(1)缘起
bfax@smth.org发了一个字符串转换程序,引起了热烈讨论。原程序如下:
1
2Function B2G(prestr As String) As String
3 Dim i, j As Integer
4 Const GB_Lib = "" //几千个字符吧,因为字符串长度限制,原程序是由GB_lib1,GB_lib2GB_lib4四个字符串构成的,为了简化问题,只用一个字符串代替。
5 Const BIG5_Lib = "" //与GB_Lib中简体字一一对应的繁体字
6
7 For i = 1 To prestr.Length
8 j= Instr(1, BIG5_Lib1, GetChar(prestr, i))
9 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib1,j),GetChar(GB_Lib1,j))
10 j= Instr(1, BIG5_Lib2, GetChar(prestr, i))
11 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib2,j),GetChar(GB_Lib2,j))
12 j= Instr(1, BIG5_Lib3, GetChar(prestr, i))
13 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib3,j),GetChar(GB_Lib3,j))
14 j= Instr(1, BIG5_Lib4, GetChar(prestr, i))
15 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib4,j),GetChar(GB_Lib4,j))
16 Next
17 Return prestr
18End Function
2Function B2G(prestr As String) As String
3 Dim i, j As Integer
4 Const GB_Lib = "" //几千个字符吧,因为字符串长度限制,原程序是由GB_lib1,GB_lib2GB_lib4四个字符串构成的,为了简化问题,只用一个字符串代替。
5 Const BIG5_Lib = "" //与GB_Lib中简体字一一对应的繁体字
6
7 For i = 1 To prestr.Length
8 j= Instr(1, BIG5_Lib1, GetChar(prestr, i))
9 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib1,j),GetChar(GB_Lib1,j))
10 j= Instr(1, BIG5_Lib2, GetChar(prestr, i))
11 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib2,j),GetChar(GB_Lib2,j))
12 j= Instr(1, BIG5_Lib3, GetChar(prestr, i))
13 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib3,j),GetChar(GB_Lib3,j))
14 j= Instr(1, BIG5_Lib4, GetChar(prestr, i))
15 If j<>0 Then prestr=prestr.Replace(GetChar(BIG5_Lib4,j),GetChar(GB_Lib4,j))
16 Next
17 Return prestr
18End Function
(2) 分析问题
写测试程序测试在我的1.5M迅驰本本测试,替换效率为30万字/s,该程序采用Replace,这样对每一个字符都要扫描GB_Lib字符串中的几千个字符,性能自然上不去。需要寻找更好的数据结构和算法,降低每一个字符串的操作时间。
.net类库里有一个很好的东西可以拿来直接用:Hashtable。也就是说,把每一个简体字作为key,每一个繁体字作为value。这样处理每个字符的时候只需要看它在不在Hashtable的key里面,在的话就找出对应的value替换,否则就不做任何操作。这样做的代价是Hashtable初始化的耗时,不过初始化顶多也就一次嘛。程序如下:
1public class ConvertDemo
2{
3 private static Hashtable _libTable;
4
5 static ConvertDemo()
6 {
7 InitHashTable();
8 }
9
10 static string GB_lib="";
11
12 static string BIG5_lib="";
13
14 private static void InitHashTable()
15 {
16 _libTable = new Hashtable();
17 PushIntoHashtable(_libTable,GB_lib,BIG5_lib);
18 }
19
20 private static void PushIntoHashtable(Hashtable t, string g , string b)
21 {
22 for (int i=0;i<g.Length;i++)
23 {
24 t.Add(g[i],b[i]);
25 }
26 }
27
28 private static char ConvertChar(char input)
29 {
30 if (_libTable.ContainsKey(input)) return (char)_libTable[input];
31 else return input;
32 }
33
34 public static string ConvertText(string inputString)
35 {
36 StringBuilder sb = new StringBuilder(inputString);
37 for (int i=0;i<inputString.Length;i++)
38 {
39 sb[i] = ConvertChar(inputString[i]);
40 }
41 return sb.ToString();
42 }
43}
2{
3 private static Hashtable _libTable;
4
5 static ConvertDemo()
6 {
7 InitHashTable();
8 }
9
10 static string GB_lib="";
11
12 static string BIG5_lib="";
13
14 private static void InitHashTable()
15 {
16 _libTable = new Hashtable();
17 PushIntoHashtable(_libTable,GB_lib,BIG5_lib);
18 }
19
20 private static void PushIntoHashtable(Hashtable t, string g , string b)
21 {
22 for (int i=0;i<g.Length;i++)
23 {
24 t.Add(g[i],b[i]);
25 }
26 }
27
28 private static char ConvertChar(char input)
29 {
30 if (_libTable.ContainsKey(input)) return (char)_libTable[input];
31 else return input;
32 }
33
34 public static string ConvertText(string inputString)
35 {
36 StringBuilder sb = new StringBuilder(inputString);
37 for (int i=0;i<inputString.Length;i++)
38 {
39 sb[i] = ConvertChar(inputString[i]);
40 }
41 return sb.ToString();
42 }
43}
测试性能,结果为300万字/秒。性能提高了10倍。
(3)用relector看Hashtable源代码,消除无用操作,继续优化
还能不能继续优化呢?ConvertChar (char input)执行次数最多,是对性能最有影响的方法。用reflector反编译Hashtable的get_Item(object key)方法:
1public virtual object get_Item(object key)
2{
3 uint num1;
4 uint num2;
5 Hashtable.bucket bucket1;
6 if (key == null)
7 {
8 throw new ArgumentNullException("key", Environment.GetResourceString("ArgumentNull_Key"));
9 }
10 Hashtable.bucket[] bucketArray1 = this.buckets;
11 uint num3 = this.InitHash(key, bucketArray1.Length, out num1, out num2);
12 int num4 = 0;
13 int num5 = (int) (num1 % bucketArray1.Length);
14 do
15 {
16 bucket1 = bucketArray1[num5];
17 if (bucket1.key == null)
18 {
19 return null;
20 }
21 if (((bucket1.hash_coll & 0x7fffffff) == num3) && this.KeyEquals(key, bucket1.key))
22 {
23 return bucket1.val;
24 }
25 num5 = (int) ((num5 + num2) % ((ulong) bucketArray1.Length));
26 }
27 while ((bucket1.hash_coll < 0) && (++num4 < bucketArray1.Length));
28 return null;
29}
2{
3 uint num1;
4 uint num2;
5 Hashtable.bucket bucket1;
6 if (key == null)
7 {
8 throw new ArgumentNullException("key", Environment.GetResourceString("ArgumentNull_Key"));
9 }
10 Hashtable.bucket[] bucketArray1 = this.buckets;
11 uint num3 = this.InitHash(key, bucketArray1.Length, out num1, out num2);
12 int num4 = 0;
13 int num5 = (int) (num1 % bucketArray1.Length);
14 do
15 {
16 bucket1 = bucketArray1[num5];
17 if (bucket1.key == null)
18 {
19 return null;
20 }
21 if (((bucket1.hash_coll & 0x7fffffff) == num3) && this.KeyEquals(key, bucket1.key))
22 {
23 return bucket1.val;
24 }
25 num5 = (int) ((num5 + num2) % ((ulong) bucketArray1.Length));
26 }
27 while ((bucket1.hash_coll < 0) && (++num4 < bucketArray1.Length));
28 return null;
29}
我的天天天天天天天天天天天天天........好长呀,先不管这个。哦,方法并不抛出异常,如果key不存在就直接返回null。这样的话,采用ContainsKey(...)判断key是否存在就是多次一举了。
把ConvertChar (char input)改为:
1private static char ConvertChar(char input)
2{
3object temp = _libTable[input];
4return temp == null?input:(char)temp;
5}
2{
3object temp = _libTable[input];
4return temp == null?input:(char)temp;
5}
这样大概能节省一半的操作。
测试结果验证了我这一想法。性能一下提高了40%,达到了500万字/s
注:上面程序有小bug,后来发现的。