BloomFilter.h

/*
 * Copyright(c) Sophist Solutions, Inc. 1990-2026.  All rights reserved
 */
#ifndef _Stroika_Foundation_Cache_BloomFilter_h_
#define _Stroika_Foundation_Cache_BloomFilter_h_ 1
 
#include "Stroika/Foundation/StroikaPreComp.h"
 
#include <type_traits>
 
#include "Stroika/Foundation/Characters/ToString.h"
#include "Stroika/Foundation/Common/Common.h"
#include "Stroika/Foundation/Common/TypeHints.h"
#include "Stroika/Foundation/Containers/Sequence.h"
#include "Stroika/Foundation/Containers/Set.h"
#include "Stroika/Foundation/Cryptography/Digest/Hash.h"
 
/**
 *  \note Code-Status:  <a href="Code-Status.md#Beta">Beta</a>
 */
 
namespace Stroika::Foundation::Cache {
 
    /**
     * \brief a Bloom filter is a probablistic set, which returns either "probably in set" or "definitely not in set"
     *
     * Description:
     *      http://en.wikipedia.org/wiki/Bloom_filter
     *          fHashFunction.size () corresponds to 'k'
     *          fBitCount corresponds to 'm'
     * 
     * \note - when the bloom filter starts getting full, it becomes less useful. A good way to decide
     *         when its time to recreate a larger bloom filter is with:
     *          \code
     *              runToProbOfFalsePositive = 0.5;   // ?? depends on your tolerance and cost of resizing bloom filter
     *              if (addressesProbablyUsed.GetStatistics ().ProbabilityOfFalsePositive () >= runToProbOfFalsePositive) {
     *                  ... resize bloom filter - means losing all existing data...
     *          \endcode
     * 
     *  \note BloomFilter does NOT support ICache<> API, due to not having a Lookup () function - because it
     *        can generate false positives. MAY want to reconsider this stand???? --LGP 2026-03-27
     */
    template <typename T>
    class BloomFilter {
    public:
        using HashResultType = uint64_t; // often something smaller will be used, but best to pick a wider type here, and @todo maybe make it a template parameter
 
    public:
        /**
         *  \note uses std::function instead of directly storing function-object (losing efficiency) because must maintain
         *        an array of these, and thats trickier if we allow them to have different types.
         */
        using HashFunctionType = function<HashResultType (T)>;
 
    public:
        /**
         * \brief kDefaultDesiredFalsePositivityRate is used for constructors given expectedMaxSetSize to compute the appropriate #bits for the BloomFilter or for explicit calls to OptimizeBitSize()
         */
        static inline constexpr double kDefaultDesiredFalsePositivityRate = 0.1;
 
    public:
        /**
         *  The constructor with an explicit set of hash functions and bitCount uses those hash functions and bitCount.
         *  The constructor taking expectedMaxSetSize calls OptimizeBitSize and OptimizeNumberOfHashFunctions to calculate
         *  optimimum numbers of hash functions and bitstorage for the given expected set size. It also uses DeriveIndependentHashFunctions ()
         *  to automatically create (semi) independent hash functions from the original argument one (this maybe could use tuning/work).
         * 
         *  \note - for the hash functions, you can simple use std::hash<T>, but the hash implementation on gcc (libstdc++) works quite badly
         *          as a crytographic hash for integer types. So the default is to use the Stroika hash function.
         *          You can directly use Cryptography::Digest::Hash<> instead (the default), and this will generaly work better where its
         *          defined.
         */
        BloomFilter (const Containers::Sequence<HashFunctionType>& hashFunctions, size_t bitCount);
        BloomFilter (size_t expectedMaxSetSize, const HashFunctionType& defaultHashFunction = Cryptography::Digest::Hash<T>{},
                     double desiredFalsePositivityRate = kDefaultDesiredFalsePositivityRate);
        BloomFilter (BloomFilter&& src) noexcept = default;
        BloomFilter (const BloomFilter& src)     = default;
 
    public:
        nonvirtual BloomFilter& operator= (BloomFilter&& src) noexcept = default;
        nonvirtual BloomFilter& operator= (const BloomFilter& src)     = default;
 
    public:
        static unsigned int OptimizeBitSize (size_t nElements, double desiredFalsePositiveProbability = kDefaultDesiredFalsePositivityRate);
 
    public:
        static unsigned int OptimizeNumberOfHashFunctions (size_t setSize, size_t bitSize);
 
    public:
        /**
         * From http://en.wikipedia.org/wiki/Bloom_filter: 
         *      (1-exp(-kn/m))^^k
         * 
         *      n elements inserted so far (other letter/vars see struct docs above)
         * 
         *  This formula is not good - see https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=903775#:~:text=The%20probability%20of%20a%20false,mapped%20into%20the%20Bloom%20filter).
         *  and its especially not good because it should mostly depend on number of bits set, not nElementsInsertedSoFar.
         *  IF/WHEN I can fix that, maybe make this a method of the 'Statistics' object.
         * 
         *  See comments in GetFractionFull() about alternative better possible approach.
         */
        static double     ProbabilityOfFalsePositive (size_t hashFunctionCount, size_t bitCount, size_t nElementsInsertedSoFar);
        nonvirtual double ProbabilityOfFalsePositive (size_t nElementsInsertedSoFar) const;
 
    public:
        static Containers::Sequence<HashFunctionType> DeriveIndependentHashFunctions (const HashFunctionType& h, size_t repeatCount);
 
    public:
        /**
         *  return true if this was probably a change, and false if it was probably already present
         */
        nonvirtual bool Add (Common::ArgByValueType<T> elt);
 
    public:
        /**
         */
        nonvirtual void clear ();
 
    public:
        /**
         *  \brief returns true if this (probably) Contains 'elt' - and false otherwise.
         *
         *  False positive retrieval results are possible, but false negatives are not;
         * 
         *  Call GetEffectiveOptions ().ProbabilityOfFalsePositive (nEltsAdded) to an estimate of the probability of false positives.
         */
        nonvirtual bool ProbablyContains (Common::ArgByValueType<T> elt) const;
 
        [[deprecated ("Since Stroika v3.0d10 use ProbablyContains")]] bool Contains (Common::ArgByValueType<T> elt) const
        {
            return ProbablyContains (elt);
        }
 
    public:
        struct Statistics;
 
    public:
        /**
         */
        nonvirtual Statistics GetStatistics () const;
 
    private:
        vector<HashFunctionType> fHashFunctions_;
        vector<bool>             fBits_;                         // @todo use Set_Bitstring
        unsigned int             fActualAddCalls_{};             // to support useful statistics, not strictly needed
        unsigned int             fApparentlyDistinctAddCalls_{}; // ""
    };
 
    /**
     */
    template <typename T>
    struct BloomFilter<T>::Statistics {
        size_t       fHashFunctions;
        size_t       fBitCount;
        size_t       fBitsSet;
        unsigned int fActualAddCalls;
        unsigned int fApparentlyDistinctAddCalls;
 
    public:
        /**
         *  \brief return number from 0..1 indicating 'percent' of the bloom filter full so its clear when to refresh/resize it (or at least when its going to start getting lots of false positives)
         * 
         *  @todo see comments on ProbabilityOfFalsePositive () - should be able to compute ProbabilityOfFalsePositive based just on this info
         *  (roughly pct bits unset / number of hash functions). But I cannot find a reference for that so leaving this as is for now.
         */
        nonvirtual double GetFractionFull () const;
 
    public:
        /**
         */
        nonvirtual double ProbabilityOfFalsePositive () const;
 
    public:
        /**
         *  @see Characters::ToString ()
         */
        nonvirtual Characters::String ToString () const;
    };
 
}
 
/*
 ********************************************************************************
 ***************************** Implementation Details ***************************
 ********************************************************************************
 */
#include "BloomFilter.inl"
 
#endif /*_Stroika_Foundation_Cache_BloomFilter_h_ */