package ResultsDataWriteGroups;

use strict;
use warnings;
use diagnostics;

use FindBin;
use lib "$FindBin::Bin/../Modules";
use Filters;

use feature 'say';

sub query {
    my ( $experiment, $experiment_dir, @filter_config_names ) = @_;

    my $extra =
      Filters::build_filter_clause( $experiment_dir, @filter_config_names );

    # The write branch applies the data filters on the trace and the
    # instruction filters on the pilot, so the clause is built per column.
    my $data_extra =
      Filters::build_filter_clause_restricted( $experiment_dir, qr/^g\./,
        @filter_config_names );
    my $instr_extra =
      Filters::build_filter_clause_restricted( $experiment_dir, qr/^p\./,
        @filter_config_names );

    # This is the same as ResultsData.pm (how much faultspace
    # area ended up in each possible outcome), but applies the
    # possible writegroups fix.
    #
    # At first I have written this query similar to ResultsData.pm, with
    # an inner join to reconstruct the "fixed" fspgroup table within the query.
    # This was extremely slow since it resulted in a large cross product
    # between the write groups/classes and the write pilots injections
    # (also I don't think the indices were helping the way I wrote the query).
    #
    # This version now constructs the result in two
    # disjunct branches (as the data is disjunct in known_outcome = 0/1).
    my $resulttype_order =
        "FIELD(resulttype, 'OK_MARKER', 'FAIL_MARKER', 'DETECTED_MARKER',"
      . " 'GROUP1_MARKER', 'GROUP2_MARKER', 'GROUP3_MARKER', 'GROUP4_MARKER',"
      . " 'TIMEOUT', 'TRAP', 'WRITE_TEXTSEGMENT', 'ACCESS_OUTERSPACE',"
      . " 'SDC', 'UNKNOWN')";

    my $querystring = "SELECT
benchmark, resulttype, SUM(faults) AS faults
FROM (
    SELECT v.benchmark AS benchmark, rc.resulttype AS resulttype, SUM(cl.len * rc.cnt) AS faults
    FROM variant v
    JOIN (
        SELECT g.variant_id, p.id AS pilot_id, SUM(g.time2 - g.time1 + 1) AS len
        FROM trace g
        JOIN fsppilot p ON p.variant_id = g.variant_id
            AND p.known_outcome = 0
            AND p.instr2 = g.instr2
            AND p.data_physical_address = g.data_physical_address
        WHERE 1 = 1$extra
        GROUP BY g.variant_id, p.id
    ) cl ON cl.variant_id = v.id
    JOIN (
        SELECT r.pilot_id, r.resulttype, COUNT(*) AS cnt
        FROM result_GenericExperimentMessage r
        GROUP BY r.pilot_id, r.resulttype
    ) rc ON rc.pilot_id = cl.pilot_id
    WHERE v.variant = '$experiment'
    GROUP BY v.id, rc.resulttype

    UNION ALL

    SELECT v.benchmark AS benchmark, rc.resulttype AS resulttype, wl.len * rc.cnt AS faults
    FROM variant v
    JOIN (
        SELECT g.variant_id, SUM(g.time2 - g.time1 + 1) AS len
        FROM trace g
        WHERE g.accesstype = 'W'$data_extra
        GROUP BY g.variant_id
    ) wl ON wl.variant_id = v.id
    JOIN (
        SELECT p.variant_id, r.resulttype, COUNT(*) AS cnt
        FROM fsppilot p
        JOIN result_GenericExperimentMessage r ON r.pilot_id = p.id
        WHERE p.known_outcome = 1$instr_extra
        GROUP BY p.variant_id, r.resulttype
    ) rc ON rc.variant_id = v.id
    WHERE v.variant = '$experiment'
) x
GROUP BY benchmark, resulttype
ORDER BY benchmark, $resulttype_order;";

    say $querystring;

    return $querystring;
}

sub args { return "--batch --raw"; }

sub filename {
    my @filter_config_names = grep { defined && length } @_;
    my $suffix =
      @filter_config_names ? "_" . join( "+", sort @filter_config_names ) : "";
    return "resultsdata_writegroups${suffix}.csv";
}

sub postprocess { $_[0] =~ s/\t/,/g; }

1;