/**
* Excel has funny behaviour when the some elements in the search vector are the wrong type.
*
*/
private static int PerformBinarySearch(ValueVector vector, LookupValueComparer lookupComparer)
{
// both low and high indexes point to values assumed too low and too high.
BinarySearchIndexes bsi = new BinarySearchIndexes(vector.Size);
while (true)
{
int midIx = bsi.GetMidIx();
if (midIx < 0)
{
return(bsi.GetLowIx());
}
CompareResult cr = lookupComparer.CompareTo(vector.GetItem(midIx));
if (cr.IsTypeMismatch)
{
int newMidIx = HandleMidValueTypeMismatch(lookupComparer, vector, bsi, midIx);
if (newMidIx < 0)
{
continue;
}
midIx = newMidIx;
cr = lookupComparer.CompareTo(vector.GetItem(midIx));
}
if (cr.IsEqual)
{
return(FindLastIndexInRunOfEqualValues(lookupComparer, vector, midIx, bsi.GetHighIx()));
}
bsi.narrowSearch(midIx, cr.IsLessThan);
}
}
/**
* Excel seems to handle mismatched types initially by just stepping 'mid' ix forward to the
* first compatible value.
* @param midIx 'mid' index (value which has the wrong type)
* @return usually -1, signifying that the BinarySearchIndex has been narrowed to the new mid
* index. Zero or greater signifies that an exact match for the lookup value was found
*/
private static int HandleMidValueTypeMismatch(LookupValueComparer lookupComparer, ValueVector vector,
BinarySearchIndexes bsi, int midIx)
{
int newMid = midIx;
int highIx = bsi.GetHighIx();
while (true)
{
newMid++;
if (newMid == highIx)
{
// every element from midIx to highIx was the wrong type
// move highIx down to the low end of the mid values
bsi.narrowSearch(midIx, true);
return(-1);
}
CompareResult cr = lookupComparer.CompareTo(vector.GetItem(newMid));
if (cr.IsLessThan && newMid == highIx - 1)
{
// move highIx down to the low end of the mid values
bsi.narrowSearch(midIx, true);
return(-1);
// but only when "newMid == highIx-1"? slightly weird.
// It would seem more efficient to always do this.
}
if (cr.IsTypeMismatch)
{
// keep stepping over values Until the right type Is found
continue;
}
if (cr.IsEqual)
{
return(newMid);
}
// Note - if moving highIx down (due to lookup<vector[newMid]),
// this execution path only moves highIx it down as far as newMid, not midIx,
// which would be more efficient.
bsi.narrowSearch(newMid, cr.IsLessThan);
return(-1);
}
}
/**
* Excel has funny behaviour when the some elements in the search vector are the wrong type.
*
*/
private static int PerformBinarySearch(ValueVector vector, LookupValueComparer lookupComparer)
{
// both low and high indexes point to values assumed too low and too high.
BinarySearchIndexes bsi = new BinarySearchIndexes(vector.Size);
while (true)
{
int midIx = bsi.GetMidIx();
if (midIx < 0)
{
return bsi.GetLowIx();
}
CompareResult cr = lookupComparer.CompareTo(vector.GetItem(midIx));
if (cr.IsTypeMismatch)
{
int newMidIx = HandleMidValueTypeMismatch(lookupComparer, vector, bsi, midIx);
if (newMidIx < 0)
{
continue;
}
midIx = newMidIx;
cr = lookupComparer.CompareTo(vector.GetItem(midIx));
}
if (cr.IsEqual)
{
return FindLastIndexInRunOfEqualValues(lookupComparer, vector, midIx, bsi.GetHighIx());
}
bsi.narrowSearch(midIx, cr.IsLessThan);
}
}
/**
* Excel seems to handle mismatched types initially by just stepping 'mid' ix forward to the
* first compatible value.
* @param midIx 'mid' index (value which has the wrong type)
* @return usually -1, signifying that the BinarySearchIndex has been narrowed to the new mid
* index. Zero or greater signifies that an exact match for the lookup value was found
*/
private static int HandleMidValueTypeMismatch(LookupValueComparer lookupComparer, ValueVector vector,
BinarySearchIndexes bsi, int midIx)
{
int newMid = midIx;
int highIx = bsi.GetHighIx();
while (true)
{
newMid++;
if (newMid == highIx)
{
// every element from midIx to highIx was the wrong type
// move highIx down to the low end of the mid values
bsi.narrowSearch(midIx, true);
return -1;
}
CompareResult cr = lookupComparer.CompareTo(vector.GetItem(newMid));
if (cr.IsLessThan && newMid == highIx - 1)
{
// move highIx down to the low end of the mid values
bsi.narrowSearch(midIx, true);
return -1;
// but only when "newMid == highIx-1"? slightly weird.
// It would seem more efficient to always do this.
}
if (cr.IsTypeMismatch)
{
// keep stepping over values Until the right type Is found
continue;
}
if (cr.IsEqual)
{
return newMid;
}
// Note - if moving highIx down (due to lookup<vector[newMid]),
// this execution path only moves highIx it down as far as newMid, not midIx,
// which would be more efficient.
bsi.narrowSearch(newMid, cr.IsLessThan);
return -1;
}
}