/* addfile.h -- Copyright 1989, 1994, 1996 Liam R. Quin.
 * All Rights Reserved.
 * This code is NOT in the public domain.
 * See the file COPYRIGHT for full details.
 */

/* $Id: addfile.h,v 1.7 1996/08/14 16:50:11 lee Exp $
 *
 * This file defines the parameter to DumpCache.
 *
 */

#define MAXPROBES	15
    /* If we don't find an empty cache slot after this many tries,
     * we decide the hash table is full and clear it out, then try again.
     */

#define DUMP_CACHE	00	/* normal usage */
#define DUMP_SYNC	01	/* write all cached entries to disk */
#define DUMP_NOFREE	02	/* don't bother calling free() */
#define DUMP_FAST	04	/* only do a fast dump */

/* when the word-cache fills up, we first try and dump all of
 * the words that occur DUMP_FAST_THRESH or fewer times.  This generally
 * clears up a lot of slots in the hash table.  If that hasn't cleared
 * enough memory, we then dump the single most frequent word, before
 * increasing the dump threshhold and trying again.
 * Now that updating the index is possible, I am not sure that this is
 * the best way to continue, but DUMP_FAST_THRESH is certainly an effective
 * way to clear slots in the cache.
 * MUST_DUMP is used to start the dump process.
 */

#ifndef DUMP_FAST_THRESH
# define DUMP_FAST_THRESH	5
#endif

/* MUST_FREE_SLOTS is true when we need more free slots in the symbol table
 * used to cache words being indexed.
 *
 * We require 80% of the slots to be free, so that the hash function doen't
 * spend ages finding a free slot.
 */
#ifndef MUST_FREE_SLOTS
# define MUST_FREE_SLOTS(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	(SlotsUsed * 10 > TotalSlots * 8)
#endif

/* MUST_FREE_MORE_SLOTS is true when we have already started to dump the
 * cache to disk and we are checking to see when to stop.  This value is set
 * so that we don't immediately dump the cache again -- it must take a while
 * to fill up, no?
 * We require at least 80% of the slots to be empty, so that finding a slot
 * for a new word doesn't take too long.
 *
 * Usually there are plenty of slots to go round, but if you index something
 * with a huge vocabulary (/usr/dict/words, for example!) where there are
 * lots of words each occurring infrequently, you may run low on slots, and
 * with that sort of input, you really do need lots of slots free.
 *
 * So dumping all but 20% is only a performance hit when there are lots of
 * infrequent words, and that's when we need to do it most anyway.
 */
#ifndef MUST_FREE_MORE_SLOTS
# define MUST_FREE_MORE_SLOTS(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	(SlotsUsed * 5 > TotalSlots)
#endif

/* MUST_DUMP evaluates to non-zero when lqaddfile must dump its cache;;
 * MUST_DUMP_MORE is used to see if it's OK to stop dumping.
 */
#ifndef MUST_DUMP
# define MUST_DUMP(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	( (InCacheCount + 1 >= MaxInCache) || \
	    MUST_FREE_SLOTS(InCacheCount,MaxInCache,SlotsUsed,TotalSlots))
#endif

/* MUST_DUMP_MORE returns 0 when the cache is suffiently dumped to
 * enable more indexing to proceed.
 */
#ifndef MUST_DUMP_MORE
# define MUST_DUMP_MORE(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	( (InCacheCount * 100 >= MaxInCache * 60) /* allow 60% full */ || \
	    MUST_FREE_MORE_SLOTS(InCacheCount,MaxInCache,SlotsUsed,TotalSlots))
#endif

/* FILE_DUMP returns 1 if we should dump the cache between files:
 */
#ifndef MUST_FILE_DUMP
# define MUST_FILE_DUMP(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	( (InCacheCount * 100 >= MaxInCache * 98)  || \
	  (SlotsUsed * 100 > TotalSlots * 75)) /* allow 75% full */
#endif

/* FILE_DUMP_MORE returns 1 if we haven't finished dumping between files:
 */
#ifndef MUST_FILE_DUMP_MORE
# define MUST_FILE_DUMP_MORE(InCacheCount,MaxInCache,SlotsUsed,TotalSlots) \
	( (InCacheCount * 100 >= MaxInCache * 90) /* allow 90% full */ || \
	    MUST_FREE_MORE_SLOTS(InCacheCount,MaxInCache,SlotsUsed,TotalSlots))
#endif