* See the file LICENSE for details.
* Heikki Levanto
*
- * $Id: merge-d.c,v 1.13 1999-08-18 13:59:19 heikki Exp $
+ * $Id: merge-d.c,v 1.16 1999-08-24 10:12:02 heikki Exp $
*
- * todo
- * - Clean up log levels
+ * missing
+ *
+ * optimize
* - Input filter: Eliminate del-ins pairs, tell if only one entry (or none)
- * - single-entry optimizing
+ * - single-entry optimizing (keep the one entry in the dict, no block)
* - study and optimize block sizes (later)
* - Clean up the different ways diffs are handled in writing and reading
* - Keep a merge-count in the firstpp, and if the block has already been
* merged, reduce it to a larger size even if it could fit in a small one!
* - Keep minimum freespace in the category table, and use that in reduce!
+ * - pass a space-needed for separateDiffBlock and reduce to be able to
+ * reserve more room for diffs, or to force a separate (larger?) block
+ * - Idea: Simplify the structure, so that the first block is always diffs.
+ * On small blocks, that is all we have. Once a block has been merged, we
+ * allocate the first main block and a (new) firstblock ffor diffs. From
+ * that point on the word has two blocks for it.
+ * - On allocating more blocks (in append), check the order of blocks, and
+ * if needed, swap them.
+ * - In merge, merge also with the input data.
*
* bugs
- * - Still has not been able to run a complete long test on bagel!
*
* caveat
* There is a confusion about the block addresses. cat or type is the category,
* Needs cleaning! The way diff blocks are handled in append and reading is
* quite different, and likely to give maintenance problems.
*
- * log levels (set isamd=x in zebra.cfg (or what ever cfg file you use) )
+ * log levels (set isamddebug=x in zebra.cfg (or what ever cfg file you use) )
* 0 = no logging. Default
* 1 = no logging here. isamd logs overall statistics
* 2 = Each call to isamd_append with start address and no more
* 7 = Log each record as it passes the system (once)
* 8 = Log raw and (de)coded data
* 9 = Anything else that may be useful
- * .. = Anything needed toi hunt a specific bug
+ * .. = Anything needed to hunt a specific bug
* (note that all tests in the code are like debug>3, which means 4 or above!)
*/
char hexbuff[64];
+ ++(is->files[0].no_fbuilds);
+
firstpp=pp=isamd_pp_open(is, isamd_addr(0,is->max_cat));
firstpp->size = firstpp->offset = ISAMD_BLOCK_OFFSET_1;
ISAMD_PP firstpp; /* the new first, the one we write into */
ISAMD_PP pp;
void *encoder_data;
+
+ ++(readpp->is->files[0].no_merges);
/* set up diffs as they should be for reading */
readpp->offset= ISAMD_BLOCK_OFFSET_1;
} /* while read */
- /* TODO: while pkey is an insert, and after last key inserted, append it */
- /* will prevent multiple merges on large insert runs */
-
/* set things up so that append can continue */
isamd_reduceblock(firstpp);
firstpp->diffs=0;
char *c_ptr = codebuff;
int codelen;
int merge_rc;
+ int mergecount=0;
+
+ ++(is->files[0].no_appds);
firstpp=isamd_pp_open(is, ipos);
if (is->method->debug >2)
logf(LOG_LOG,"isamd_appd: block pp=%p buf=%p [%d]:%s",
pp, pp->buf,
difflenidx, hexdump(&pp->buf[difflenidx],8,0));
+ if (mergecount++)
+ ++(is->files[0].no_remerges);
merge_rc = merge (&firstpp, &pp, &i_key);
if (0!=merge_rc)
return merge_rc; /* merge handled them all ! */
else
retval = append_diffs(is,ipos,data);
- if (0) /*!*/
- {
- void *p1=xmalloc(100);
- void *p2=xmalloc(100);
- void *p3=xmalloc(100);
- logf(LOG_LOG,"Traversing xmalloc stuff. p1=%p p2=%p p3=%p",p1,p2,p3);
- xmalloc_trav("end of append"); /*!*/
- assert(!"foo");
- }
-
return retval;
} /* isamd_append */
/*
* $Log: merge-d.c,v $
- * Revision 1.13 1999-08-18 13:59:19 heikki
+ * Revision 1.16 1999-08-24 10:12:02 heikki
+ * Comments about optimising
+ *
+ * Revision 1.15 1999/08/22 08:26:34 heikki
+ * COmments
+ *
+ * Revision 1.14 1999/08/20 12:25:58 heikki
+ * Statistics in isamd
+ *
+ * Revision 1.13 1999/08/18 13:59:19 heikki
* Fixed another unlikely difflen bug
*
* Revision 1.12 1999/08/18 13:28:17 heikki