Prefer binary format for arrays

This improves performance decoding text[].

pgtype/array_codec.go

@@ -47,7 +47,16 @@ func (c *ArrayCodec) FormatSupported(format int16) bool {
 }
 
 func (c *ArrayCodec) PreferredFormat() int16 {
-	return c.ElementType.Codec.PreferredFormat()
+	// The binary format should always be preferred for arrays if it is supported. Usually, this will happen automatically
+	// because most types that support binary prefer it. However, text, json, and jsonb support binary but prefer the text
+	// format. This is because it is simpler for jsonb and PostgreSQL can be significantly faster using the text format
+	// for text-like data types than binary. However, arrays appear to always be faster in binary.
+	//
+	// https://www.postgresql.org/message-id/CAMovtNoHFod2jMAKQjjxv209PCTJx5Kc66anwWvX0mEiaXwgmA%40mail.gmail.com
+	if c.ElementType.Codec.FormatSupported(BinaryFormatCode) {
+		return BinaryFormatCode
+	}
+	return TextFormatCode
 }
 
 func (c *ArrayCodec) PlanEncode(m *Map, oid uint32, format int16, value any) EncodePlan {

pgtype/array_codec_test.go

@@ -342,3 +342,11 @@ func TestArrayCodecDecodeTextArrayWithTextOfNULL(t *testing.T) {
 		}
 	})
 }
+
+func TestArrayCodecDecodeTextArrayPrefersBinaryFormat(t *testing.T) {
+	defaultConnTestRunner.RunTest(context.Background(), t, func(ctx context.Context, t testing.TB, conn *pgx.Conn) {
+		sd, err := conn.Prepare(ctx, "", `select '{"foo", "NULL", " NULL "}'::text[]`)
+		require.NoError(t, err)
+		require.Equal(t, int16(1), conn.TypeMap().FormatCodeForOID(sd.Fields[0].DataTypeOID))
+	})
+}