;
+ const char *src = p->bytes + p->offset;
pp->block[pp->level] = p;
- if (p->bytes[0]) /* leaf */
+ pp->total_size += p->size;
+ pp->no_blocks++;
+ if (p->leaf)
break;
+
decode_ptr (&src, &pos);
- p->offset = src - (char*) p->bytes;
+ p->offset = src - p->bytes;
pp->level++;
+ pp->accessed_nodes[pp->level]++;
}
pp->block[pp->level+1] = 0;
+ pp->maxlevel=pp->level;
+ if (level)
+ *level = pp->level;
return pp;
}
-void isamb_pp_close (ISAMB_PP pp)
+ISAMB_PP isamb_pp_open (ISAMB isamb, ISAMB_P pos, int scope)
+{
+ return isamb_pp_open_x (isamb, pos, 0, scope);
+}
+
+void isamb_pp_close_x (ISAMB_PP pp, int *size, int *blocks)
{
int i;
if (!pp)
return;
+ yaz_log(YLOG_DEBUG,"isamb_pp_close lev=%d returned "ZINT_FORMAT" values,"
+ "skipped "ZINT_FORMAT,
+ pp->maxlevel, pp->skipped_numbers, pp->returned_numbers);
+ for (i=pp->maxlevel;i>=0;i--)
+ if ( pp->skipped_nodes[i] || pp->accessed_nodes[i])
+ yaz_log(YLOG_DEBUG,"isamb_pp_close level leaf-%d: "
+ ZINT_FORMAT" read, "ZINT_FORMAT" skipped", i,
+ pp->accessed_nodes[i], pp->skipped_nodes[i]);
+ pp->isamb->skipped_numbers += pp->skipped_numbers;
+ pp->isamb->returned_numbers += pp->returned_numbers;
+ for (i=pp->maxlevel;i>=0;i--)
+ {
+ pp->isamb->accessed_nodes[i] += pp->accessed_nodes[i];
+ pp->isamb->skipped_nodes[i] += pp->skipped_nodes[i];
+ }
+ if (size)
+ *size = pp->total_size;
+ if (blocks)
+ *blocks = pp->no_blocks;
for (i = 0; i <= pp->level; i++)
close_block (pp->isamb, pp->block[i]);
xfree (pp->block);
xfree (pp);
}
+int isamb_block_info (ISAMB isamb, int cat)
+{
+ if (cat >= 0 && cat < isamb->no_cat)
+ return isamb->file[cat].head.block_size;
+ return -1;
+}
+
+void isamb_pp_close (ISAMB_PP pp)
+{
+ isamb_pp_close_x (pp, 0, 0);
+}
+
+/* simple recursive dumper .. */
+static void isamb_dump_r (ISAMB b, ISAMB_P pos, void (*pr)(const char *str),
+ int level)
+{
+ char buf[1024];
+ char prefix_str[1024];
+ if (pos)
+ {
+ struct ISAMB_block *p = open_block (b, pos);
+ sprintf(prefix_str, "%*s " ZINT_FORMAT " cat=%d size=%d max=%d items="
+ ZINT_FORMAT, level*2, "",
+ pos, p->cat, p->size, b->file[p->cat].head.block_max,
+ p->no_items);
+ (*pr)(prefix_str);
+ sprintf(prefix_str, "%*s " ZINT_FORMAT, level*2, "", pos);
+ if (p->leaf)
+ {
+ while (p->offset < p->size)
+ {
+ const char *src = p->bytes + p->offset;
+ char *dst = buf;
+ (*b->method->codec.decode)(p->decodeClientData, &dst, &src);
+ (*b->method->log_item)(YLOG_DEBUG, buf, prefix_str);
+ p->offset = src - (char*) p->bytes;
+ }
+ assert(p->offset == p->size);
+ }
+ else
+ {
+ const char *src = p->bytes + p->offset;
+ ISAMB_P sub;
+ zint item_len;
+
+ decode_ptr (&src, &sub);
+ p->offset = src - (char*) p->bytes;
+
+ isamb_dump_r(b, sub, pr, level+1);
+
+ while (p->offset < p->size)
+ {
+ decode_ptr (&src, &item_len);
+ (*b->method->log_item)(YLOG_DEBUG, src, prefix_str);
+ src += item_len;
+ decode_ptr (&src, &sub);
+
+ p->offset = src - (char*) p->bytes;
+
+ isamb_dump_r(b, sub, pr, level+1);
+ }
+ }
+ close_block(b,p);
+ }
+}
+
+void isamb_dump (ISAMB b, ISAMB_P pos, void (*pr)(const char *str))
+{
+ isamb_dump_r(b, pos, pr, 0);
+}
+
int isamb_pp_read (ISAMB_PP pp, void *buf)
{
- char *dst = buf;
- char *src;
- struct ISAMB_block *p = pp->block[pp->level];
- if (!p)
- return 0;
+ return isamb_pp_forward(pp, buf, 0);
+}
+
- while (p->offset == p->size)
+static int isamb_pp_on_right_node(ISAMB_PP pp, int level, const void *untilbuf)
+{ /* looks one node higher to see if we should be on this node at all */
+ /* useful in backing off quickly, and in avoiding tail descends */
+ /* call with pp->level to begin with */
+ struct ISAMB_block *p;
+ int cmp;
+ const char *src;
+ zint item_len;
+ assert(level>=0);
+ if ( level == 0) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_on_right returning true for root");
+#endif
+ return 1; /* we can never skip the root node */
+ }
+ level--;
+ p=pp->block[level];
+ assert(p->offset <= p->size);
+ if (p->offset < p->size )
{
- int pos, item_len;
- while (p->offset == p->size)
- {
- if (pp->level == 0)
- return 0;
- close_block (pp->isamb, pp->block[pp->level]);
- pp->block[pp->level] = 0;
- (pp->level)--;
- p = pp->block[pp->level];
- assert (p->bytes[0] == 0); /* must be int */
+ assert(p->offset>0);
+ src=p->bytes + p->offset;
+ decode_ptr(&src, &item_len);
+#if ISAMB_DEBUG
+ (*pp->isamb->method->codec.log_item)(YLOG_DEBUG,untilbuf,"on_leaf: until");
+ (*pp->isamb->method->codec.log_item)(YLOG_DEBUG,src,"on_leaf: value");
+#endif
+ cmp=(*pp->isamb->method->compare_item)(untilbuf,src);
+ if (cmp<pp->scope) { /* cmp<2 */
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_on_right returning true "
+ "cmp=%d lev=%d ofs=%d",cmp,level,p->offset);
+#endif
+ return 1;
}
- src = p->bytes + p->offset;
-
- decode_ptr (&src, &item_len);
+ else {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_on_right returning false "
+ "cmp=%d lev=%d ofs=%d",cmp,level,p->offset);
+#endif
+ return 0;
+ }
+ }
+ else {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_on_right at tail, looking higher "
+ "lev=%d",level);
+#endif
+ return isamb_pp_on_right_node(pp, level, untilbuf);
+ }
+} /* isamb_pp_on_right_node */
+
+static int isamb_pp_read_on_leaf(ISAMB_PP pp, void *buf)
+{ /* reads the next item on the current leaf, returns 0 if end of leaf*/
+ struct ISAMB_block *p = pp->block[pp->level];
+ char *dst;
+ const char *src;
+ assert(pp);
+ assert(buf);
+ if (p->offset == p->size) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_read_on_leaf returning 0 on node %d",p->pos);
+#endif
+ return 0; /* at end of leaf */
+ }
+ src=p->bytes + p->offset;
+ dst=buf;
+ (*pp->isamb->method->codec.decode)(p->decodeClientData,&dst, &src);
+ p->offset = src - (char*) p->bytes;
+#if ISAMB_DEBUG
+ (*pp->isamb->method->codec.log_item)(YLOG_DEBUG, buf, "read_on_leaf returning 1");
+#endif
+ pp->returned_numbers++;
+ return 1;
+} /* read_on_leaf */
+
+static int isamb_pp_forward_on_leaf(ISAMB_PP pp, void *buf, const void *untilbuf)
+{ /* forwards on the current leaf, returns 0 if not found */
+ int cmp;
+ int skips=0;
+ while (1){
+ if (!isamb_pp_read_on_leaf(pp,buf))
+ return 0;
+ /* FIXME - this is an extra function call, inline the read? */
+ cmp=(*pp->isamb->method->compare_item)(untilbuf,buf);
+ if (cmp <pp->scope){ /* cmp<2 found a good one */
+#if ISAMB_DEBUG
+ if (skips)
+ yaz_log(YLOG_DEBUG, "isam_pp_fwd_on_leaf skipped %d items",skips);
+#endif
+ pp->returned_numbers++;
+ return 1;
+ }
+ if (!skips)
+ if (!isamb_pp_on_right_node(pp, pp->level, untilbuf))
+ return 0; /* never mind the rest of this leaf */
+ pp->skipped_numbers++;
+ skips++;
+ }
+} /* forward_on_leaf */
+
+static int isamb_pp_climb_level(ISAMB_PP pp, ISAMB_P *pos)
+{ /* climbs higher in the tree, until finds a level with data left */
+ /* returns the node to (consider to) descend to in *pos) */
+ struct ISAMB_block *p = pp->block[pp->level];
+ const char *src;
+ zint item_len;
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_climb_level starting "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ assert(pp->level >= 0);
+ assert(p->offset <= p->size);
+ if (pp->level==0)
+ {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_climb_level returning 0 at root");
+#endif
+ return 0;
+ }
+ assert(pp->level>0);
+ close_block(pp->isamb, pp->block[pp->level]);
+ pp->block[pp->level]=0;
+ (pp->level)--;
+ p=pp->block[pp->level];
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_climb_level climbed to level %d node %d ofs=%d",
+ pp->level, p->pos, p->offset);
+#endif
+ assert(!p->leaf);
+ assert(p->offset <= p->size);
+ if (p->offset == p->size ) {
+ /* we came from the last pointer, climb on */
+ if (!isamb_pp_climb_level(pp,pos))
+ return 0;
+ p=pp->block[pp->level];
+ }
+ else
+ {
+ /* skip the child we just came from */
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isam_pp_climb_level: skipping lev=%d ofs=%d sz=%d",
+ pp->level, p->offset, p->size);
+#endif
+ assert (p->offset < p->size );
+ src=p->bytes + p->offset;
+ decode_ptr(&src, &item_len);
src += item_len;
- decode_ptr (&src, &pos);
-
- p->offset = src - (char*) p->bytes;
+ decode_ptr(&src, pos);
+ p->offset=src - (char *)p->bytes;
+
+ }
+ return 1;
+} /* climb_level */
- ++(pp->level);
-
- while (1)
+
+static zint isamb_pp_forward_unode(ISAMB_PP pp, zint pos, const void *untilbuf)
+{ /* scans a upper node until it finds a child <= untilbuf */
+ /* pp points to the key value, as always. pos is the child read from */
+ /* the buffer */
+ /* if all values are too small, returns the last child in the node */
+ /* FIXME - this can be detected, and avoided by looking at the */
+ /* parent node, but that gets messy. Presumably the cost is */
+ /* pretty low anyway */
+ struct ISAMB_block *p = pp->block[pp->level];
+ const char *src=p->bytes + p->offset;
+ zint item_len;
+ int cmp;
+ zint nxtpos;
+#if ISAMB_DEBUG
+ int skips=0;
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward_unode starting "
+ "at level %d node %d ofs=%di sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ assert(!p->leaf);
+ assert(p->offset <= p->size);
+ if (p->offset == p->size) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward_unode returning at end "
+ "at level %d node %d ofs=%di sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return pos; /* already at the end of it */
+ }
+ while(p->offset < p->size) {
+ decode_ptr(&src,&item_len);
+ cmp=(*pp->isamb->method->compare_item)(untilbuf,src);
+ src+=item_len;
+ decode_ptr(&src,&nxtpos);
+ if (cmp<pp->scope) /* cmp<2 */
{
- pp->block[pp->level] = p = open_block (pp->isamb, pos);
-
- if (p->bytes[0]) /* leaf */
- {
- break;
- }
- src = p->bytes + p->offset;
- decode_ptr (&src, &pos);
- p->offset = src - (char*) p->bytes;
- pp->level++;
- }
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward_unode returning a hit "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return pos;
+ } /* found one */
+ pos=nxtpos;
+ p->offset=src-(char*)p->bytes;
+ (pp->skipped_nodes[pp->maxlevel - pp->level -1])++;
+#if ISAMB_DEBUG
+ skips++;
+#endif
}
- assert (p->offset < p->size);
- assert (p->bytes[0]);
- src = p->bytes + p->offset;
- (*pp->isamb->method->code_item)(ISAMC_DECODE, p->decodeClientData,
- &dst, &src);
- p->offset = src - (char*) p->bytes;
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward_unode returning at tail "
+ "at level %d node %d ofs=%d sz=%d skips=%d",
+ pp->level, p->pos, p->offset, p->size, skips);
+#endif
+ return pos; /* that's the last one in the line */
+
+} /* forward_unode */
+
+static void isamb_pp_descend_to_leaf(ISAMB_PP pp, ISAMB_P pos, const void *untilbuf)
+{ /* climbs down the tree, from pos, to the leftmost leaf */
+ struct ISAMB_block *p = pp->block[pp->level];
+ const char *src;
+ assert(!p->leaf);
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_descend_to_leaf "
+ "starting at lev %d node %d ofs=%d lf=%d u=%p",
+ pp->level, p->pos, p->offset, p->leaf, untilbuf);
+#endif
+ if (untilbuf)
+ pos=isamb_pp_forward_unode(pp,pos,untilbuf);
+ ++(pp->level);
+ assert(pos);
+ p=open_block(pp->isamb, pos);
+ pp->block[pp->level]=p;
+ ++(pp->accessed_nodes[pp->maxlevel-pp->level]);
+ ++(pp->no_blocks);
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_descend_to_leaf "
+ "got lev %d node %d lf=%d",
+ pp->level, p->pos, p->leaf);
+#endif
+ if (p->leaf)
+ return;
+ assert (p->offset==0 );
+ src=p->bytes + p->offset;
+ decode_ptr(&src, &pos);
+ p->offset=src-(char*)p->bytes;
+ isamb_pp_descend_to_leaf(pp,pos,untilbuf);
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_descend_to_leaf "
+ "returning at lev %d node %d ofs=%d lf=%d",
+ pp->level, p->pos, p->offset, p->leaf);
+#endif
+} /* descend_to_leaf */
+
+static int isamb_pp_find_next_leaf(ISAMB_PP pp)
+{ /* finds the next leaf by climbing up and down */
+ ISAMB_P pos;
+ if (!isamb_pp_climb_level(pp,&pos))
+ return 0;
+ isamb_pp_descend_to_leaf(pp, pos,0);
return 1;
}
-int isamb_pp_num (ISAMB_PP pp)
-{
+static int isamb_pp_climb_desc(ISAMB_PP pp, const void *untilbuf)
+{ /* climbs up and descends to a leaf where values >= *untilbuf are found */
+ ISAMB_P pos;
+#if ISAMB_DEBUG
+ struct ISAMB_block *p = pp->block[pp->level];
+ yaz_log(YLOG_DEBUG,"isamb_pp_climb_desc starting "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ if (!isamb_pp_climb_level(pp,&pos))
+ return 0;
+ /* see if it would pay to climb one higher */
+ if (!isamb_pp_on_right_node(pp, pp->level, untilbuf))
+ if (!isamb_pp_climb_level(pp,&pos))
+ return 0;
+ isamb_pp_descend_to_leaf(pp, pos,untilbuf);
+#if ISAMB_DEBUG
+ p = pp->block[pp->level];
+ yaz_log(YLOG_DEBUG,"isamb_pp_climb_desc done "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
return 1;
+} /* climb_desc */
+
+int isamb_pp_forward (ISAMB_PP pp, void *buf, const void *untilbuf)
+{
+#if ISAMB_DEBUG
+ struct ISAMB_block *p = pp->block[pp->level];
+ assert(p->leaf);
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward starting "
+ "at level %d node %d ofs=%d sz=%d u=%p sc=%d",
+ pp->level, p->pos, p->offset, p->size,untilbuf, scope);
+#endif
+ if (untilbuf) {
+ if (isamb_pp_forward_on_leaf( pp, buf, untilbuf)) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward (f) returning (A) "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return 1;
+ }
+ if (! isamb_pp_climb_desc( pp, untilbuf)) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward (f) returning notfound (B) "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return 0; /* could not find a leaf */
+ }
+ do{
+ if (isamb_pp_forward_on_leaf( pp, buf, untilbuf)) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward (f) returning (C) "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return 1;
+ }
+ }while ( isamb_pp_find_next_leaf(pp));
+ return 0; /* could not find at all */
+ }
+ else { /* no untilbuf, a straight read */
+ /* FIXME - this should be moved
+ * directly into the pp_read */
+ /* keeping here now, to keep same
+ * interface as the old fwd */
+ if (isamb_pp_read_on_leaf( pp, buf)) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward (read) returning (D) "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return 1;
+ }
+ if (isamb_pp_find_next_leaf(pp)) {
+#if ISAMB_DEBUG
+ yaz_log(YLOG_DEBUG,"isamb_pp_forward (read) returning (E) "
+ "at level %d node %d ofs=%d sz=%d",
+ pp->level, p->pos, p->offset, p->size);
+#endif
+ return isamb_pp_read_on_leaf(pp, buf);
+ }
+ else
+ return 0;
+ }
+} /* isam_pp_forward (new version) */
+
+void isamb_pp_pos( ISAMB_PP pp, double *current, double *total )
+{ /* return an estimate of the current position and of the total number of */
+ /* occureences in the isam tree, based on the current leaf */
+ struct ISAMB_block *p = pp->block[pp->level];
+ assert(total);
+ assert(current);
+ assert(p->leaf);
+
+ *total = pp->block[0]->no_items;
+ *current = (double) pp->returned_numbers;
+#if ISAMB_DEBUG
+ yaz_log(YLOG_LOG, "isamb_pp_pos returning: cur= %0.1f tot=%0.1f rn="
+ ZINT_FORMAT, *current, *total, pp->returned_numbers);
+#endif
}
projects / idzebra-moved-to-github.git / blobdiff