use v6.d;
use MaildirIndexer::Index;
unit class MaildirIndexer::Index::ByAddresses does MaildirIndexer::Index;
use MaildirIndexer::LogTimelineSchema;
use MaildirIndexer::Email;

# most of this is copied from
# p6-Algorithm-NaiveBayes:auth<cpan:TITSUKI>:ver<0.04>, in particular the
# Algorithm::NaiveBayes::Classifier::Bernoulli class

has Array[Str] %!addresses-for-file;
has BagHash $!count-by-address-and-mailbox .= new;
has BagHash $!known-addresses .= new;
has BagHash $!count-by-mailbox .= new;
has Numeric %!p-address-given-mailbox;
has Int $!total-count;

has Numeric %!cached-p-given-mailbox;
has Supplier $!update-cache .= new;

constant $NOT-ZERO = 1e-15;

submethod TWEAK() {
    $!update-cache.Supply.stable(10).tap({ self.do-update-cache() });
}

method dump() {
}

submethod account-for(Str @addresses,Str $mailbox,Int $step) {
    $!total-count += $step;
    $!count-by-mailbox{$mailbox} += $step;

    for @addresses -> Str $addr {
        my Pair $pair = ( $addr<> => $mailbox<> );
        $!known-addresses{$addr} += $step;
        my Numeric $count = $!count-by-address-and-mailbox{$pair} += $step;

        if ($count) {
            my Numeric $a = $NOT-ZERO + $count;
            my Numeric $b = 1 + $!count-by-mailbox{$mailbox};
            %!p-address-given-mailbox{$pair} = $a / $b;
        }
        else {
            %!p-address-given-mailbox{$pair} :delete;
        }
    }

    $!update-cache.emit(Any);
}

submethod do-update-cache() {
    MaildirIndexer::LogTimelineSchema::Index::Cache.log: {
        for $!count-by-mailbox.keys -> Str $mailbox {
            my Numeric $p = $!count-by-mailbox{$mailbox} / $!total-count;

            for $!known-addresses.keys -> Str $addr {
                my $addr-p = %!p-address-given-mailbox{$addr<> => $mailbox<>} // $NOT-ZERO;
                $p *= 1 - $addr-p;
            }

            %!cached-p-given-mailbox{$mailbox} = $p;
        }
    }
}

method add-mail(MaildirIndexer::Email:D $email, Str:D $mailbox --> Nil) {
    MaildirIndexer::LogTimelineSchema::Index::Add.log: :class('ByAddresses'),:$mailbox, -> {
        # ignore adding the same file twice, files in maildirs are
        # immutable
        return if %!addresses-for-file{ $email.path }:exists;

        my Str @addresses = $email.addresses or return;
        %!addresses-for-file{ $email.path } := @addresses;

        self.account-for(@addresses,$mailbox,1);

        return;
    }
}

method del-path(IO:D $file, Str:D $mailbox --> Nil) {
    MaildirIndexer::LogTimelineSchema::Index::Rm.log: :class('ByAddresses'),:$mailbox, -> {
        # using assignment would fail when the path isn't present in
        # the hash, because it tries to assign the (undefined)
        # Array[Str] as a single element, instead of splatting it;
        # also, binding is faster because it avoids a copy
        my Str @addresses := %!addresses-for-file{$file.path}:delete or return;

        self.account-for(@addresses,$mailbox,-1);

        return;
    }
}

submethod predict-mailbox-given-addresses(@addresses) {
    my Numeric %prediction;

    for $!count-by-mailbox.keys -> Str $mailbox {
        my Numeric $p = %!cached-p-given-mailbox{$mailbox} // $NOT-ZERO;

        for @addresses -> Str $addr {
            my $addr-p = %!p-address-given-mailbox{$addr<> => $mailbox<>} // $NOT-ZERO;
            $p *= $addr-p / ( 1- $addr-p );
        }

        %prediction{$mailbox} = $p;
    }

    return %prediction;
}

method mailbox-for-email(MaildirIndexer::Email:D $email --> Str) {
    my Str $result;
    MaildirIndexer::LogTimelineSchema::Index::Find.log: :class('ByAddresses'), -> {
        my %prediction = self.predict-mailbox-given-addresses($email.addresses);

        my @most-probable-mailboxes = %prediction.pairs.sort(*.value);

        if @most-probable-mailboxes -> $_ { $result = .[*-1].key }
    }
    return $result;
}