X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=isamc%2Fmerge.c;h=e8d92473df904c832b700c93320e073beccd9ce6;hb=5437b50633032595afe6f87dc0f989bc92a5aea8;hp=b2092c1664f4b67f4cb0790f18bb9d8276b5a911;hpb=1d81c32c736b0b108f1739b25cd46efec0cec34c;p=idzebra-moved-to-github.git diff --git a/isamc/merge.c b/isamc/merge.c index b2092c1..e8d9247 100644 --- a/isamc/merge.c +++ b/isamc/merge.c @@ -1,61 +1,76 @@ -/* - * Copyright (c) 1996-1998, Index Data. - * See the file LICENSE for details. - * Sebastian Hammer, Adam Dickmeiss - * - * $Log: merge.c,v $ - * Revision 1.6 1998-03-06 13:54:03 adam - * Fixed two nasty bugs in isc_merge. - * - * Revision 1.5 1997/02/12 20:42:43 adam - * Bug fix: during isc_merge operations, some pages weren't marked dirty - * even though they should be. At this point the merge operation marks - * a page dirty if the previous page changed at all. A better approach is - * to mark it dirty if the last key written changed in previous page. - * - * Revision 1.4 1996/11/08 11:15:31 adam - * Number of keys in chain are stored in first block and the function - * to retrieve this information, isc_pp_num is implemented. - * - * Revision 1.3 1996/11/04 14:08:59 adam - * Optimized free block usage. - * - * Revision 1.2 1996/11/01 13:36:46 adam - * New element, max_blocks_mem, that control how many blocks of max size - * to store in memory during isc_merge. - * Function isc_merge now ignores delete/update of identical keys and - * the proper blocks are then non-dirty and not written in flush_blocks. - * - * Revision 1.1 1996/11/01 08:59:15 adam - * First version of isc_merge that supports update/delete. - * - */ +/* $Id: merge.c,v 1.27 2004-11-19 10:27:09 heikki 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 #include #include #include - -#include +#include #include "isamc-p.h" struct isc_merge_block { int offset; /* offset in r_buf */ - int block; /* block number of file (0 if none) */ + zint block; /* block number of file (0 if none) */ int dirty; /* block is different from that on file */ }; +#if 0 +static void opt_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, + int last) +{ + int i, no_dirty = 0; + for (i = 0; i mb[j].block) + j_min = j; + } + assert (j_min >= 0); + tmp = mb[j_min].block; + mb[j_min].block = mb[i].block; + mb[i].block = tmp; + mb[i].dirty = 1; + } + if (!last) + mb[i].dirty = 1; +} +#endif + static void flush_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, - char *r_buf, int *firstpos, int cat, int last, - int *numkeys) + char *r_buf, zint *firstpos, int cat, int last, + zint *numkeys) { int i; for (i = 0; imethod->debug > 2) - logf (LOG_LOG, "isc: skip ptr=%d size=%d %d %d", + yaz_log (YLOG_LOG, "isc: skip ptr=%d size=%d %d " ZINT_FORMAT, i, ssize, cat, mb[i].block); ++(is->files[cat].no_skip_writes); continue; @@ -96,28 +116,27 @@ static void flush_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, src = r_buf + mb[i].offset - ISAMC_BLOCK_OFFSET_1; ssize += ISAMC_BLOCK_OFFSET_1; - memcpy (src+sizeof(int)+sizeof(ssize), numkeys, - sizeof(*numkeys)); + memcpy (src+sizeof(zint)+sizeof(ssize), numkeys, sizeof(*numkeys)); if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush ptr=%d numk=%d size=%d nextpos=%d", - i, *numkeys, (int) ssize, mb[i+1].block); + yaz_log (YLOG_LOG, "isc: flush ptr=%d numk=" ZINT_FORMAT " size=%d nextpos=" + ZINT_FORMAT, i, *numkeys, (int) ssize, mb[i+1].block); } else { src = r_buf + mb[i].offset - ISAMC_BLOCK_OFFSET_N; ssize += ISAMC_BLOCK_OFFSET_N; if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush ptr=%d size=%d nextpos=%d", + yaz_log (YLOG_LOG, "isc: flush ptr=%d size=%d nextpos=" ZINT_FORMAT, i, (int) ssize, mb[i+1].block); } - memcpy (src, &mb[i+1].block, sizeof(int)); - memcpy (src+sizeof(int), &ssize, sizeof(ssize)); + memcpy (src, &mb[i+1].block, sizeof(zint)); + memcpy (src+sizeof(zint), &ssize, sizeof(ssize)); isc_write_block (is, cat, mb[i].block, src); } } -static int get_border (ISAMC is, struct isc_merge_block *mb, int ptr, - int cat, int firstpos) +static int get_border (ISAMC is, struct isc_merge_block *mb, zint ptr, + int cat, zint firstpos) { /* Border set to initial fill or block size depending on whether we are creating a new one or updating and old one. @@ -132,7 +151,7 @@ static int get_border (ISAMC is, struct isc_merge_block *mb, int ptr, return mb[ptr].offset + fill - off; } -ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) +ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I *data) { char i_item[128], *i_item_ptr; @@ -141,20 +160,22 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) ISAMC_PP pp; char f_item[128], *f_item_ptr; int f_more; + int last_dirty = 0; + int debug = is->method->debug; struct isc_merge_block mb[200]; - int firstpos = 0; + zint firstpos = 0; int cat = 0; char r_item_buf[128]; /* temporary result output */ char *r_buf; /* block with resulting data */ int r_offset = 0; /* current offset in r_buf */ int ptr = 0; /* pointer */ void *r_clientData; /* encode client data */ - int border; - int numKeys = 0; + zint border; + zint numKeys = 0; - r_clientData = (*is->method->code_start)(ISAMC_ENCODE); + r_clientData = (*is->method->codec.start)(); r_buf = is->merge_buf + 128; pp = isc_pp_open (is, ipos); @@ -165,8 +186,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) f_more = 1; cat = pp->cat; - if (is->method->debug > 1) - logf (LOG_LOG, "isc: isc_merge begin %d %d", cat, pp->pos); + if (debug > 1) + yaz_log (YLOG_LOG, "isc: isc_merge begin %d " ZINT_FORMAT, cat, pp->pos); /* read first item from i */ i_item_ptr = i_item; @@ -190,18 +211,19 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) { /* the resulting output is of the same category as the the original - */ - if (mb[ptr].offset == r_offset) + */ + if (r_offset <= mb[ptr].offset +is->method->filecat[cat].mfill) { - /* the resulting output block is empty. Delete + /* the resulting output block is too small/empty. Delete the original (if any) - */ - if (is->method->debug > 3) - logf (LOG_LOG, "isc: release A"); + */ + if (debug > 3) + yaz_log (YLOG_LOG, "isc: release A"); if (mb[ptr].block) isc_release_block (is, pp->cat, mb[ptr].block); mb[ptr].block = pp->pos; - mb[ptr].dirty = 2; + if (!mb[ptr].dirty) + mb[ptr].dirty = 1; if (ptr > 0) mb[ptr-1].dirty = 1; } @@ -209,18 +231,18 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) { /* indicate new boundary based on the original file */ mb[++ptr].block = pp->pos; - mb[ptr].dirty = (mb[ptr-1].dirty > 1) ? 1 : 0; + mb[ptr].dirty = last_dirty; mb[ptr].offset = r_offset; - if (is->method->debug > 3) - logf (LOG_LOG, "isc: bound ptr=%d,offset=%d", + if (debug > 3) + yaz_log (YLOG_LOG, "isc: bound ptr=%d,offset=%d", ptr, r_offset); if (cat==is->max_cat && ptr >= is->method->max_blocks_mem) { /* We are dealing with block(s) of max size. Block(s) except 1 will be flushed. */ - if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush A %d sections", ptr); + if (debug > 2) + yaz_log (YLOG_LOG, "isc: flush A %d sections", ptr); flush_blocks (is, mb, ptr-1, r_buf, &firstpos, cat, 0, &pp->numKeys); mb[0].block = mb[ptr-1].block; @@ -239,6 +261,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) } border = get_border (is, mb, ptr, cat, firstpos); } + last_dirty = 0; if (!f_more) cmp = -1; else if (!i_more) @@ -256,7 +279,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) /* is next input item the same as current except for the delete flag? */ cmp = (*is->method->compare_item)(i_item, f_item); - if (!cmp && i_mode) + if (!cmp && i_mode) /* delete/insert nop? */ { /* yes! insert as if it was an insert only */ memcpy (r_item, i_item, i_item_ptr - i_item); @@ -268,6 +291,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) { /* no! delete the item */ r_item = NULL; + last_dirty = 1; mb[ptr].dirty = 2; } } @@ -295,12 +319,13 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) { if (!i_mode) /* delete item which isn't there? */ { - logf (LOG_FATAL, "Inconsistent register at offset %d", + yaz_log (YLOG_FATAL, "Inconsistent register at offset %d", r_offset); abort (); } memcpy (r_item, i_item, i_item_ptr - i_item); mb[ptr].dirty = 2; + last_dirty = 1; /* move i */ i_item_ptr = i_item; i_more = (*data->read_item)(data->clientData, &i_item_ptr, @@ -309,18 +334,19 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) if (r_item) /* insert resulting item? */ { char *r_out_ptr = r_buf + r_offset; + const char *src = r_item; int new_offset; - (*is->method->code_item)(ISAMC_ENCODE, r_clientData, - &r_out_ptr, &r_item); + (*is->method->codec.encode)(r_clientData, &r_out_ptr, &src); new_offset = r_out_ptr - r_buf; numKeys++; if (border < new_offset && border >= r_offset) { - if (is->method->debug > 2) - logf (LOG_LOG, "isc: border %d %d", ptr, border); + if (debug > 2) + yaz_log (YLOG_LOG, "isc: border %d " ZINT_FORMAT, + ptr, border); /* Max size of current block category reached ... make new virtual block entry */ mb[++ptr].block = 0; @@ -332,8 +358,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) except one will be flushed. Note: the block(s) are surely not the last one(s). */ - if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush B %d sections", ptr-1); + if (debug > 2) + yaz_log (YLOG_LOG, "isc: flush B %d sections", ptr-1); flush_blocks (is, mb, ptr-1, r_buf, &firstpos, cat, 0, &pp->numKeys); mb[0].block = mb[ptr-1].block; @@ -374,26 +400,30 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) cat++; mb[j].dirty = 1; mb[j].block = 0; + mb[ptr].offset = r_offset; for (i = 1; i < ptr; i++) { int border = is->method->filecat[cat].ifill - ISAMC_BLOCK_OFFSET_1 + mb[j].offset; - if (is->method->debug > 3) - logf (LOG_LOG, "isc: remap %d border=%d", i, border); + if (debug > 3) + yaz_log (YLOG_LOG, "isc: remap %d border=%d", i, border); if (mb[i+1].offset > border && mb[i].offset <= border) { - if (is->method->debug > 3) - logf (LOG_LOG, "isc: to %d %d", j, mb[i].offset); + if (debug > 3) + yaz_log (YLOG_LOG, "isc: to %d %d", j, mb[i].offset); mb[++j].dirty = 1; mb[j].block = 0; mb[j].offset = mb[i].offset; } } - if (is->method->debug > 2) - logf (LOG_LOG, "isc: remap from %d to %d sections to cat %d", + if (debug > 2) + yaz_log (YLOG_LOG, "isc: remap from %d to %d sections to cat %d", ptr, j, cat); ptr = j; border = get_border (is, mb, ptr, cat, firstpos); + if (debug > 3) + yaz_log (YLOG_LOG, "isc: border=" ZINT_FORMAT " r_offset=%d", + border, r_offset); } } if (mb[ptr].offset < r_offset) @@ -406,8 +436,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) { /* empty output. Release last block if any */ if (cat == pp->cat && mb[ptr].block) { - if (is->method->debug > 3) - logf (LOG_LOG, "isc: release C"); + if (debug > 3) + yaz_log (YLOG_LOG, "isc: release C"); isc_release_block (is, pp->cat, mb[ptr].block); mb[ptr].block = 0; if (ptr > 0) @@ -415,8 +445,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) } } - if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush C, %d sections", ptr); + if (debug > 2) + yaz_log (YLOG_LOG, "isc: flush C, %d sections", ptr); if (firstpos) { @@ -424,8 +454,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) has changed */ if (numKeys != isc_pp_num (pp)) { - if (is->method->debug > 2) - logf (LOG_LOG, "isc: patch num keys firstpos=%d num=%d", + if (debug > 2) + yaz_log (YLOG_LOG, "isc: patch num keys firstpos=" ZINT_FORMAT " num=" ZINT_FORMAT, firstpos, numKeys); bf_write (is->files[cat].bf, firstpos, ISAMC_BLOCK_OFFSET_N, sizeof(numKeys), &numKeys); @@ -440,11 +470,11 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) /* flush rest of block(s) in r_buf */ flush_blocks (is, mb, ptr, r_buf, &firstpos, cat, 1, &numKeys); - (*is->method->code_stop)(ISAMC_ENCODE, r_clientData); + (*is->method->codec.stop)(r_clientData); if (!firstpos) cat = 0; - if (is->method->debug > 1) - logf (LOG_LOG, "isc: isc_merge return %d %d", cat, firstpos); + if (debug > 1) + yaz_log (YLOG_LOG, "isc: isc_merge return %d " ZINT_FORMAT, cat, firstpos); isc_pp_close (pp); return cat + firstpos * 8; }