前几天在园子里面看到一篇讲StringBuilder性能的文章( https://www.cnblogs.com/wanghao72214/p/15571181.html )。文章里面给出了一个测试用例,比较StringBuilder.AppendJoin和String.Join的性能。根据该测试结果,“对于这个操作,这两种方法的速度很接近,但 StringBuilder.AppendJoin 使用的内存明显较少”。据此,该文言之凿凿地指出,应该使用“StringBuilder.AppendJoin 而不是 String.Join”。
事实果真如此吗?
搜索一下就知道,StringBuilder采用的是先预分配缓冲区,然后将要连接的字符串直接复制到缓冲区的做法。这个做法确实高效,避免了中间结果带来的时间消耗和内存占用。
那么,string.Join真的那么不堪吗?
首先看看string.Join的代码:
public static string Join(string separator, params string[] value) { if (value == null) { throw new ArgumentNullException("value"); } return string.Join(separator, value, 0, value.Length); } public unsafe static string Join(string separator, string[] value, int startIndex, int count) { if (value == null) { throw new ArgumentNullException("value"); } if (startIndex < 0) { throw new ArgumentOutOfRangeException("startIndex", Environment.GetResourceString("ArgumentOutOfRange_StartIndex")); } if (count < 0) { throw new ArgumentOutOfRangeException("count", Environment.GetResourceString("ArgumentOutOfRange_NegativeCount")); } if (startIndex > value.Length - count) { throw new ArgumentOutOfRangeException("startIndex", Environment.GetResourceString("ArgumentOutOfRange_IndexCountBuffer")); } if (separator == null) { separator = string.Empty; } if (count == 0) { return string.Empty; } int num = 0; int num2 = startIndex + count - 1; for (int i = startIndex; i <= num2; i++) { if (value[i] != null) { num += value[i].Length; } } num += (count - 1) * separator.Length; if (num < 0 || num + 1 < 0) { throw new OutOfMemoryException(); } if (num == 0) { return string.Empty; } string text = string.FastAllocateString(num); fixed (char* ptr = &text.m_firstChar) { UnSafeCharBuffer unSafeCharBuffer = new UnSafeCharBuffer(ptr, num); unSafeCharBuffer.AppendString(value[startIndex]); for (int j = startIndex + 1; j <= num2; j++) { unSafeCharBuffer.AppendString(separator); unSafeCharBuffer.AppendString(value[j]); } } return text; }
可以看到,string.Join的做法是先计算最终结果的大小,然后调用string.FastAllocateString分配空间,最后将数据直接复制到分配的缓冲区。很显然,这一过程和StringBuilder如出一辙。
但是测试结果毕竟摆在那里,那么问题在哪里呢?
看看该文的测试用例:
[Benchmark] public string UsingStringJoin() { var list = new List < string > { "A", "B", "C", "D", "E" }; var stringBuilder = new StringBuilder(); for (int i = 0; i < 10000; i++) { stringBuilder.Append(string.Join(' ', list)); } return stringBuilder.ToString(); } [Benchmark] public string UsingAppendJoin() { var list = new List < string > { "A", "B", "C", "D", "E" }; var stringBuilder = new StringBuilder(); for (int i = 0; i < 10000; i++) { stringBuilder.AppendJoin(' ', list); } return stringBuilder.ToString(); }
问题就在下面的一句:
stringBuilder.Append(string.Join(' ', list));
这句代码实际上是先用string.Join把list拼好,再调用stringBuilder.Append把string.Join的结果拼接起来。这样,string.Join分配一次内存,stringBuilder再分配一次内存,内存占用怎能不大?
当然,没码没真相,得拿编译后的IL说话。所以,根据这个用法写段测试代码:
private void BtnStartClick(object sender, EventArgs e) { string[] dummy = new string[] { "zfsdfsd", "sdfsdf" }; StringBuilder sb = new StringBuilder(); sb.Append(string.Join(",", dummy)); string s = sb.ToString(); Console.WriteLine(s); }
看看IL:
1 .method private hidebysig 2 instance void BtnStartClick ( 3 object sender, 4 class [mscorlib]System.EventArgs e 5 ) cil managed 6 { 7 // Header Size: 12 bytes 8 // Code Size: 65 (0x41) bytes 9 // LocalVarSig Token: 0x11000004 RID: 4 10 .maxstack 3 11 .locals init ( 12 [0] string[] dummy, 13 [1] class [mscorlib]System.Text.StringBuilder sb, 14 [2] string s, 15 [3] string[] CS$0$0000 16 ) 17 18 /* (34,3)-(34,4) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 19 /* 0x00000340 00 */ IL_0000: nop 20 /* (35,4)-(35,52) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 21 /* 0x00000341 18 */ IL_0001: ldc.i4.2 22 /* 0x00000342 8D1D000001 */ IL_0002: newarr [mscorlib]System.String 23 /* 0x00000347 0D */ IL_0007: stloc.3 24 /* 0x00000348 09 */ IL_0008: ldloc.3 25 /* 0x00000349 16 */ IL_0009: ldc.i4.0 26 /* 0x0000034A 7201000070 */ IL_000A: ldstr "zfsdfsd" 27 /* 0x0000034F A2 */ IL_000F: stelem.ref 28 /* 0x00000350 09 */ IL_0010: ldloc.3 29 /* 0x00000351 17 */ IL_0011: ldc.i4.1 30 /* 0x00000352 7211000070 */ IL_0012: ldstr "sdfsdf" 31 /* 0x00000357 A2 */ IL_0017: stelem.ref 32 /* 0x00000358 09 */ IL_0018: ldloc.3 33 /* 0x00000359 0A */ IL_0019: stloc.0 34 /* (37,4)-(37,41) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 35 /* 0x0000035A 731600000A */ IL_001A: newobj instance void [mscorlib]System.Text.StringBuilder::.ctor() 36 /* 0x0000035F 0B */ IL_001F: stloc.1 37 /* (38,4)-(38,38) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 38 /* 0x00000360 07 */ IL_0020: ldloc.1 39 /* 0x00000361 721F000070 */ IL_0021: ldstr "," 40 /* 0x00000366 06 */ IL_0026: ldloc.0 41 /* 0x00000367 281700000A */ IL_0027: call string [mscorlib]System.String::Join(string, string[]) 42 /* 0x0000036C 6F1800000A */ IL_002C: callvirt instance class [mscorlib]System.Text.StringBuilder [mscorlib]System.Text.StringBuilder::Append(string) 43 /* 0x00000371 26 */ IL_0031: pop 44 /* (40,4)-(40,27) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 45 /* 0x00000372 07 */ IL_0032: ldloc.1 46 /* 0x00000373 6F1900000A */ IL_0033: callvirt instance string [mscorlib]System.Object::ToString() 47 /* 0x00000378 0C */ IL_0038: stloc.2 48 /* (42,4)-(42,25) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 49 /* 0x00000379 08 */ IL_0039: ldloc.2 50 /* 0x0000037A 281A00000A */ IL_003A: call void [mscorlib]System.Console::WriteLine(string) 51 /* 0x0000037F 00 */ IL_003F: nop 52 /* (45,3)-(45,4) d:\Work_Private\IoT\ClientSimulator\MainForm.cs */ 53 /* 0x00000380 2A */ IL_0040: ret 54 } // end of method MainForm::BtnStartClick
从第41和42行可以清楚看到,代码先调用了String.Join,然后是StringBuilder.Append.
所以事情很清楚了。错误的代码得出了不符合本意的测试结果,根据这个结果得到的结论自然也是错误的。
实际上,根据MS的文档,“修改 StringBuilder 时,除非达到容量,否则对象不会为自己重新分配空间。 当达到容量时,将自动分配新的空间且容量翻倍。”,可以看出,在边界情况下,使用StringBuilder耗费的空间反而比Join要大。当然,考虑到内存对齐的因素,Join也会有部分内存浪费,但这实在是微不足道的。
那么,应该使用StringBuilder还是Join呢?
很简单,按照具体情况决定。如果要拼接的是现成的字符串数组,自然应该用Join。否则的话,还是用StringBuilder省事点。