From f11b025b77d618ef2b0eeb77536ba2fe9ece21a1 Mon Sep 17 00:00:00 2001
From: Christoph Urlacher <christoph.urlacher@protonmail.com>
Date: Sat, 19 Sep 2026 11:43:19 +0200
Subject: [PATCH] Align ResultsDataWriteGroups query with
 global-occurrences-onwrite.sh

---
 scripts/Queries/ResultsDataWriteGroups.pm | 50 ++++++++++-------------
 1 file changed, 22 insertions(+), 28 deletions(-)

diff --git a/scripts/Queries/ResultsDataWriteGroups.pm b/scripts/Queries/ResultsDataWriteGroups.pm
index 078e23c..58d38b6 100644
--- a/scripts/Queries/ResultsDataWriteGroups.pm
+++ b/scripts/Queries/ResultsDataWriteGroups.pm
@@ -16,27 +16,23 @@ sub query {
     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.
+    # Filters are applied per column because of the disjunct branch union
     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 );
+    $instr_extra =~ s/\bp\.injection_instr_absolute\b/g.instr2_absolute/g;
 
     # This is the same as ResultsData.pm (how much faultspace
-    # area ended up in each possible outcome), but applies the
-    # possible writegroups fix.
+    # area ended up in each possible outcome), but includes the
+    # "inject-on-write" fault model like in global-occurrences-onwrite.sh
     #
-    # 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).
+    # A fault in a written value is treated as a fault in the next access to
+    # the same address.
+    # FAIL weights the write area by SUM(t.width), this query uses
+    # (g.time2 - g.time1 + 1) instead.
     my $resulttype_order =
         "FIELD(resulttype, 'OK_MARKER', 'FAIL_MARKER', 'DETECTED_MARKER',"
       . " 'GROUP1_MARKER', 'GROUP2_MARKER', 'GROUP3_MARKER', 'GROUP4_MARKER',"
@@ -68,22 +64,20 @@ FROM (
 
     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 = '$experiment'
+    SELECT v.benchmark AS benchmark, r.resulttype AS resulttype,
+        SUM(g.time2 - g.time1 + 1) AS faults
+    FROM trace g
+    JOIN variant v ON v.id = g.variant_id
+    JOIN trace t ON t.variant_id = g.variant_id
+        AND t.data_physical_address = g.data_physical_address
+        AND t.instr1 = g.instr2 + 1
+    JOIN fspgroup h ON h.variant_id = t.variant_id
+        AND h.data_physical_address = t.data_physical_address
+        AND h.instr2 = t.instr2
+        AND h.fspmethod_id = (SELECT id FROM fspmethod WHERE method = 'basic')
+    JOIN result_GenericExperimentMessage r ON r.pilot_id = h.pilot_id
+    WHERE g.accesstype = 'W'
+        AND v.variant = '$experiment'$data_extra$instr_extra
 ) x
 GROUP BY benchmark, resulttype
 ORDER BY benchmark, $resulttype_order;";