001/*
002 * Copyright (C) 2008 The Guava Authors
003 *
004 * Licensed under the Apache License, Version 2.0 (the "License");
005 * you may not use this file except in compliance with the License.
006 * You may obtain a copy of the License at
007 *
008 * http://www.apache.org/licenses/LICENSE-2.0
009 *
010 * Unless required by applicable law or agreed to in writing, software
011 * distributed under the License is distributed on an "AS IS" BASIS,
012 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
013 * See the License for the specific language governing permissions and
014 * limitations under the License.
015 */
016
017package com.google.common.escape;
018
019import static com.google.common.base.Preconditions.checkNotNull;
020
021import com.google.common.annotations.Beta;
022import com.google.common.annotations.GwtCompatible;
023
024/**
025 * An {@link Escaper} that converts literal text into a format safe for
026 * inclusion in a particular context (such as an XML document). Typically (but
027 * not always), the inverse process of "unescaping" the text is performed
028 * automatically by the relevant parser.
029 *
030 * <p>For example, an XML escaper would convert the literal string {@code
031 * "Foo<Bar>"} into {@code "Foo&lt;Bar&gt;"} to prevent {@code "<Bar>"} from
032 * being confused with an XML tag. When the resulting XML document is parsed,
033 * the parser API will return this text as the original literal string {@code
034 * "Foo<Bar>"}.
035 *
036 * <p><b>Note:</b> This class is similar to {@link CharEscaper} but with one
037 * very important difference. A CharEscaper can only process Java
038 * <a href="http://en.wikipedia.org/wiki/UTF-16">UTF16</a> characters in
039 * isolation and may not cope when it encounters surrogate pairs. This class
040 * facilitates the correct escaping of all Unicode characters.
041 *
042 * <p>As there are important reasons, including potential security issues, to
043 * handle Unicode correctly if you are considering implementing a new escaper
044 * you should favor using UnicodeEscaper wherever possible.
045 *
046 * <p>A {@code UnicodeEscaper} instance is required to be stateless, and safe
047 * when used concurrently by multiple threads.
048 *
049 * <p>Popular escapers are defined as constants in classes like {@link
050 * com.google.common.html.HtmlEscapers} and {@link com.google.common.xml.XmlEscapers}.
051 * To create your own escapers extend this class and implement the {@link #escape(int)}
052 * method.
053 *
054 * @author David Beaumont
055 * @since 15.0
056 */
057@Beta
058@GwtCompatible
059public abstract class UnicodeEscaper extends Escaper {
060  /** The amount of padding (chars) to use when growing the escape buffer. */
061  private static final int DEST_PAD = 32;
062
063  /** Constructor for use by subclasses. */
064  protected UnicodeEscaper() {}
065
066  /**
067   * Returns the escaped form of the given Unicode code point, or {@code null}
068   * if this code point does not need to be escaped. When called as part of an
069   * escaping operation, the given code point is guaranteed to be in the range
070   * {@code 0 <= cp <= Character#MAX_CODE_POINT}.
071   *
072   * <p>If an empty array is returned, this effectively strips the input
073   * character from the resulting text.
074   *
075   * <p>If the character does not need to be escaped, this method should return
076   * {@code null}, rather than an array containing the character representation
077   * of the code point. This enables the escaping algorithm to perform more
078   * efficiently.
079   *
080   * <p>If the implementation of this method cannot correctly handle a
081   * particular code point then it should either throw an appropriate runtime
082   * exception or return a suitable replacement character. It must never
083   * silently discard invalid input as this may constitute a security risk.
084   *
085   * @param cp the Unicode code point to escape if necessary
086   * @return the replacement characters, or {@code null} if no escaping was
087   *     needed
088   */
089  protected abstract char[] escape(int cp);
090
091  /**
092   * Scans a sub-sequence of characters from a given {@link CharSequence},
093   * returning the index of the next character that requires escaping.
094   *
095   * <p><b>Note:</b> When implementing an escaper, it is a good idea to override
096   * this method for efficiency. The base class implementation determines
097   * successive Unicode code points and invokes {@link #escape(int)} for each of
098   * them. If the semantics of your escaper are such that code points in the
099   * supplementary range are either all escaped or all unescaped, this method
100   * can be implemented more efficiently using {@link CharSequence#charAt(int)}.
101   *
102   * <p>Note however that if your escaper does not escape characters in the
103   * supplementary range, you should either continue to validate the correctness
104   * of any surrogate characters encountered or provide a clear warning to users
105   * that your escaper does not validate its input.
106   *
107   * <p>See {@link com.google.common.net.PercentEscaper} for an example.
108   *
109   * @param csq a sequence of characters
110   * @param start the index of the first character to be scanned
111   * @param end the index immediately after the last character to be scanned
112   * @throws IllegalArgumentException if the scanned sub-sequence of {@code csq}
113   *     contains invalid surrogate pairs
114   */
115  protected int nextEscapeIndex(CharSequence csq, int start, int end) {
116    int index = start;
117    while (index < end) {
118      int cp = codePointAt(csq, index, end);
119      if (cp < 0 || escape(cp) != null) {
120        break;
121      }
122      index += Character.isSupplementaryCodePoint(cp) ? 2 : 1;
123    }
124    return index;
125  }
126
127  /**
128   * Returns the escaped form of a given literal string.
129   *
130   * <p>If you are escaping input in arbitrary successive chunks, then it is not
131   * generally safe to use this method. If an input string ends with an
132   * unmatched high surrogate character, then this method will throw
133   * {@link IllegalArgumentException}. You should ensure your input is valid <a
134   * href="http://en.wikipedia.org/wiki/UTF-16">UTF-16</a> before calling this
135   * method.
136   *
137   * <p><b>Note:</b> When implementing an escaper it is a good idea to override
138   * this method for efficiency by inlining the implementation of
139   * {@link #nextEscapeIndex(CharSequence, int, int)} directly. Doing this for
140   * {@link com.google.common.net.PercentEscaper} more than doubled the
141   * performance for unescaped strings (as measured by {@link
142   * CharEscapersBenchmark}).
143   *
144   * @param string the literal string to be escaped
145   * @return the escaped form of {@code string}
146   * @throws NullPointerException if {@code string} is null
147   * @throws IllegalArgumentException if invalid surrogate characters are
148   *         encountered
149   */
150  @Override
151  public String escape(String string) {
152    checkNotNull(string);
153    int end = string.length();
154    int index = nextEscapeIndex(string, 0, end);
155    return index == end ? string : escapeSlow(string, index);
156  }
157
158  /**
159   * Returns the escaped form of a given literal string, starting at the given
160   * index.  This method is called by the {@link #escape(String)} method when it
161   * discovers that escaping is required.  It is protected to allow subclasses
162   * to override the fastpath escaping function to inline their escaping test.
163   * See {@link CharEscaperBuilder} for an example usage.
164   *
165   * <p>This method is not reentrant and may only be invoked by the top level
166   * {@link #escape(String)} method.
167   *
168   * @param s the literal string to be escaped
169   * @param index the index to start escaping from
170   * @return the escaped form of {@code string}
171   * @throws NullPointerException if {@code string} is null
172   * @throws IllegalArgumentException if invalid surrogate characters are
173   *         encountered
174   */
175  protected final String escapeSlow(String s, int index) {
176    int end = s.length();
177
178    // Get a destination buffer and setup some loop variables.
179    char[] dest = Platform.charBufferFromThreadLocal();
180    int destIndex = 0;
181    int unescapedChunkStart = 0;
182
183    while (index < end) {
184      int cp = codePointAt(s, index, end);
185      if (cp < 0) {
186        throw new IllegalArgumentException(
187            "Trailing high surrogate at end of input");
188      }
189      // It is possible for this to return null because nextEscapeIndex() may
190      // (for performance reasons) yield some false positives but it must never
191      // give false negatives.
192      char[] escaped = escape(cp);
193      int nextIndex = index + (Character.isSupplementaryCodePoint(cp) ? 2 : 1);
194      if (escaped != null) {
195        int charsSkipped = index - unescapedChunkStart;
196
197        // This is the size needed to add the replacement, not the full
198        // size needed by the string.  We only regrow when we absolutely must.
199        int sizeNeeded = destIndex + charsSkipped + escaped.length;
200        if (dest.length < sizeNeeded) {
201          int destLength = sizeNeeded + (end - index) + DEST_PAD;
202          dest = growBuffer(dest, destIndex, destLength);
203        }
204        // If we have skipped any characters, we need to copy them now.
205        if (charsSkipped > 0) {
206          s.getChars(unescapedChunkStart, index, dest, destIndex);
207          destIndex += charsSkipped;
208        }
209        if (escaped.length > 0) {
210          System.arraycopy(escaped, 0, dest, destIndex, escaped.length);
211          destIndex += escaped.length;
212        }
213        // If we dealt with an escaped character, reset the unescaped range.
214        unescapedChunkStart = nextIndex;
215      }
216      index = nextEscapeIndex(s, nextIndex, end);
217    }
218
219    // Process trailing unescaped characters - no need to account for escaped
220    // length or padding the allocation.
221    int charsSkipped = end - unescapedChunkStart;
222    if (charsSkipped > 0) {
223      int endIndex = destIndex + charsSkipped;
224      if (dest.length < endIndex) {
225        dest = growBuffer(dest, destIndex, endIndex);
226      }
227      s.getChars(unescapedChunkStart, end, dest, destIndex);
228      destIndex = endIndex;
229    }
230    return new String(dest, 0, destIndex);
231  }
232
233  /**
234   * Returns the Unicode code point of the character at the given index.
235   *
236   * <p>Unlike {@link Character#codePointAt(CharSequence, int)} or
237   * {@link String#codePointAt(int)} this method will never fail silently when
238   * encountering an invalid surrogate pair.
239   *
240   * <p>The behaviour of this method is as follows:
241   * <ol>
242   * <li>If {@code index >= end}, {@link IndexOutOfBoundsException} is thrown.
243   * <li><b>If the character at the specified index is not a surrogate, it is
244   *     returned.</b>
245   * <li>If the first character was a high surrogate value, then an attempt is
246   *     made to read the next character.
247   *     <ol>
248   *     <li><b>If the end of the sequence was reached, the negated value of
249   *         the trailing high surrogate is returned.</b>
250   *     <li><b>If the next character was a valid low surrogate, the code point
251   *         value of the high/low surrogate pair is returned.</b>
252   *     <li>If the next character was not a low surrogate value, then
253   *         {@link IllegalArgumentException} is thrown.
254   *     </ol>
255   * <li>If the first character was a low surrogate value,
256   *     {@link IllegalArgumentException} is thrown.
257   * </ol>
258   *
259   * @param seq the sequence of characters from which to decode the code point
260   * @param index the index of the first character to decode
261   * @param end the index beyond the last valid character to decode
262   * @return the Unicode code point for the given index or the negated value of
263   *         the trailing high surrogate character at the end of the sequence
264   */
265  protected static int codePointAt(CharSequence seq, int index, int end) {
266    checkNotNull(seq);
267    if (index < end) {
268      char c1 = seq.charAt(index++);
269      if (c1 < Character.MIN_HIGH_SURROGATE ||
270          c1 > Character.MAX_LOW_SURROGATE) {
271        // Fast path (first test is probably all we need to do)
272        return c1;
273      } else if (c1 <= Character.MAX_HIGH_SURROGATE) {
274        // If the high surrogate was the last character, return its inverse
275        if (index == end) {
276          return -c1;
277        }
278        // Otherwise look for the low surrogate following it
279        char c2 = seq.charAt(index);
280        if (Character.isLowSurrogate(c2)) {
281          return Character.toCodePoint(c1, c2);
282        }
283        throw new IllegalArgumentException(
284            "Expected low surrogate but got char '" + c2 +
285            "' with value " + (int) c2 + " at index " + index +
286            " in '" + seq + "'");
287      } else {
288        throw new IllegalArgumentException(
289            "Unexpected low surrogate character '" + c1 +
290            "' with value " + (int) c1 + " at index " + (index - 1) +
291            " in '" + seq + "'");
292      }
293    }
294    throw new IndexOutOfBoundsException("Index exceeds specified range");
295  }
296
297  /**
298   * Helper method to grow the character buffer as needed, this only happens
299   * once in a while so it's ok if it's in a method call.  If the index passed
300   * in is 0 then no copying will be done.
301   */
302  private static char[] growBuffer(char[] dest, int index, int size) {
303    char[] copy = new char[size];
304    if (index > 0) {
305      System.arraycopy(dest, 0, copy, 0, index);
306    }
307    return copy;
308  }
309}