michael@0: /*
michael@0:  **********************************************************************
michael@0:  *   Copyright (C) 2005-2012, International Business Machines
michael@0:  *   Corporation and others.  All Rights Reserved.
michael@0:  **********************************************************************
michael@0:  */
michael@0: 
michael@0: #include "unicode/utypes.h"
michael@0: 
michael@0: #if !UCONFIG_NO_CONVERSION
michael@0: 
michael@0: #include "csrutf8.h"
michael@0: #include "csmatch.h"
michael@0: 
michael@0: U_NAMESPACE_BEGIN
michael@0: 
michael@0: CharsetRecog_UTF8::~CharsetRecog_UTF8()
michael@0: {
michael@0:     // nothing to do
michael@0: }
michael@0: 
michael@0: const char *CharsetRecog_UTF8::getName() const
michael@0: {
michael@0:     return "UTF-8";
michael@0: }
michael@0: 
michael@0: UBool CharsetRecog_UTF8::match(InputText* input, CharsetMatch *results) const {
michael@0:     bool hasBOM = FALSE;
michael@0:     int32_t numValid = 0;
michael@0:     int32_t numInvalid = 0;
michael@0:     const uint8_t *inputBytes = input->fRawInput;
michael@0:     int32_t i;
michael@0:     int32_t trailBytes = 0;
michael@0:     int32_t confidence;
michael@0: 
michael@0:     if (input->fRawLength >= 3 && 
michael@0:         inputBytes[0] == 0xEF && inputBytes[1] == 0xBB && inputBytes[2] == 0xBF) {
michael@0:             hasBOM = TRUE;
michael@0:     }
michael@0: 
michael@0:     // Scan for multi-byte sequences
michael@0:     for (i=0; i < input->fRawLength; i += 1) {
michael@0:         int32_t b = inputBytes[i];
michael@0: 
michael@0:         if ((b & 0x80) == 0) {
michael@0:             continue;   // ASCII
michael@0:         }
michael@0: 
michael@0:         // Hi bit on char found.  Figure out how long the sequence should be
michael@0:         if ((b & 0x0E0) == 0x0C0) {
michael@0:             trailBytes = 1;
michael@0:         } else if ((b & 0x0F0) == 0x0E0) {
michael@0:             trailBytes = 2;
michael@0:         } else if ((b & 0x0F8) == 0xF0) {
michael@0:             trailBytes = 3;
michael@0:         } else {
michael@0:             numInvalid += 1;
michael@0: 
michael@0:             if (numInvalid > 5) {
michael@0:                 break;
michael@0:             }
michael@0: 
michael@0:             trailBytes = 0;
michael@0:         }
michael@0: 
michael@0:         // Verify that we've got the right number of trail bytes in the sequence
michael@0:         for (;;) {
michael@0:             i += 1;
michael@0: 
michael@0:             if (i >= input->fRawLength) {
michael@0:                 break;
michael@0:             }
michael@0: 
michael@0:             b = inputBytes[i];
michael@0: 
michael@0:             if ((b & 0xC0) != 0x080) {
michael@0:                 numInvalid += 1;
michael@0:                 break;
michael@0:             }
michael@0: 
michael@0:             if (--trailBytes == 0) {
michael@0:                 numValid += 1;
michael@0:                 break;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:     }
michael@0: 
michael@0:     // Cook up some sort of confidence score, based on presense of a BOM
michael@0:     //    and the existence of valid and/or invalid multi-byte sequences.
michael@0:     confidence = 0;
michael@0:     if (hasBOM && numInvalid == 0) {
michael@0:         confidence = 100;
michael@0:     } else if (hasBOM && numValid > numInvalid*10) {
michael@0:         confidence = 80;
michael@0:     } else if (numValid > 3 && numInvalid == 0) {
michael@0:         confidence = 100;
michael@0:     } else if (numValid > 0 && numInvalid == 0) {
michael@0:         confidence = 80;
michael@0:     } else if (numValid == 0 && numInvalid == 0) {
michael@0:         // Plain ASCII.
michael@0:         confidence = 10;
michael@0:     } else if (numValid > numInvalid*10) {
michael@0:         // Probably corruput utf-8 data.  Valid sequences aren't likely by chance.
michael@0:         confidence = 25;
michael@0:     }
michael@0: 
michael@0:     results->set(input, this, confidence);
michael@0:     return (confidence > 0);
michael@0: }
michael@0: 
michael@0: U_NAMESPACE_END
michael@0: #endif