view release on metacpan or  search on metacpan

lib/Algorithm/MinPerfHashTwoLevel.pm  view on Meta::CPAN

    if ( !defined $self->{seed} ) {
                       #1234567812345678
        $self->{seed}= "MinPerfHash2Levl";
        delete $self->{state};
    }
    return $self->{seed};
}

sub set_seed {
    my ($self,$value)= @_;
    my $seed= $self->_seed($value);
    return $self->{constructor_seed}= $seed;
}

sub get_seed {
    my ($self)= @_;
    return $self->_seed();
}


sub get_state {
    my $self= shift;
    return $self->{state} ||= seed_state($self->_seed());
}

sub get_failed_seeds {
    my $self= shift;
    return @{$self->{failed_seeds}||[]};
}

1;
__END__

=head1 NAME

Algorithm::MinPerfHashTwoLevel - construct a "two level" minimal perfect hash

=head1 SYNOPSIS

  use Algorithm::MinPerfHashTwoLevel;
  my $buckets= Algorithm::MinPerfHashTwoLevel->compute(\%source_hash);

=head1 DESCRIPTION

This module implements an algorithm to construct (relatively efficiently) a
minimal perfect hash using the "two level" algorithm. A perfect hash is one
which has no collisions for any keys, a minimal perfect hash has exactly the
same number of buckets as it has keys. The "two level" algorithm involves
computing two hash values for each key. The first is used as an initial lookup
into the bucket array to find a mask value which is used to modify the second
hash value to determine the actual bucket to read from. This module computes
the appropriate mask values.

In this implementation only one 64 bit hash value is computed, but the high
and low 32 bits are used as if there were two hash values. The hash function
used is SipHash 1-3. The full 64 bit hash is called h0, the high 32 bits are
called h1, and the low 32 bits are called h2.)

Computing the hash and mask is done in C (via XS).

The process for looking up a value in a two level hash with n buckets is
as follows:

    0. compute the h0 for the key. (giving: h1 = h0 >> 32; h2 = h0 & 0xFFFFFFFF;)
    1. compute idx1 = h1 % n;
    2. find the xor_val for bucket[idx1]
    3. if the xor_val is zero we are done, the key is not in the hash
    4. compute idx2:
        if int(xor_val) < 0
            idx2 = -xor_val-1
        else
            idx2 = INTHASH(h2 ^ xor_val) % n;
    5. compare the key data associated with bucket[idx2] with the key provided
    6. if they match return the desired value, otherwise the key is not in the hash.

In essence this module performs the task of computing the xor_val for
each bucket such that the idx2 for every element is unique, it does it in C/XS so
that it is fast.

The INTHASH() function used is:

    x = ((x >> 16) ^ x) * 0x45d9f3b;
    x = ((x >> 16) ^ x) * 0x45d9f3b;
    x = ((x >> 16) ^ x);

which is just a simple 32 bit integer hash function I found at
https://stackoverflow.com/a/12996028, but any decent reversible
integer hash function would do. 

*NOTE* in Perl a given string may have differing binary representations
if it is encoded as utf8 or not. This module uses the same conventions
as Perl itself, which is that keys are stored in their minimal form when
possible, and are only stored in their unicode (utf8) form when they
cannot be downgraded to latin-1. This ensures that the unicode and latin-1
representations of a given string are treated as the same key. This module
deals with this by "normalizing" the keys and values into latin-1, but
tracking the representation as a flag. See key_normalized and key_is_utf8
(and their 'val' equivalents) documented in the construct method.

=head2 METHODS

=over 4

=item new

Construct a new Algorithm::MinPerfHashTwoLevel object. Optional arguments
which may be provided are 'source_hash' which is a hash reference to use
as the source for the minimal perfect hash, 'seed' which is expected to be
a 16 byte string, and 'debug' which is expected to be 0 or 1, as well
as variant, which may be 5 (use version v0.14 for variants before 5). 
The default is 5.

=item compute

Compute the buckets for a two level minimal perfect hash. Either operates
on the 'source_hash' passed into the constructor, or requires one to passed
in as an argument.

Returns an array of hashes containing information about each bucket:

          {