<%
'為了支持原創,請保留該處注釋,謝謝!
'作者:草上飛
'獲取主域名
Function getDomainUrl(url)
tempurl=replace(url,"http://","")
if instr(tempurl,"/")>0 then
tempurl=left(tempurl,instr(tempurl,"/")-1)
end If
getDomainurl=tempurl
End Function
Function GetHttpPage(HttpUrl)
If IsNull(HttpUrl)=True Or Len(HttpUrl)<18 Or HttpUrl="$False$" Then
GetHttpPage="$False$"
Exit Function
End If
Dim Http
Set Http=server.createobject("MSXML2.XMLHTTP")
Http.open "GET",HttpUrl,False
Http.Send()
If Http.Readystate<>4 then
Set Http=Nothing
GetHttpPage="$False$"
Exit function
End if
GetHTTPPage=Http.responseText
Set Http=Nothing
If Err.number<>0 then
Err.Clear
End If
End Function
'==================================================
'函數名:ScriptHtml
'作 用:過濾html標記
'參 數:ConStr ------ 要過濾的字符串
' TagName ------要過濾的標簽
' FType 1表示過濾左邊標簽 2表示過濾左右標簽及中間的值 3表示過濾左邊標簽和右邊標簽,保留內容。
'==================================================
Function ScriptHtml(Byval ConStr,TagName,FType,includestr)
Dim Re
Set Re=new RegExp
Re.IgnoreCase =true
Re.Global=True
Select Case FType
Case 1
Re.Pattern="<" & TagName & "([^>])*("&includestr&"){1,}([^>])*>"
ConStr=Re.Replace(ConStr,"")
Case 2
Re.Pattern="<" & TagName & "([^>])*("&includestr&"){1,}([^>])*>.*?</" & TagName & "([^>])*>"
'response.write constr&"<br>"
ConStr=Re.Replace(ConStr,"")
'response.write server.htmlencode(constr)&"<br>"
Case 3
Re.Pattern="<" & TagName & "([^>])*("&includestr&"){1,}([^>])*>"
ConStr=Re.Replace(ConStr,"")
Re.Pattern="</" & TagName & "([^>])*>"
ConStr=Re.Replace(ConStr,"")
End Select
ScriptHtml=ConStr
Set Re=Nothing
End Function
'==================================================
'函數名:GetBody
'作 用:截取字符串
'參 數:ConStr ------將要截取的字符串
'參 數:StartStr ------開始字符串
'參 數:OverStr ------結束字符串
'參 數:IncluL ------是否包含StartStr
'參 數:IncluR ------是否包含OverStr
'==================================================
Function GetBody(ConStr,StartStr,OverStr,IncluL,IncluR)
If ConStr="$False$" or ConStr="" or IsNull(ConStr)=True Or StartStr="" or IsNull(StartStr)=True Or OverStr="" or IsNull(OverStr)=True Then
GetBody="$False$"
Exit Function
End If
Dim ConStrTemp
Dim Start,Over
ConStrTemp=Lcase(ConStr)
StartStr=Lcase(StartStr)
OverStr=Lcase(OverStr)
Start = InStrB(1, ConStrTemp, StartStr, vbBinaryCompare)
'response.write Start&"<br>"&IncluL&"<br>"
'response.end
If Start<=0 then
GetBody="$False$"
Exit Function
Else
If IncluL=False Then
Start=Start+LenB(StartStr)
End If
End If
Over=InStrB(Start,ConStrTemp,OverStr,vbBinaryCompare)
'response.write Over
'response.end
'response.write Start&" "&Over&" "&Over-Start
'response.end
If Over<=0 Or Over<=Start then
GetBody="$False$"
Exit Function
Else
If IncluR=True Then
Over=Over+LenB(OverStr)
End If
End If
GetBody=MidB(ConStr,Start,Over-Start)
'response.write getBody
'response.end
End Function
'==================================================
'函數名:GetArray
'作 用:提取鏈接地址,以$Array$分隔
'參 數:ConStr ------提取地址的原字符
'參 數:StartStr ------開始字符串
'參 數:OverStr ------結束字符串
'參 數:IncluL ------是否包含StartStr
'參 數:IncluR ------是否包含OverStr
'==================================================
Function GetArray(Byval ConStr,StartStr,OverStr,IncluL,IncluR)
If ConStr="$False$" or ConStr="" Or IsNull(ConStr)=True or StartStr="" Or OverStr="" or IsNull(StartStr)=True Or IsNull(OverStr)=True Then
GetArray="$False$"
Exit Function
End If
Dim TempStr,TempStr2,objRegExp,Matches,Match
TempStr=""
Set objRegExp = New Regexp
objRegExp.IgnoreCase = True
objRegExp.Global = True
objRegExp.Pattern = "("&StartStr&").+?("&OverStr&")"
Set Matches =objRegExp.Execute(ConStr)
For Each Match in Matches
TempStr=TempStr & "$Array$" & Match.Value
Next
Set Matches=nothing
If TempStr="" Then
GetArray="$False$"
Exit Function
End If
TempStr=Right(TempStr,Len(TempStr)-7)
If IncluL=False then
objRegExp.Pattern =StartStr
TempStr=objRegExp.Replace(TempStr,"")
End if
If IncluR=False then
objRegExp.Pattern =OverStr
TempStr=objRegExp.Replace(TempStr,"")
End if
Set objRegExp=nothing
Set Matches=nothing
If TempStr="" then
GetArray="$False$"
Else
GetArray=TempStr
End if
End Function
Function getAlexaRank(weburl)
tempurl=getDomainUrl(weburl)
'讀取http://client.alexa.com/common/css/scramble.css中的數據
alexacss="http://client.alexa.com/common/css/scramble.css"
strAlexaCss=GetHttpPage(alexacss)
'response.write strAlexaCss
'response.end
alexarankqueryurl="http://www.alexa.com/data/details/traffic_details/"&tempurl
strAlexaContent=GetHttpPage(alexarankqueryurl)
rankcontent=getBody(strAlexaContent,"Information Service.-->","<!-- google_ad_section_end(name=default) -->",false,false)
'獲取其中的span的class
strspan=GetArray(rankcontent,"<span class=""","""",false,false)
'response.write rankcontent&"<br>"
'response.write strspan&"<br>"
'response.end
If strspan<>"$False$" Then
aspan=split(strspan,"$Array$")
For i=0 To UBound(aspan)
'response.write "."&aspan(i)
'判定aspan(i)即span的class是否在alexacss中存在,如果存在,則需要將這個span和span中的數據去掉。
If InStr(strAlexaCss,"."&aspan(i))>=1 Then
'response.write aspan(i)&"<br>"
'response.end
'表示屬性為none.需要替換掉。
rankcontent=ScriptHtml(rankcontent,"span",2,aspan(i))
Else
rankcontent=ScriptHtml(rankcontent,"span",1,aspan(i))
End if
Next
'替換上面少去掉的右邊的span標簽。
rankcontent=Replace(rankcontent,"</span>","")
End If
If rankcontent="$False$" Then
rankcontent="No Data"
End if
getAlexaRank=Replace(rankcontent,",","")
End Function
url=request.querystring("url")
%>
<form name="alexaform" method=get>
輸入網址:<input type="" name="url" value="<%=url%>" size=40> <input type="submit" value="查 詢">
</form>
<%
If url<>"" Then
response.write "您的網站在ALEXA的排名為:"
response.flush
rank=getAlexaRank(url)
response.write rank
End if
%>