From c2168d89a8b835b2482fbbb20033ae3469f60392 Mon Sep 17 00:00:00 2001
From: Jille Timmermans <jille@hexon.nl>
Date: Fri, 13 Sep 2024 09:41:44 +0200
Subject: [PATCH] Fallback code: Always zero memory beforehand

This makes the API contract stricter.
---
 README.md     |  1 +
 purego.go     | 11 +++++++++++
 typeswitch.go | 10 +++++-----
 3 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/README.md b/README.md
index 14e8757..c7bc212 100644
--- a/README.md
+++ b/README.md
@@ -10,6 +10,7 @@ The easiest way to explain is to just show the fallback code.
 
 ```go
 func goVectorEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+  clear(dst)
   for i, v := range rows {
     if search == v {
       dst[i/8] |= 1 << (i % 8)
diff --git a/purego.go b/purego.go
index 8ec3759..0b1981b 100644
--- a/purego.go
+++ b/purego.go
@@ -1,6 +1,7 @@
 package vectorcmp
 
 func goVectorEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+	zero(dst)
 	for i, v := range rows {
 		if search == v {
 			dst[i/8] |= 1 << (i % 8)
@@ -9,6 +10,7 @@ func goVectorEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, ro
 }
 
 func goVectorGreaterThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+	zero(dst)
 	for i, v := range rows {
 		if search > v {
 			dst[i/8] |= 1 << (i % 8)
@@ -17,6 +19,7 @@ func goVectorGreaterThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search
 }
 
 func goVectorLessThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+	zero(dst)
 	for i, v := range rows {
 		if search < v {
 			dst[i/8] |= 1 << (i % 8)
@@ -25,6 +28,7 @@ func goVectorLessThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search T,
 }
 
 func goVectorGreaterEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+	zero(dst)
 	for i, v := range rows {
 		if search >= v {
 			dst[i/8] |= 1 << (i % 8)
@@ -33,9 +37,16 @@ func goVectorGreaterEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, searc
 }
 
 func goVectorLesserEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows []T) {
+	zero(dst)
 	for i, v := range rows {
 		if search <= v {
 			dst[i/8] |= 1 << (i % 8)
 		}
 	}
 }
+
+func zero(dst []byte) {
+	for i := range dst {
+		dst[i] = 0
+	}
+}
diff --git a/typeswitch.go b/typeswitch.go
index ad35008..3e4fce2 100644
--- a/typeswitch.go
+++ b/typeswitch.go
@@ -1,9 +1,9 @@
 package vectorcmp
 
 // VectorEquals compares each entry in rows to search and enables the corresponding bit in dst if equal.
-// dst must be zeroed before this call.
 // Prefer calling VectorEquals8/16/32/64 directly.
 //
+//	clear(dst)
 //	for i, v := range rows {
 //	  if search == v {
 //	    dst[i/8] |= 1 << (i % 8)
@@ -25,9 +25,9 @@ func VectorEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, rows
 }
 
 // VectorGreaterThan compares each entry in rows to search and enables the corresponding bit in dst if search > rows[i].
-// dst must be zeroed before this call.
 // Prefer calling VectorGreaterThan8/16/32/64 directly.
 //
+//	clear(dst)
 //	for i, v := range rows {
 //	  if search > v {
 //	    dst[i/8] |= 1 << (i % 8)
@@ -49,9 +49,9 @@ func VectorGreaterThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search T,
 }
 
 // VectorLessThan compares each entry in rows to search and enables the corresponding bit in dst if search < rows[i].
-// dst must be zeroed before this call.
 // Prefer calling VectorLessThan8/16/32/64 directly.
 //
+//	clear(dst)
 //	for i, v := range rows {
 //	  if search < v {
 //	    dst[i/8] |= 1 << (i % 8)
@@ -73,9 +73,9 @@ func VectorLessThan[T uint8 | uint16 | uint32 | uint64](dst []byte, search T, ro
 }
 
 // VectorGreaterEquals compares each entry in rows to search and enables the corresponding bit in dst if search >= rows[i].
-// dst must be zeroed before this call.
 // Prefer calling VectorGreaterEquals8/16/32/64 directly.
 //
+//	clear(dst)
 //	for i, v := range rows {
 //	  if search >= v {
 //	    dst[i/8] |= 1 << (i % 8)
@@ -97,9 +97,9 @@ func VectorGreaterEquals[T uint8 | uint16 | uint32 | uint64](dst []byte, search
 }
 
 // VectorLesserEquals compares each entry in rows to search and enables the corresponding bit in dst if search <= rows[i].
-// dst must be zeroed before this call.
 // Prefer calling VectorLesserEquals8/16/32/64 directly.
 //
+//	clear(dst)
 //	for i, v := range rows {
 //	  if search <= v {
 //	    dst[i/8] |= 1 << (i % 8)