X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=isamc%2Fmerge.c;h=aa7be55f46c2a8cac7d768a28611b27f68569a17;hb=7afdc32625057ad7146fd043f5bcac1eeb3d53b6;hp=5970471cd8308968a3f4f52f48bfbc16160890bb;hpb=5ed7bb5a4a7e46271cc161304b114eae54b53674;p=idzebra-moved-to-github.git diff --git a/isamc/merge.c b/isamc/merge.c index 5970471..aa7be55 100644 --- a/isamc/merge.c +++ b/isamc/merge.c @@ -462,7 +462,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) return cat + firstpos * 8; } -char *hexdump(unsigned char *p, int len, char *buff) { +static char *hexdump(unsigned char *p, int len, char *buff) { static char localbuff[128]; char bytebuff[8]; if (!buff) buff=localbuff; @@ -476,544 +476,311 @@ char *hexdump(unsigned char *p, int len, char *buff) { return buff; } -ISAMC_P isamh_append (ISAMH is, ISAMH_P ipos, ISAMH_I data) + +/* isamh - heikki's append-only isam + * Idea: When allocating a new block, allocate memory for a very large block + * (maximal blocksize). When done, see if you can shrink it to some + * smaller size. First-time indexing will go in optimal blocks, and + * following small additions will go to the end of the last of the + * maximal ones. Only later, when new blocks need to be allocated, it + * may make sense to reserve some extra space... +*/ + + +static void isamh_reduceblock(ISAMH is, ISAMH_PP pp, int numKeys) { + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_reduce: block p=%d c=%d o=%d nk=%d ", + pp->pos, pp->cat, pp->offset, numKeys); + if (pp->pos != 0) + return; /* already allocated in some size */ + while ( ( pp->cat > 0 ) && + ( pp->offset < is->method->filecat[pp->cat-1].bsize) && + ( numKeys < is->method->filecat[pp->cat-1].mblocks) ) + pp->cat--; + pp->pos = isamh_alloc_block(is,pp->cat) ; + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_reduced block p=%d to c=%d o=%d nk=%d bs=%d", + pp->pos, pp->cat, pp->offset, numKeys, + is->method->filecat[pp->cat].bsize); +} /* reduceblock */ - ISAMH_PP pp; - char f_item[128]; - char *f_item_ptr=f_item; - int fmore=1; +ISAMC_P isamh_append (ISAMH is, ISAMH_P ipos, ISAMH_I data) +{ + + ISAMH_PP pp; /* always the last pp in the chain */ + ISAMH_PP firstpp; /* always the first one in the chain, may ==pp */ + ISAMH_PP prevpp; /* the one that points to pp, may be null or ==firstpp */ + /* needed to postpone the writing of its next field until */ + /* pp itself has been categorized */ char i_item[128]; char *i_item_ptr; - int i_more=1, i_mode, i; - - char *r_out_ptr; + int i_more=1, i_mode; char codebuffer[128]; char *codeptr; + char *bufptr; int codelen; - ISAMH_PP firstpp; void *r_clientData; /* encode client data */ int newblock; int newcat; - int numKeys = 0; + int maxkeys; int maxsize; int retval; + int maxcat; pp = firstpp = isamh_pp_open (is, ipos); + prevpp =0; /* only used when new blocks allocated */ + assert (*is->method->code_reset); + maxcat=0; /* find the largest block size for default allocation */ + for ( newcat=0; is->method->filecat[newcat].mblocks > 0; newcat++) + if ( is->method->filecat[newcat].bsize > + is->method->filecat[maxcat].bsize) + maxcat = newcat; + if ( 0==ipos) { /* new block */ - pp->cat=0; - pp->pos = isamh_alloc_block(is,pp->cat); + pp->cat=maxcat; /* start large... */ + pp->pos = 0; /* not allocated yet */ pp->size= pp->offset = ISAMH_BLOCK_OFFSET_1 ; - logf(LOG_LOG,"isamh_append: starting with new block"); + pp->buf=xrealloc(pp->buf,is->method->filecat[maxcat].bsize); + r_clientData = (*is->method->code_start)(ISAMH_ENCODE); + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append: starting with new block"); } else { /* existing block */ - if (firstpp->lastblock == firstpp->pos) + if (isamh_block(firstpp->lastblock) == firstpp->pos) + /*!!! TODO: BUG: Compare whole addresses !!! (later) */ { /* only one block, we have it already */ pp->offset=ISAMH_BLOCK_OFFSET_1; - logf(LOG_LOG,"isamh_append: starting with one block"); + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append: starting with one block %d",pp->pos); } else - { /* TODO: Read the last block (into what buffer?) */ + { + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append: starting with multiple blocks %d>%d>%d", + firstpp->pos,isamh_block(firstpp->next),isamh_block(firstpp->lastblock)); + pp=isamh_pp_open(is,firstpp->lastblock); + /* dirty, but this can also read a N-block. Just clear extra values*/ + pp->lastblock=0; pp->offset=ISAMH_BLOCK_OFFSET_N; - logf(LOG_LOG,"isamh_append: starting with multiple blocks"); } /* get last */ - /* read pointers in it to synchronize the encoder ??!! */ + r_clientData = (*is->method->code_start)(ISAMH_ENCODE); + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append: scanning to end of block %d %d->%d", + pp->pos, pp->offset, pp->size); codeptr=codebuffer; - //while () { - //} + while (pp->offsetsize) { + codeptr=codebuffer; + bufptr=pp->buf + pp->offset; + (*is->method->code_item)(ISAMH_DECODE, r_clientData, &codeptr, &bufptr); + codelen = bufptr - (pp->buf+pp->offset) ; + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append: dec at %d %d/%d:%s", + pp->offset, codelen, codeptr-codebuffer, + hexdump(codebuffer,codeptr-codebuffer,0) ); + pp->offset += codelen; + } } /* existing block */ - r_clientData = (*is->method->code_start)(ISAMH_ENCODE); i_item_ptr = i_item; i_more = (*data->read_item)(data->clientData,&i_item_ptr,&i_mode); - logf(LOG_LOG,"isamh_append 1: m=%d l=%d %s", - i_mode, i_item_ptr-i_item, hexdump(i_item,i_item_ptr-i_item,0)); + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append 1: m=%d l=%d %s", + i_mode, i_item_ptr-i_item, hexdump(i_item,i_item_ptr-i_item,0)); maxsize = is->method->filecat[pp->cat].bsize; while(i_more) { - codeptr = codebuffer; - i_item_ptr=i_item; - (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); - codelen = codeptr-codebuffer; + if (i_mode) + { /* insert key, ignore all delete keys time being... */ + codeptr = codebuffer; + i_item_ptr=i_item; + (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); + codelen = codeptr-codebuffer; - assert( (codelen < 128) && (codelen>0)); + assert( (codelen < 128) && (codelen>0)); - logf(LOG_LOG,"isamh_append: coded into %d:%s", - codelen,hexdump(codebuffer,codelen,0)); - - - if ( pp->offset + codelen > maxsize ) - { - logf(LOG_LOG,"isamh_append: need new block: %d > %d ", - pp->offset + codelen, maxsize ); - newcat = pp->cat; /* TODO - grow that block some day... */ - newblock = isamh_alloc_block(is,newcat); - pp->next = newblock; - if (firstpp!=pp) - { /* not first block, write to disk already now */ - isamh_buildlaterblock(pp); - isamh_write_block(is,pp->cat,pp->pos,pp->buf); - //if (cat != newcat) - // realloc buf !!!! - } - else - { /* we had only one block, allocate a second buffer */ - pp = (ISAMH_PP) xmalloc (sizeof(*pp)); - assert(pp); - *pp = *firstpp; /* copy most fields directly over */ - pp->buf = (char *) xmalloc (is->method->filecat[newcat].bsize); - } - pp->cat = newcat; - pp->pos = newblock; - pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; - pp->next=0; - logf(LOG_LOG,"isamh_append: got a new block %d",pp->pos); - - /* reset the encoding, and code again */ - (*is->method->code_reset)(r_clientData); - codeptr = codebuffer; - i_item_ptr=i_item; - (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); - codelen = codeptr-codebuffer; - logf(LOG_LOG,"isamh_append: coded again %d:%s", - codelen,hexdump(codebuffer,codelen,0)); - - } /* new block needed */ - - /* ok, now we can write it */ - memcpy(&(pp->buf[pp->offset]), codebuffer, codelen); - pp->offset += codelen; - pp->size += codelen; - firstpp->numKeys++; + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append: coded into %d:%s (nk=%d)", + codelen,hexdump(codebuffer,codelen,0),firstpp->numKeys); + + if ( pp->offset + codelen > maxsize ) + { /* oops, block full, do something */ + newcat = maxcat; /* start with a large block again */ + maxkeys = is->method->filecat[pp->cat].mblocks; /* max keys */ + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append: need new block: %d > %d (k:%d/%d)", + pp->offset + codelen, maxsize, firstpp->numKeys,maxkeys ); + + newblock = 0; + pp->next = 0; + + /* Four possibilities: */ + if (prevpp!=0) + { /* 1: we have a prevpp that can go on the disk */ + /* reduce pp, set next ptr, and store block */ + /* set pp up as new last block, and remember the just */ + /* filled as (pp) the new prev */ + assert(pp!=firstpp); + isamh_reduceblock(is,pp,firstpp->numKeys); + prevpp->next=isamh_addr(pp->pos,pp->cat); + isamh_buildlaterblock(prevpp); + isamh_write_block(is,prevpp->cat,prevpp->pos,prevpp->buf); + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append N1: Wrote prevpp (%d:%d) -> %d:%d", + prevpp->pos, prevpp->cat, pp->pos, pp->cat); + isamh_pp_close(prevpp); /* it's done its job */ + prevpp = pp; + pp=isamh_pp_open(is,isamh_addr(0,maxcat)); /* start a large one */ + pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; + pp->next=0; + pp->lastblock=0; + } + else if ( (firstpp!=pp) && (firstpp->next != 0)) + { /* 2: we are working at end of list, but have no prevpp */ + /* some block (already on disk) points to pp */ + /* set the newly filled block as prev, don't save yet */ + /* allocate new pp */ + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append N2: set up a new prevpp (%d:%d)", + pp->pos, pp->cat); + prevpp = pp; + pp=isamh_pp_open(is,isamh_addr(0,maxcat)); /* start a large one */ + pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; + pp->next=0; + pp->lastblock=0; + } + else if ( (firstpp!=pp) && (firstpp->next==0)) + { /* 3: we have earlier allocated pp as the second block */ + /* reduce it to get its address. Set first->next to it */ + /* move it to prevpp, and create a new pp */ + isamh_reduceblock(is,pp,firstpp->numKeys); + firstpp->next=isamh_addr(pp->pos,pp->cat); + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append N3: set up a new firstpp->next (%d:%d)", + pp->pos, pp->cat); + prevpp=pp; + pp=isamh_pp_open(is,isamh_addr(0,maxcat)); /* start a large one */ + pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; + pp->next=0; + pp->lastblock=0; + } + else + { /* 4: firstpp itself has got full. */ + /* allocate a new pp for it, store nothing */ + assert(firstpp==pp); + assert(firstpp->next==0); + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append N4: allocated new pp for full first (%d:%d)", + firstpp->pos, firstpp->cat); + pp=isamh_pp_open(is,isamh_addr(0,maxcat)); /* start a large one */ + pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; + pp->next=0; + pp->lastblock=0; + } - /* and try to read the next element */ + maxsize = is->method->filecat[pp->cat].bsize; + pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; + pp->next=0; + pp->lastblock=0; + if (pp->is->method->debug > 2) + logf(LOG_LOG,"isamh_append: got a new block c=%d p=%d",pp->cat,pp->pos); + + + /* reset the encoding, and code again */ + (*is->method->code_reset)(r_clientData); + codeptr = codebuffer; + i_item_ptr=i_item; + (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); + codelen = codeptr-codebuffer; + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append: coded again %d:%s (nk=%d)", + codelen,hexdump(codebuffer,codelen,0),firstpp->numKeys); + + } /* block full */ + + /* ok, now we can write it */ + memcpy(&(pp->buf[pp->offset]), codebuffer, codelen); + pp->offset += codelen; + pp->size += codelen; + firstpp->numKeys++; + } /* not a delete */ + + /* try to read the next element */ i_item_ptr = i_item; i_more = (*data->read_item)(data->clientData,&i_item_ptr,&i_mode); - logf(LOG_LOG,"isamh_append 2: m=%d l=%d %s", + if (pp->is->method->debug > 3) + logf(LOG_LOG,"isamh_append 2: m=%d l=%d %s", i_mode, i_item_ptr-i_item, hexdump(i_item,i_item_ptr-i_item,0)); - + + } /* while */ + + isamh_reduceblock(is,pp,firstpp->numKeys); + /* may be the same as firstpp! */ + + if (prevpp) + { /* Write the prev block we have been holding off */ + assert(prevpp->pos); + prevpp->next = isamh_addr(pp->pos, pp->cat); + isamh_buildlaterblock(prevpp); + isamh_write_block(is,prevpp->cat,prevpp->pos,prevpp->buf); + if (firstpp->next==0) /* can happen if extending many blocks */ + firstpp->next=isamh_addr(prevpp->pos,prevpp->cat); + isamh_pp_close(prevpp); } - /* Write the last (partial) block, if needed. */ if (pp!=firstpp) { + pp->next=0; /* just to be sure */ + if (firstpp->next==0) + firstpp->next=isamh_addr(pp->pos,pp->cat); isamh_buildlaterblock(pp); isamh_write_block(is,pp->cat,pp->pos,pp->buf); } /* update first block and write it */ - firstpp->lastblock = pp->pos; + firstpp->lastblock = isamh_addr(pp->pos,pp->cat); + isamh_reduceblock(is,firstpp,firstpp->numKeys); isamh_buildfirstblock(firstpp); isamh_write_block(is,firstpp->cat,firstpp->pos,firstpp->buf); /* release the second block, if we allocated one */ if ( firstpp != pp ) - { - xfree(pp->buf); - xfree(pp); - } - - retval = firstpp->pos*8 + firstpp->cat; + isamh_pp_close(pp); + + /* get return value (before it disappears at close! */ + retval = isamh_addr(firstpp->pos,firstpp->cat); - isamh_pp_close(firstpp); + isamh_pp_close(firstpp); return retval; } /* isamh_append */ -ISAMC_P test_isamh_append (ISAMH is, ISAMH_P ipos, ISAMH_I data) -/* test routines while fighting it */ -{ - /* ipos is always ==0, in my test, as I have no earlier base to insert */ - /* into. The key extractor calls this only once for each key to be inserted */ - - - ISAMH_PP pp; - char f_item[128]; - char *f_item_ptr=f_item; - int fmore=1; - - char i_item[128]; - char *i_item_ptr; - int i_more=1, i_mode, i; - - pp = isamh_pp_open (is, ipos); - logf (LOG_LOG, "isamh_append:scannig fmore loop (ipos=%d)",ipos); - while (fmore) - { - f_item_ptr=f_item; - fmore = isamh_read_item (pp,&f_item_ptr); - logf (LOG_LOG, "isamh_append: fmore=%d len=%d", - fmore, f_item_ptr-f_item); - } /* while fmore */ - - logf (LOG_LOG, "isamh_append:scannig imore loop"); - - while(i_more) { - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData,&i_item_ptr,&i_mode); - logf(LOG_LOG,"isamh_append: mode=%d len=%d",i_mode, i_item_ptr-i_item); - } - - isamh_pp_close(pp); -} /* foo isamh_append */ - -#ifdef SKIPOLDISAM - -ISAMC_P old_isamh_append (ISAMH is, ISAMH_P ipos, ISAMH_I data) -{ - - char i_item[128], *i_item_ptr; - int i_more, i_mode, i; - - ISAMH_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; - 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; - - r_clientData = (*is->method->code_start)(ISAMH_ENCODE); - r_buf = is->merge_buf + 128; - - pp = isamh_pp_open (is, ipos); - /* read first item from file. make sure f_more indicates no boundary */ - f_item_ptr = f_item; - f_more = isamh_read_item (pp, &f_item_ptr); - if (f_more > 0) - f_more = 1; - cat = pp->cat; - - if (debug > 1) - logf (LOG_LOG, "isc: isamh_append begin %d %d %d", cat, pp->pos, ipos); - - /* read first item from i */ - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData, &i_item_ptr, &i_mode); - - mb[ptr].block = pp->pos; /* is zero if no block on disk */ - mb[ptr].dirty = 0; - mb[ptr].offset = 0; - - border = is->method->filecat[cat].bsize; - /* border = get_border (is, mb, ptr, cat, firstpos); */ - while (i_more || f_more) - { - char *r_item = r_item_buf; - int cmp; - - if (f_more > 1) - { - /* block to block boundary in the original file. */ - f_more = 1; - if (cat == pp->cat) - { - /* the resulting output is of the same category as the - the original - */ - -#ifdef SKIPTHIS /* should not happen when just appending new records */ - if (r_offset <= mb[ptr].offset +is->method->filecat[cat].mfill) -#else - if (0) -#endif - { - /* the resulting output block is too small/empty. Delete - the original (if any) - */ - if (debug > 3) - logf (LOG_LOG, "isc: release A"); - if (mb[ptr].block) - isamh_release_block (is, pp->cat, mb[ptr].block); - mb[ptr].block = pp->pos; - if (!mb[ptr].dirty) - mb[ptr].dirty = 1; - if (ptr > 0) - mb[ptr-1].dirty = 1; - } - else - { - - /* indicate new boundary based on the original file */ - mb[++ptr].block = pp->pos; - mb[ptr].dirty = last_dirty; - mb[ptr].offset = r_offset; - if (debug > 3) - logf (LOG_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 (debug > 2) - logf (LOG_LOG, "isc: flush A %d sections", ptr); - flush_blocks ((ISAMC)is, mb, ptr-1, r_buf, &firstpos, cat, - 0, &pp->numKeys); - mb[0].block = mb[ptr-1].block; - mb[0].dirty = mb[ptr-1].dirty; - memcpy (r_buf, r_buf + mb[ptr-1].offset, - mb[ptr].offset - mb[ptr-1].offset); - mb[0].offset = 0; - - mb[1].block = mb[ptr].block; - mb[1].dirty = mb[ptr].dirty; - mb[1].offset = mb[ptr].offset - mb[ptr-1].offset; - ptr = 1; - r_offset = mb[ptr].offset; - } - } - } - /*border = get_border (is, mb, ptr, cat, firstpos);*/ - border = is->method->filecat[cat].bsize; - } - last_dirty = 0; - if (!f_more) - cmp = -1; - else if (!i_more) - cmp = 1; - else - cmp = (*is->method->compare_item)(i_item, f_item); - if (cmp == 0) /* insert i=f */ - { - if (!i_mode) /* delete item? */ - { - /* move i */ - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData, &i_item_ptr, - &i_mode); - /* 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) /* delete/insert nop? */ - { - /* yes! insert as if it was an insert only */ - memcpy (r_item, i_item, i_item_ptr - i_item); - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData, &i_item_ptr, - &i_mode); - } - else - { - /* no! delete the item */ - r_item = NULL; - last_dirty = 1; - mb[ptr].dirty = 2; - } - } - else - { - memcpy (r_item, f_item, f_item_ptr - f_item); - - /* move i */ - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData, &i_item_ptr, - &i_mode); - } - /* move f */ - f_item_ptr = f_item; - f_more = isamh_read_item (pp, &f_item_ptr); - } - else if (cmp > 0) /* insert f */ - { - memcpy (r_item, f_item, f_item_ptr - f_item); - /* move f */ - f_item_ptr = f_item; - f_more = isamh_read_item (pp, &f_item_ptr); - } - else /* insert i */ - { - if (!i_mode) /* delete item which isn't there? */ - { - logf (LOG_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, - &i_mode); - } - if (r_item) /* insert resulting item? */ - { - char *r_out_ptr = r_buf + r_offset; - int new_offset; - - (*is->method->code_item)(ISAMH_ENCODE, r_clientData, - &r_out_ptr, &r_item); - new_offset = r_out_ptr - r_buf; - - numKeys++; - - if (border < new_offset && border >= r_offset) - { - if (debug > 2) - logf (LOG_LOG, "isc: border %d %d", ptr, border); - /* Max size of current block category reached ... - make new virtual block entry */ - mb[++ptr].block = 0; - mb[ptr].dirty = 1; - mb[ptr].offset = 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 one will be flushed. Note: the block(s) are - surely not the last one(s). - */ - if (debug > 2) - logf (LOG_LOG, "isc: flush B %d sections", ptr-1); - flush_blocks ((ISAMC)is, mb, ptr-1, r_buf, &firstpos, cat, - 0, &pp->numKeys); - mb[0].block = mb[ptr-1].block; - mb[0].dirty = mb[ptr-1].dirty; - memcpy (r_buf, r_buf + mb[ptr-1].offset, - mb[ptr].offset - mb[ptr-1].offset); - mb[0].offset = 0; - - mb[1].block = mb[ptr].block; - mb[1].dirty = mb[0].dirty; - mb[1].offset = mb[ptr].offset - mb[ptr-1].offset; - memcpy (r_buf + mb[1].offset, r_buf + r_offset, - new_offset - r_offset); - new_offset = (new_offset - r_offset) + mb[1].offset; - ptr = 1; - } - border = is->method->filecat[cat].bsize; - /* get_border (is, mb, ptr, cat, firstpos); */ - } - r_offset = new_offset; - } -#ifdef SKIPTHIS /* categories are handled differently in isamH */ - /* to be implemented later... */ - - if (cat < is->max_cat && ptr >= is->method->filecat[cat].mblocks) - { - /* Max number blocks in current category reached -> - must switch to next category (with larger block size) - */ - int j = 0; - - (is->files[cat].no_remap)++; - /* delete all original block(s) read so far */ - for (i = 0; i < ptr; i++) - if (mb[i].block) - isamh_release_block (is, pp->cat, mb[i].block); - /* also delete all block to be read in the future */ - pp->deleteFlag = 1; - - /* remap block offsets */ - assert (mb[j].offset == 0); - 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 - - ISAMH_BLOCK_OFFSET_1 + mb[j].offset; - if (debug > 3) - logf (LOG_LOG, "isc: remap %d border=%d", i, border); - if (mb[i+1].offset > border && mb[i].offset <= border) - { - if (debug > 3) - logf (LOG_LOG, "isc: to %d %d", j, mb[i].offset); - mb[++j].dirty = 1; - mb[j].block = 0; - mb[j].offset = mb[i].offset; - } - } - if (debug > 2) - logf (LOG_LOG, "isc: remap from %d to %d sections to cat %d", - ptr, j, cat); - ptr = j; - border = is->method->filecat[cat].bsize; - /*border = get_border (is, mb, ptr, cat, firstpos);*/ - if (debug > 3) - logf (LOG_LOG, "isc: border=%d r_offset=%d", border, r_offset); - } -#endif /* skipthis */ - - } - if (mb[ptr].offset < r_offset) - { /* make the final boundary offset */ - mb[++ptr].dirty = 1; - mb[ptr].block = 0; - mb[ptr].offset = r_offset; - } - else - { /* empty output. Release last block if any */ - if (cat == pp->cat && mb[ptr].block) - { - if (debug > 3) - logf (LOG_LOG, "isc: release C"); - isamh_release_block (is, pp->cat, mb[ptr].block); - mb[ptr].block = 0; - if (ptr > 0) - mb[ptr-1].dirty = 1; - } - } - - if (debug > 2) - logf (LOG_LOG, "isc: flush C, %d sections", ptr); - - if (firstpos) - { - /* we have to patch initial block with num keys if that - has changed */ - if (numKeys != isamh_pp_num (pp)) - { - if (debug > 2) - logf (LOG_LOG, "isc: patch num keys firstpos=%d num=%d", - firstpos, numKeys); - bf_write (is->files[cat].bf, firstpos, ISAMH_BLOCK_OFFSET_N, - sizeof(numKeys), &numKeys); - } - } - else if (ptr > 0) - { /* we haven't flushed initial block yet and there surely are some - blocks to flush. Make first block dirty if numKeys differ */ - if (numKeys != isamh_pp_num (pp)) - mb[0].dirty = 1; - } - /* flush rest of block(s) in r_buf */ - flush_blocks ((ISAMC)is, mb, ptr, r_buf, &firstpos, cat, 1, &numKeys); - - (*is->method->code_stop)(ISAMH_ENCODE, r_clientData); - if (!firstpos) - cat = 0; - if (debug > 1) - logf (LOG_LOG, "isc: isamh_append return %d %d", cat, firstpos); - isamh_pp_close (pp); - return cat + firstpos * 8; -} -#endif /* SKIPOLDISAM */ /* * $Log: merge.c,v $ - * Revision 1.13 1999-07-06 09:37:05 heikki + * Revision 1.19 1999-07-14 12:12:07 heikki + * Large-block isam-h (may not work too well... Abandoning for isam-d) + * + * Revision 1.17 1999/07/13 14:22:17 heikki + * Better allocation strategy in isamh_merge + * + * Revision 1.16 1999/07/08 14:23:27 heikki + * Fixed a bug in isamh_pp_read and cleaned up a bit + * + * Revision 1.15 1999/07/07 09:36:04 heikki + * Fixed an assertion in isamh + * + * Revision 1.13 1999/07/06 09:37:05 heikki * Working on isamh - not ready yet. * * Revision 1.12 1999/06/30 15:03:55 heikki