forked from tzaeschke/tinspin-indexes
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathTODO.txt
17 lines (16 loc) · 903 Bytes
/
TODO.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
CritBit
=======
- Remove Node.posFirstBit this should always be (parent.posDiff+1)
- Memory optimisation: Nodes should have 'long' instead of 'long[]' for infixes.
Infixes that cross 64bit boundaries can be stored in two separate nodes.
This would introduce a special case where a node may have only a single child
which contains a continuation of the infix -> 'null' would not indicate that value is 'null'
-> use NULL indicator object, see next point
- Memory optimisation: Merge children and subnodes into single field, -> saves
2*32bits per Node.
- Fix warnings in CritBit.x.checkMatchOrigKD()
- Fix warnings in CritBit.x.unsetBitAfterSplit()
- Why do Point queries scale much worse than kd-Tree? Because of merging???? O(k)
- Profile:
- Consider using table for maskDst in CritBit.x.unsetBitAfterSplit() / setBitAfterSplit()
--> Aternatively, consider operating on merged value?