Regex compiler: fall back to TrivialPattern when possible

While at it, let's get rid of the unescaping in TrivialPattern which was
buggy anyway: special operators such as \b were misinterpreted as trivial
patterns.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

test/regex/Compiler.java

@@ -166,6 +166,11 @@ class Compiler implements PikeVMOpcodes {
       throw new IllegalArgumentException("Unclosed groups: ("
         + (groups.size() - 1) + "): " + regex);
     }
-    return new RegexPattern(regex, 0, new Output(root).toVM());
+    PikeVM vm = new Output(root).toVM();
+    String plain = vm.isPlainString();
+    if (plain != null) {
+      return new TrivialPattern(regex, plain, 0);
+    }
+    return new RegexPattern(regex, 0, vm);
   }
 }

test/regex/Pattern.java

@@ -43,9 +43,6 @@ public abstract class Pattern implements PikeVMOpcodes {
   }
 
   public static Pattern compile(String regex, int flags) {
-    try {
-      return new TrivialPattern(regex, flags);
-    } catch (UnsupportedOperationException handledBelow) { }
     if (flags != 0) {
       throw new UnsupportedOperationException("TODO");
     }

test/regex/PikeVM.java

@@ -332,4 +332,41 @@ class PikeVM implements PikeVMOpcodes {
     }
     return foundMatch;
   }
+
+  /**
+   * Determines whether this machine recognizes a pattern without special
+   * operators.
+   * <p>
+   * In case that the regular expression is actually a plain string without any
+   * special operators, we can avoid using a full-blown Pike VM and instead fall
+   * back to using the much faster {@link TrivialPattern}.
+   * </p>
+   * 
+   * @return the string to match, or null if the machine recognizes a
+   *         non-trivial pattern
+   */
+  public String isPlainString() {
+    // we expect the machine to start with SAVE_OFFSET 0 and
+    // end with SAVE_OFFSET 1
+    int start = 0;
+    if (start + 1 < program.length &&
+        program[start] == SAVE_OFFSET && program[start + 1] == 0) {
+      start += 2;
+    }
+    int end = program.length;
+    if (end > start + 1 &&
+        program[end - 2] == SAVE_OFFSET && program[end - 1] == 1) {
+      end -= 2;
+    }
+    for (int i = start; i < end; ++ i) {
+      if (program[i] < 0) {
+        return null;
+      }
+    }
+    char[] array = new char[end - start];
+    for (int i = start; i < end; ++ i) {
+      array[i - start] = (char)program[i];
+    }
+    return new String(array);
+  }
 }

test/regex/TrivialPattern.java

@@ -22,102 +22,15 @@ import java.util.LinkedList;
  */
 public class TrivialPattern extends Pattern {
 
-  private final String trivialPattern;
+  private final String unescaped;
 
-  TrivialPattern(String pattern, int flags) {
+  TrivialPattern(String pattern, String unescaped, int flags) {
     super(pattern, flags);
-    this.trivialPattern = trivial(pattern);
-  }
-
-  private static String trivial(String pattern) {
-    StringBuffer buffer = new StringBuffer();
-    for (int i = 0; i < pattern.length(); ++i) {
-      char c = pattern.charAt(i);
-      switch (c) {
-      case '\\':
-        if (++i == pattern.length()) {
-          break;
-        }
-        c = pattern.charAt(i);
-        if (c == '0') {
-          int len = digits(pattern, ++i, 3, 8);
-          if (len == 3 && pattern.charAt(i) > '3') {
-            --len;
-          }
-          c = (char)Integer.parseInt(pattern.substring(i, i + len), 8);
-          i += len - 1;
-        } else if (c == 'x' || c == 'u') {
-          int len = digits(pattern, ++i, 4, 16);
-          c = (char)Integer.parseInt(pattern.substring(i, i + len), 16);
-          i += len - 1;
-        } else {
-          c = unescape(pattern.charAt(i));
-        }
-        if (c != -1) {
-          break;
-        }
-        // fallthru
-      case '.':
-      case '*':
-      case '+':
-      case '?':
-      case '|':
-      case '[':
-      case ']':
-      case '{':
-      case '}':
-      case '(':
-      case ')':
-      case '^':
-      case '$':
-        throw new UnsupportedOperationException
-          ("only trivial regular expressions are supported so far (" + pattern + ")");
-      }
-      buffer.append(c);
-    }
-    return buffer.toString();
-  }
-
-  private static int digits(String s, int offset, int maxLength, int base) {
-    for (int i = 0; ; ++i) {
-      if (i == maxLength || offset + i >= s.length()) {
-        return i;
-      }
-      int value = s.charAt(offset + i) - '0';
-      if (value < 0) {
-        return i;
-      }
-      if (base > 10 && value >= 10) {
-        value += 10 - (value >= 'a' - '0' ? 'a' - '0' : 'A' - '0');
-      }
-      if (value >= base) {
-        return i;
-      }
-    }
-  }
-
-  private static char unescape(char c) {
-    switch (c) {
-    case '\\':
-       return c;
-    case 'a':
-       return 0x0007;
-    case 'e':
-       return 0x001B;
-    case 'f':
-       return 0x000C;
-    case 'n':
-       return 0x000A;
-    case 'r':
-       return 0x000D;
-    case 't':
-       return 0x0009;
-    }
-    return (char)-1;
+    this.unescaped = unescaped;
   }
 
   public Matcher matcher(CharSequence input) {
-    return new TrivialMatcher(trivialPattern, input);
+    return new TrivialMatcher(unescaped, input);
   }
 
   public String[] split(CharSequence input, int limit) {
@@ -135,7 +48,7 @@ public class TrivialPattern extends Pattern {
     List<CharSequence> list = new LinkedList<CharSequence>();
     int index = 0;
     int trailing = 0;
-    int patternLength = trivialPattern.length();
+    int patternLength = unescaped.length();
     while (index < input.length() && list.size() < limit - 1) {
       int i;
       if (patternLength == 0) {
@@ -145,7 +58,7 @@ public class TrivialPattern extends Pattern {
           i = index + 1;
         }
       } else {
-        i = indexOf(input, trivialPattern, index);
+        i = indexOf(input, unescaped, index);
       }
 
       if (i >= 0) {