home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Monster Media 1994 #1
/
monster.zip
/
monster
/
OS2
/
DB02_SRC.ZIP
/
BUCKET.CC
< prev
next >
Wrap
C/C++ Source or Header
|
1993-11-05
|
11KB
|
443 lines
/**************************************************************************
* Source Id :
*
* $Id: bucket.cc,v 1.31 1993/10/05 07:29:18 kevinl Exp $
*-------------------------------------------------------------------------
* Project Notes :
*
* Diamond Base
* ============
* A solid database implementation, spurred on by the continuing
* Metal (Lead) Base saga.
*
* Project Team :
* A. Davison
* K. Lentin
* D. Platt
*
* Project Commenced : 05-02-1993
*
*-------------------------------------------------------------------------
* Module Notes :
*
* Bucket Module
*
* Original Author : Andy & Kev (with inspiration from Daz)
*
*-------------------------------------------------------------------------
* Revision History:
*
* $Log: bucket.cc,v $
// Revision 1.31 1993/10/05 07:29:18 kevinl
// Fixed negative offset problem in bucket::del
//
// Revision 1.30 1993/07/20 13:05:14 kevinl
// No longer complains about unused 'in' in sploosh.
//
// Revision 1.29 1993/07/20 12:59:30 kevinl
// Some unisgned int vs long problems
//
// Revision 1.28 1993/06/23 05:21:22 kevinl
// Mallocs are now in angular brackets
//
// Revision 1.27 1993/06/20 13:43:05 kevinl
// Fixed multiple mallocs
//
// Revision 1.26 1993/05/26 01:01:39 kevinl
// MALLOC_H_MISSING
//
// Revision 1.25 1993/05/11 14:44:50 kevinl
// Added version number output
//
// Revision 1.24 1993/05/06 04:00:19 kevinl
// SASC define for malloc.h
//
// Revision 1.23 1993/05/03 23:33:37 kevinl
// long aligned data
//
// Revision 1.22 1993/05/03 01:33:03 kevinl
// Cosmetic (mainly) changes for CC
//
// Revision 1.21 1993/05/01 14:29:12 kevinl
// Replaced linear searchForKey with a binary search
// Got rid of ints
//
// Revision 1.20 1993/04/30 07:45:45 kevinl
// Some optimizing
//
// Revision 1.19 1993/04/28 11:30:15 kevinl
// Fixed up inlines
//
// Revision 1.18 1993/04/25 09:52:39 kevinl
// Fixed some dberror calls
//
// Revision 1.17 1993/04/24 07:15:27 kevinl
// Comments!
//
// Revision 1.16 1993/04/19 15:49:37 kevinl
// empty() now handles internal vs leaf nodes correctly
//
// Revision 1.15 1993/04/15 07:57:57 kevinl
// Added findIndex, del and empty
// Shortened full()
//
// Revision 1.14 1993/04/08 10:57:27 kevinl
// played with bcopy prototype to placate linux.
//
// Revision 1.13 1993/04/04 23:56:25 kevinl
// Rounded off keyLengths
// searchForKey returns orrect indicies
//
// Revision 1.12 1993/03/29 08:20:09 darrenp
// Added new malloc library support
//
// Revision 1.11 1993/03/18 08:33:58 kevinl
// Fixed up reinitiaised default argument
//
// Revision 1.10 1993/03/17 14:27:42 kevinl
// Added index number to bucket constructor
// Added keyNum entry to all object key accesses
//
// Revision 1.9 1993/03/15 13:58:35 kevinl
// Added setParent
// changed some int's to long
//
// Revision 1.8 1993/03/08 00:30:50 kevinl
// Replaced BORLANDC with __BORLANDC__
//
// Revision 1.7 1993/02/17 12:15:53 kevinl
// Added getLink and setLink to solve alignment problems
// some tidying
//
// Revision 1.6 1993/02/13 00:55:33 kevinl
// replaced some long's with bucketId
//
// Revision 1.5 1993/02/11 03:53:38 kevinl
// a few fixes.
//
// Revision 1.4 1993/02/10 07:56:34 darrenp
// commented now.
//
// Revision 1.3 1993/02/08 14:33:24 davison
// Andy: wrote sploosh
// Kev: minor fixes to sploosh, comments, some output routines for debugging
//
// Revision 1.2 1993/02/08 00:12:28 kevinl
// Kevin adds a constructor...
//
// Revision 1.1 1993/02/06 16:50:36 davison
// Initial revision
//
**************************************************************************/
#if !defined(MALLOC_H_MISSING) && !defined(MALLOC_H_INCLUDED)
extern "C" {
#include <malloc.h>
}
#define MALLOC_H_INCLUDED
#endif
#include <defs.h>
#include <object.h>
#include <iostream.h>
// Borland needs mem.h for memmove.
#ifdef __BORLANDC__
#include <mem.h>
#endif
// Ultrix hides it's bcopy prototype in here
#ifdef __ultrix
#include <memory.h>
#endif
char* bucket::verStr(void)
{
return "$Id: bucket.cc,v 1.31 1993/10/05 07:29:18 kevinl Exp $";
}
// The constructor.
bucket::bucket(bucketId newId, long newKeyLength, long newSize, e_leaf newLeaf, bucketId newParentId, bucketId newNextBucket, bucketId newPrevBucket, long newKeyNum)
{
// Allocate the data for the bucket and grab a pointer to the header data.
// Size is rounded up to the size of a long for allocation purposes
// only. This is array of longs to satisfy potential alignment problems
// with casting this to a headerData*
data = new long[((size = newSize)+sizeof(long)-1) / sizeof(long)];
headerData *h = (headerData*)data;
// Store the characteristics of the bucket
h->leaf = newLeaf;
h->parentId = newParentId;
h->nextBucket = newNextBucket;
h->prevBucket = newPrevBucket;
h->activeKeys = 0;
id = newId;
keyLength = newKeyLength;
keyNum = newKeyNum;
// Each tuple contains a key, the length of which is defined by
// the object and stored by the constructor, and a long link.
// Round keyLength up to a multiple of 4
myTupleSize = ((keyLength+3)&(~3)) + sizeof(long);
// The number of keys is as many tuples as will fit in the bucket, taking
// the header and initial pointer into account.
numKeys = (size - headerSize() - sizeof(long)) / tupleSize();
}
bucket::~bucket()
{
delete data;
}
// A little info for the programmer if s/he desires
void
bucket::info(void)
{
cout << "HeaderSize = " << dec << headerSize() << endl
<< "TupleSize = " << tupleSize() << endl
<< "ActiveKeys = " << header()->activeKeys << endl
<< "Previous = " << header()->prevBucket << " Next = " << header()->nextBucket << endl;
}
// Used during debugging. NB Key is assumed to start with a long
void
bucket::dump(void)
{
cout << "Bucket Dump" << endl;
cout << "Link: " << getLink(0) << endl;
for (long i = 1; i <= header()->activeKeys; i++)
cout << "Data: " << hex << *((long *)getKeyAddr(i)) << endl
<< "Link: " << dec << getLink(i) << endl;
}
// Actually return the data in the link itself
bucketId
bucket::getLink(long pos)
{
bucketId link;
bcopy(getLinkAddr(pos), (char*)&link, sizeof(bucketId));
return link;
}
// Store a new link
void
bucket::setLink(long pos, bucketId link)
{
bcopy((char*)&link, getLinkAddr(pos), sizeof(bucketId));
}
// The key in theObject and the link are inserted into the bucket in the
// appropriate spot.
void
bucket::insert(object& theObject,long link)
{
long pos = 0;
// This returns the position where theObject should be inserted in the bucket.
searchForKey(theObject,pos);
// Shift the existing data along one.
pKeyType tempKey = getKeyAddr(pos+1); // to stop it being called twice
bcopy((char *)tempKey,(char *)getKeyAddr(pos+2), (header()->activeKeys-pos)*tupleSize());
// Now put the key in.
pKeyType key = theObject.getKey(keyNum);
bcopy((char *)key,(char *)tempKey,keyLength);
delete key;
// Now the Link.
setLink(pos+1, link);
// Now we have one more...
header()->activeKeys++;
}
// Delete a key from a bucket
dbError
bucket::del(long pos)
{
// Can't delete an entry that is not there.
// Are there pos in the bucket?
if (pos > header()->activeKeys || pos < 0)
return dbErr(db_range);
// All after pos roll over and one falls out...
// NB For Inner nodes, we move starting at a link, not a key. This is
// so that the less than link at the start of the bucket is maintained
// accurate. The number to move is therefore 1 tuple less plus the extra
// link.
if (header()->leaf)
bcopy((char*)getKeyAddr(pos+1), (char*)getKeyAddr(pos), (header()->activeKeys - pos) * tupleSize());
else
if (header()->activeKeys > pos)
bcopy((char*)getLinkAddr(pos+1), (char*)getLinkAddr(pos), (header()->activeKeys - pos - 1) * tupleSize() + sizeof(long));
// And there was one less entry in the bucket.
header()->activeKeys--;
return dbErr(db_ok);
}
// Find a key in the bucket or return where it should go.
// If the key is not found then return the entry before where it would be
// If the key is found, return the spot where it is
// This function has been replaced by the binary chop version below.
#if 0
bool
bucket::searchForKey(object& theObject, long& ptr)
{
long i=1;
while (i <= header()->activeKeys) // NB pairs are numbered 1 to n
{
pKeyType key = getKeyAddr(i);
if (theObject.isLessThan(key, keyNum) )
{
ptr = i-1; // Needs to be inserted before i
return FALSE;
}
else if (theObject.isEqual(key, keyNum) )
{
ptr = i;
return TRUE;
}
i++;
}
ptr=i-1;
return FALSE;
}
#endif
bool
bucket::searchForKey(object& theObject, long& ptr)
{
long high = header()->activeKeys;
long low = 1; // NB pairs are numbered 1 to n
while (low <= high)
{
long mid = (low + high) / 2;
pKeyType key = getKeyAddr(mid);
if (theObject.isEqual(key, keyNum) )
{
ptr = mid;
return TRUE;
}
if (theObject.isLessThan(key, keyNum) )
high = mid - 1;
else
low = mid + 1;
}
// OK, now high is less than low. The key must therefore go after
// position high
ptr = high;
return FALSE;
}
// Search a bucket for a particular link and return it's position. This is
// used during delete, when trying to find where, in it's parent, a deleted
// bucket was positioned.
void
bucket::findIndex(long& ind, bool& found)
{
found = false;
for (long i=0; i <= header()->activeKeys; i++)
if (getLink(i) == ind)
{
found = true;
ind = i;
break;
}
}
// When a bucket is full, it contents need to be split into two buckets.
// The calling function creates the new bucket and fixes the forward and
// backward links.
// In the case of a leaf bucket, the first key in the new bucket is returned
// in 'out'. In the case of an inner node, the key deleted is returned.
void
bucket::sploosh(bucket* newBucket, pKeyType in, pKeyType out, object& theObject,
long link)
{
in = in;
long newKeyPos=0;
long delFromPos=header()->activeKeys/2+1;
// pKeyType medianKey = getKeyAddr(header()->activeKeys / 2);
// This loop is a little ugly! It determines where the key must be inserted.
for(long i=1;(!newKeyPos) && i<=header()->activeKeys;i++)
{
// is in < current key ?
if (theObject.isLessThan(getKeyAddr(i), keyNum)) // && (newKeyPos!=0))
newKeyPos = i;
}
if (!newKeyPos)
newKeyPos = i;
// Now put the excess values into the new bucket.
bcopy((char *)getKeyAddr(delFromPos),(char *)newBucket->getKeyAddr(1),
(header()->activeKeys-delFromPos+1)*tupleSize());
newBucket->setActiveKeys(header()->activeKeys-delFromPos+1);
setActiveKeys(delFromPos-1);
// Now add the new key where appropriate...
if (newKeyPos <= delFromPos)
insert(theObject,link);
else
newBucket->insert(theObject,link);
#ifdef __CC
memcpy(out,newBucket->getKeyAddr(1),(int)keyLength);
#else
memcpy(out,newBucket->getKeyAddr(1),keyLength);
#endif
// If this is an inner node then move everything in the new bucket down one.
// Note that it is not the first tuple that is deleted, but the link number
// 0 and key number 1. This deletes the key that will be sent back in 'out'
// and leaves the correct link in the extra spot.
if (header()->leaf == INNER)
{
bcopy( newBucket->getLinkAddr(1),(char *)newBucket->firstTuple(),
(newBucket->header()->activeKeys-1)*tupleSize()+sizeof(long));
newBucket->setActiveKeys(newBucket->header()->activeKeys-1);
}
}