fixed "sometimes, to output invalid confidence value" bug.

src/cchardet/cchardet.pyx

@@ -33,11 +33,10 @@ def detect(char *msg):
         return csd_close(csd)
 
 def detect_with_confidence(char *msg):
-    # TODO: fix "sometimes, to output invalid confidence value" bug.
     cdef csd_t csd = csd_open()
     cdef int length = strlen(msg)
     cdef int result = csd_consider(csd, msg, length)
-    cdef float confidence
+    cdef float confidence = 0.0
     cdef const_char_ptr detected_charset
     # ref: charsetdetect.cpp
     if result == -1: # Error, signal with a negative number

src/ext/libcharsetdetect/mozilla/extensions/universalchardet/src/base/nsUniversalDetector.cpp

@@ -45,6 +45,8 @@
 #include "nsEscCharsetProber.h"
 #include "nsLatin1Prober.h"
 
+#include <iostream>
+
 nsUniversalDetector::nsUniversalDetector(PRUint32 aLanguageFilter)
 {
   mDone = PR_FALSE;
@@ -292,6 +294,8 @@ void nsUniversalDetector::DataEnd2(float *confidence)
   {
     mDone = PR_TRUE;
     Report(mDetectedCharset);
+    // no confidence
+    *confidence = (float)0.99;
     return;
   }
 
@@ -316,10 +320,11 @@ void nsUniversalDetector::DataEnd2(float *confidence)
         }
       }
       //do not report anything because we are not confident of it, that's in fact a negative answer
-      if (maxProberConfidence > MINIMUM_THRESHOLD)
+      if (maxProberConfidence > MINIMUM_THRESHOLD){
         Report(mCharSetProbers[maxProber]->GetCharSetName());
         *confidence = maxProberConfidence;
       }
+    }
     break;
   case eEscAscii:
     break;