From 9a47dab78020e2df9ad8df03d40b18f9b1e89f57 Mon Sep 17 00:00:00 2001
From: Matt Joiner <anacrolix@gmail.com>
Date: Mon, 16 Sep 2024 17:43:33 +1000
Subject: [PATCH 1/2] Fix really slow BSI.ClearValues

---
 roaring64/bsi64.go | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)

diff --git a/roaring64/bsi64.go b/roaring64/bsi64.go
index 5d6019db..812ee4c7 100644
--- a/roaring64/bsi64.go
+++ b/roaring64/bsi64.go
@@ -973,21 +973,10 @@ func ClearBits(foundSet, target *Bitmap) {
 
 // ClearValues removes the values found in foundSet
 func (b *BSI) ClearValues(foundSet *Bitmap) {
-
-	var wg sync.WaitGroup
-	wg.Add(1)
-	go func() {
-		defer wg.Done()
-		b.eBM.AndNot(foundSet)
-	}()
-	for i := 0; i < b.BitCount(); i++ {
-		wg.Add(1)
-		go func(j int) {
-			defer wg.Done()
-			b.bA[j].AndNot(foundSet)
-		}(i)
+	b.eBM.AndNot(foundSet)
+	for i := range b.bA {
+		b.bA[i].AndNot(foundSet)
 	}
-	wg.Wait()
 }
 
 // NewBSIRetainSet - Construct a new BSI from a clone of existing BSI, retain only values contained in foundSet

From e5faa9e54325d828a5aa2e745c73f92b0a7962da Mon Sep 17 00:00:00 2001
From: Matt Joiner <anacrolix@gmail.com>
Date: Mon, 9 Mar 2026 21:35:13 +1100
Subject: [PATCH 2/2] Document why ClearValues is serial

The goroutine-per-bit-plane approach it replaced was slower in practice.
Goroutine creation overhead dominates for typical BSI sizes, and the cost
compounds when ClearValues is called once per term in a deletion pass over
an entire index (the pattern used in caterwaul's DeleteGroups).
---
 roaring64/bsi64.go | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/roaring64/bsi64.go b/roaring64/bsi64.go
index 812ee4c7..484dd8ea 100644
--- a/roaring64/bsi64.go
+++ b/roaring64/bsi64.go
@@ -971,7 +971,13 @@ func ClearBits(foundSet, target *Bitmap) {
 	target.AndNot(foundSet)
 }
 
-// ClearValues removes the values found in foundSet
+// ClearValues removes from the BSI all values whose column IDs are in
+// foundSet, modifying the BSI in place.
+//
+// The implementation is intentionally serial. A previous goroutine-per-bit-plane
+// approach was slower in practice: goroutine creation overhead dominated for
+// typical BSI sizes, and the cost compounds when ClearValues is called in a
+// tight loop (e.g. once per term across an entire index during a deletion pass).
 func (b *BSI) ClearValues(foundSet *Bitmap) {
 	b.eBM.AndNot(foundSet)
 	for i := range b.bA {