bitvector

Uncompressed, dynamically resizeable bitset, similar to java.util.BitSet.

• Allocation-free enumeration of set bits
• 4 byte words, avoids Long emulation in JS

Based on Adrian Papari initial implementation and has been enhanced to add new functionality and modern Kotlin support:

• introduce new operations (first, last, clear, invert and more)
• split into 2 implementations: BitVector and MutableBitVector
• make use of kotlin stdlib countOneBits, countLeadingZeroBits

implementation("io.github.adokky:bitvector:0.9")

val bv: BitVector = bitsOf(1, 2, 56, 64, 128, 129, 130, 131, 420)

val bv = MutableBitVector()
bv[142] = true // or bv.set(142)
assert(142 in bv)

bv.unset(142)  // or bv[142] = false
assert(142 !in bv)

val a = bitsOf(0, 1, 2, 3, 120, 130)
val b = bitsOf(0, 1, 2, 120, 121, 122, 123, 130)

assert(a.and(b) == bitsOf(0, 1, 2, 120, 130))

// 1, 4 contained in other 
bitsOf(1, 4) in bitsOf(0, 1, 4, 5, 6) 

bitsOf(*bits).forEachBit { println("bit $it says hi") }

// can be combined with filter/map etc
bitsOf(*bits).forEach { println("bit $it says hi") }

Mapping true bit positions into integers, inserting each into an IntBag (thin wrapper around int[]).

artemis-odb's BitVector was the basis for this implementation. The benchmark setup favors the artemis implementation, as it provides an optimized toIntBag method: it serves as a reference for best possible performance.

See jmh-logs for the full logs.

Discrepancy to artemis BitVector is unwelcome. The implementation is for the most part the same, except that this implementation uses int for words, instead of long. 4 or 8 byte words did not have a significant impact on performance.

The for-loop performs poorly due to all the Integer boxing, extra indirection and allocation, compared to forEachBit.

java.util.BitSet suffers from not having a way of enumerating all bits at once, instead relying on repeatedly calling nextSetBit.