bidiorder.java

处理PDF

        //
        // Although the scan proceeds left to right, and changes the type values
        // in a way that would appear to affect the computations later in the scan,
        // there is actually no problem.  A change in the current value can only
        // affect the value to its immediate right, and only affect it if it is
        // ES or CS.  But the current value can only change if the value to its
        // right is not ES or CS.  Thus either the current value will not change,
        // or its change will have no effect on the remainder of the analysis.
        
        for (int i = start + 1; i < limit - 1; ++i) {
            if (resultTypes[i] == ES || resultTypes[i] == CS) {
                byte prevSepType = resultTypes[i-1];
                byte succSepType = resultTypes[i+1];
                if (prevSepType == EN && succSepType == EN) {
                    resultTypes[i] = EN;
                } else if (resultTypes[i] == CS && prevSepType == AN && succSepType == AN) {
                    resultTypes[i] = AN;
                }
            }
        }
        
        // Rule W5.
        for (int i = start; i < limit; ++i) {
            if (resultTypes[i] == ET) {
                // locate end of sequence
                int runstart = i;
                int runlimit = findRunLimit(runstart, limit, new byte[] { ET });
                
                // check values at ends of sequence
                byte t = runstart == start ? sor : resultTypes[runstart - 1];
                
                if (t != EN) {
                    t = runlimit == limit ? eor : resultTypes[runlimit];
                }
                
                if (t == EN) {
                    setTypes(runstart, runlimit, EN);
                }
                
                // continue at end of sequence
                i = runlimit;
            }
        }
        
        // Rule W6.
        for (int i = start; i < limit; ++i) {
            byte t = resultTypes[i];
            if (t == ES || t == ET || t == CS) {
                resultTypes[i] = ON;
            }
        }
        
        // Rule W7.
        for (int i = start; i < limit; ++i) {
            if (resultTypes[i] == EN) {
                // set default if we reach start of run
                byte prevStrongType = sor;
                for (int j = i - 1; j >= start; --j) {
                    byte t = resultTypes[j];
                    if (t == L || t == R) { // AL's have been removed
                        prevStrongType = t;
                        break;
                    }
                }
                if (prevStrongType == L) {
                    resultTypes[i] = L;
                }
            }
        }
    }
    
    /**
     * 6) resolving neutral types
     * Rules N1-N2.
     */
    private void resolveNeutralTypes(int start, int limit, byte level, byte sor, byte eor) {
        
        for (int i = start; i < limit; ++i) {
            byte t = resultTypes[i];
            if (t == WS || t == ON || t == B || t == S) {
                // find bounds of run of neutrals
                int runstart = i;
                int runlimit = findRunLimit(runstart, limit, new byte[] {B, S, WS, ON});
                
                // determine effective types at ends of run
                byte leadingType;
                byte trailingType;
                
                if (runstart == start) {
                    leadingType = sor;
                } else {
                    leadingType = resultTypes[runstart - 1];
                    if (leadingType == L || leadingType == R) {
                        // found the strong type
                    } else if (leadingType == AN) {
                        leadingType = R;
                    } else if (leadingType == EN) {
                        // Since EN's with previous strong L types have been changed
                        // to L in W7, the leadingType must be R.
                        leadingType = R;
                    }
                }
                
                if (runlimit == limit) {
                    trailingType = eor;
                } else {
                    trailingType = resultTypes[runlimit];
                    if (trailingType == L || trailingType == R) {
                        // found the strong type
                    } else if (trailingType == AN) {
                        trailingType = R;
                    } else if (trailingType == EN) {
                        trailingType = R;
                    }
                }
                
                byte resolvedType;
                if (leadingType == trailingType) {
                    // Rule N1.
                    resolvedType = leadingType;
                } else {
                    // Rule N2.
                    // Notice the embedding level of the run is used, not
                    // the paragraph embedding level.
                    resolvedType = typeForLevel(level);
                }
                
                setTypes(runstart, runlimit, resolvedType);
                
                // skip over run of (former) neutrals
                i = runlimit;
            }
        }
    }
    
    /**
     * 7) resolving implicit embedding levels
     * Rules I1, I2.
     */
    private void resolveImplicitLevels(int start, int limit, byte level, byte sor, byte eor) {
        if ((level & 1) == 0) { // even level
            for (int i = start; i < limit; ++i) {
                byte t = resultTypes[i];
                // Rule I1.
                if (t == L ) {
                    // no change
                } else if (t == R) {
                    resultLevels[i] += 1;
                } else { // t == AN || t == EN
                    resultLevels[i] += 2;
                }
            }
        } else { // odd level
            for (int i = start; i < limit; ++i) {
                byte t = resultTypes[i];
                // Rule I2.
                if (t == R) {
                    // no change
                } else { // t == L || t == AN || t == EN
                    resultLevels[i] += 1;
                }
            }
        }
    }
    
    //
    // Output
    //
    
    public byte[] getLevels() {
        return getLevels(new int[]{textLength});
    }
    
    /**
     * Return levels array breaking lines at offsets in linebreaks. <br>
     * Rule L1.
     * <p>
     * The returned levels array contains the resolved level for each
     * bidi code passed to the constructor.
     * <p>
     * The linebreaks array must include at least one value.
     * The values must be in strictly increasing order (no duplicates)
     * between 1 and the length of the text, inclusive.  The last value
     * must be the length of the text.
     *
     * @param linebreaks the offsets at which to break the paragraph
     * @return the resolved levels of the text
     */
    public byte[] getLevels(int[] linebreaks) {
        
        // Note that since the previous processing has removed all
        // P, S, and WS values from resultTypes, the values referred to
        // in these rules are the initial types, before any processing
        // has been applied (including processing of overrides).
        //
        // This example implementation has reinserted explicit format codes
        // and BN, in order that the levels array correspond to the
        // initial text.  Their final placement is not normative.
        // These codes are treated like WS in this implementation,
        // so they don't interrupt sequences of WS.
        
        validateLineBreaks(linebreaks, textLength);
        
        byte[] result = (byte[])resultLevels.clone(); // will be returned to caller
        
        // don't worry about linebreaks since if there is a break within
        // a series of WS values preceding S, the linebreak itself
        // causes the reset.
        for (int i = 0; i < result.length; ++i) {
            byte t = initialTypes[i];
            if (t == B || t == S) {
                // Rule L1, clauses one and two.
                result[i] = paragraphEmbeddingLevel;
                
                // Rule L1, clause three.
                for (int j = i - 1; j >= 0; --j) {
                    if (isWhitespace(initialTypes[j])) { // including format codes
                        result[j] = paragraphEmbeddingLevel;
                    } else {
                        break;
                    }
                }
            }
        }
        
        // Rule L1, clause four.
        int start = 0;
        for (int i = 0; i < linebreaks.length; ++i) {
            int limit = linebreaks[i];
            for (int j = limit - 1; j >= start; --j) {
                if (isWhitespace(initialTypes[j])) { // including format codes
                    result[j] = paragraphEmbeddingLevel;
                } else {
                    break;
                }
            }
            
            start = limit;
        }
        
        return result;
    }
    
    /**
     * Return reordering array breaking lines at offsets in linebreaks.
     * <p>
     * The reordering array maps from a visual index to a logical index.
     * Lines are concatenated from left to right.  So for example, the
     * fifth character from the left on the third line is
     * <pre> getReordering(linebreaks)[linebreaks[1] + 4]</pre>
     * (linebreaks[1] is the position after the last character of the
     * second line, which is also the index of the first character on the
     * third line, and adding four gets the fifth character from the left).
     * <p>
     * The linebreaks array must include at least one value.
     * The values must be in strictly increasing order (no duplicates)
     * between 1 and the length of the text, inclusive.  The last value
     * must be the length of the text.
     *
     * @param linebreaks the offsets at which to break the paragraph.
     */
    public int[] getReordering(int[] linebreaks) {
        validateLineBreaks(linebreaks, textLength);
        
        byte[] levels = getLevels(linebreaks);
        
        return computeMultilineReordering(levels, linebreaks);
    }
    
    /**
     * Return multiline reordering array for a given level array.
     * Reordering does not occur across a line break.
     */
    private static int[] computeMultilineReordering(byte[] levels, int[] linebreaks) {
        int[] result = new int[levels.length];
        
        int start = 0;
        for (int i = 0; i < linebreaks.length; ++i) {
            int limit = linebreaks[i];
            
            byte[] templevels = new byte[limit - start];
            System.arraycopy(levels, start, templevels, 0, templevels.length);
            
            int[] temporder = computeReordering(templevels);
            for (int j = 0; j < temporder.length; ++j) {
                result[start + j] = temporder[j] + start;
            }
            
            start = limit;
        }
        
        return result;
    }
    
    /**
     * Return reordering array for a given level array.  This reorders a single line.
     * The reordering is a visual to logical map.  For example,
     * the leftmost char is string.charAt(order[0]).
     * Rule L2.
     */
    private static int[] computeReordering(byte[] levels) {
        int lineLength = levels.length;
        
        int[] result = new int[lineLength];
        
        // initialize order
        for (int i = 0; i < lineLength; ++i) {
            result[i] = i;
        }
        
        // locate highest level found on line.
        // Note the rules say text, but no reordering across line bounds is performed,
        // so this is sufficient.
        byte highestLevel = 0;
        byte lowestOddLevel = 63;
        for (int i = 0; i < lineLength; ++i) {
            byte level = levels[i];
            if (level > highestLevel) {
                highestLevel = level;
            }
            if (((level & 1) != 0) && level < lowestOddLevel) {
                lowestOddLevel = level;
            }
        }