fix(testing): save measurements only in the benchmark framework

Author: not-matthias

testing/fork.sh

@@ -122,6 +122,7 @@ find . -type f -name "*.go" -not -name "*_test.go" -exec sed -i 's|"testing"|tes
 restore_files "${CODSPEED_FILES[@]}"
 
 apply_patch "patches/benchmark_stopbenchmark_fail.patch" 10 ".."
+apply_patch "patches/benchmark_stoptimer_mitigation.patch" 10 ".."
 
 
 # Run pre-commit and format the code

testing/patches/benchmark_stoptimer_mitigation.patch

@@ -0,0 +1,68 @@
+diff --git a/testing/testing/benchmark.go b/testing/testing/benchmark.go
+index 8fdea90..9473443 100644
+--- a/testing/testing/benchmark.go
++++ b/testing/testing/benchmark.go
+@@ -93,6 +93,10 @@ type codspeed struct {
+ 	startTimestamp  uint64
+ 	startTimestamps []uint64
+ 	stopTimestamps  []uint64
++
++	// Indicates whether a measurement has been saved already. This aims to prevent saving measurements
++	// twice, because `b.Loop()` saves them internally as well but is also called from runN
++	savedMeasurement bool
+ }
+
+ // B is a type passed to [Benchmark] functions to manage benchmark
+@@ -160,6 +164,7 @@ func (b *B) StartTimerWithoutMarker() {
+ 		// b.startBytes = memStats.TotalAlloc
+ 		b.start = highPrecisionTimeNow()
+ 		b.timerOn = true
++		b.savedMeasurement = false
+ 		// b.loop.i &^= loopPoisonTimer
+ 	}
+ }
+@@ -186,14 +191,22 @@ func (b *B) StopTimerWithoutMarker() {
+ 		b.timerOn = false
+ 		// If we hit B.Loop with the timer stopped, fail.
+ 		// b.loop.i |= loopPoisonTimer
++	}
++}
+
+-		// For b.N loops: This will be called in runN which sets b.N to the number of iterations.
+-		// For b.Loop() loops: loopSlowPath sets b.N to 0 to prevent b.N loops within b.Loop. However, since
+-		// we're starting/stopping the timer for each iteration in the b.Loop() loop, we can use 1 as
+-		// the number of iterations for this round.
+-		b.codspeedItersPerRound = append(b.codspeedItersPerRound, max(int64(b.N), 1))
+-		b.codspeedTimePerRoundNs = append(b.codspeedTimePerRoundNs, timeSinceStart)
++func (b *B) SaveMeasurement() {
++	if b.savedMeasurement {
++		return
+ 	}
++	b.savedMeasurement = true
++
++	// For b.N loops: This will be called in runN which sets b.N to the number of iterations.
++	// For b.Loop() loops: loopSlowPath sets b.N to 0 to prevent b.N loops within b.Loop. However, since
++	// we're starting/stopping the timer for each iteration in the b.Loop() loop, we can use 1 as
++	// the number of iterations for this round.
++	timeSinceStart := highPrecisionTimeSince(b.start)
++	b.codspeedItersPerRound = append(b.codspeedItersPerRound, max(int64(b.N), 1))
++	b.codspeedTimePerRoundNs = append(b.codspeedTimePerRoundNs, timeSinceStart)
+ }
+
+ func (b *B) StopTimer() {
+@@ -296,6 +309,7 @@ func (b *B) __codspeed_root_frame__runN(n int) {
+ 	b.ResetTimer()
+ 	b.StartTimer()
+ 	b.benchFunc(b)
++	b.SaveMeasurement()
+ 	b.StopTimer()
+ 	b.previousN = n
+ 	b.previousDuration = b.duration
+@@ -614,6 +628,7 @@ func (b *B) loopSlowPath() bool {
+ // whereas b.N-based benchmarks must run the benchmark function (and any
+ // associated setup and cleanup) several times.
+ func (b *B) Loop() bool {
++	b.SaveMeasurement()
+ 	b.StopTimerWithoutMarker()
+ 	// This is written such that the fast path is as fast as possible and can be
+ 	// inlined.

testing/testing/benchmark.go

@@ -93,6 +93,10 @@ type codspeed struct {
 	startTimestamp  uint64
 	startTimestamps []uint64
 	stopTimestamps  []uint64
+
+	// Indicates whether a measurement has been saved already. This aims to prevent saving measurements
+	// twice, because `b.Loop()` saves them internally as well but is also called from runN
+	savedMeasurement bool
 }
 
 // B is a type passed to [Benchmark] functions to manage benchmark
@@ -160,6 +164,7 @@ func (b *B) StartTimerWithoutMarker() {
 		// b.startBytes = memStats.TotalAlloc
 		b.start = highPrecisionTimeNow()
 		b.timerOn = true
+		b.savedMeasurement = false
 		// b.loop.i &^= loopPoisonTimer
 	}
 }
@@ -186,14 +191,22 @@ func (b *B) StopTimerWithoutMarker() {
 		b.timerOn = false
 		// If we hit B.Loop with the timer stopped, fail.
 		// b.loop.i |= loopPoisonTimer
+	}
+}
 
-		// For b.N loops: This will be called in runN which sets b.N to the number of iterations.
-		// For b.Loop() loops: loopSlowPath sets b.N to 0 to prevent b.N loops within b.Loop. However, since
-		// we're starting/stopping the timer for each iteration in the b.Loop() loop, we can use 1 as
-		// the number of iterations for this round.
-		b.codspeedItersPerRound = append(b.codspeedItersPerRound, max(int64(b.N), 1))
-		b.codspeedTimePerRoundNs = append(b.codspeedTimePerRoundNs, timeSinceStart)
+func (b *B) SaveMeasurement() {
+	if b.savedMeasurement {
+		return
 	}
+	b.savedMeasurement = true
+
+	// For b.N loops: This will be called in runN which sets b.N to the number of iterations.
+	// For b.Loop() loops: loopSlowPath sets b.N to 0 to prevent b.N loops within b.Loop. However, since
+	// we're starting/stopping the timer for each iteration in the b.Loop() loop, we can use 1 as
+	// the number of iterations for this round.
+	timeSinceStart := highPrecisionTimeSince(b.start)
+	b.codspeedItersPerRound = append(b.codspeedItersPerRound, max(int64(b.N), 1))
+	b.codspeedTimePerRoundNs = append(b.codspeedTimePerRoundNs, timeSinceStart)
 }
 
 func (b *B) StopTimer() {
@@ -296,6 +309,7 @@ func (b *B) __codspeed_root_frame__runN(n int) {
 	b.ResetTimer()
 	b.StartTimer()
 	b.benchFunc(b)
+	b.SaveMeasurement()
 	b.StopTimer()
 	b.previousN = n
 	b.previousDuration = b.duration
@@ -614,6 +628,7 @@ func (b *B) loopSlowPath() bool {
 // whereas b.N-based benchmarks must run the benchmark function (and any
 // associated setup and cleanup) several times.
 func (b *B) Loop() bool {
+	b.SaveMeasurement()
 	b.StopTimerWithoutMarker()
 	// This is written such that the fast path is as fast as possible and can be
 	// inlined.