Regex: support look-behind patterns

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

test/Regex.java

@@ -60,5 +60,6 @@ public class Regex {
     expectGroups("a|(b|c)", "a", (String)null);
     expectGroups("a|(b|c)", "c", "c");
     expectGroups("(?=a)a", "a");
+    expectGroups(".*(o)(?<=[A-Z][a-z]*)", "Hello", "o");
   }
 }

test/regex/Compiler.java

@@ -28,7 +28,7 @@ class Compiler implements PikeVMOpcodes {
     private int groupCount = -1;
     private int findPreambleSize;
     private ArrayList<CharacterMatcher> classes;
-    private ArrayList<PikeVM> lookaheads;
+    private ArrayList<PikeVM> lookarounds;
 
     public Output(Expression expr) {
       // try-run to determine the code size
@@ -37,7 +37,7 @@ class Compiler implements PikeVMOpcodes {
       offset = 0;
       groupCount = -1;
       classes = new ArrayList<CharacterMatcher>();
-      lookaheads = new ArrayList<PikeVM>();
+      lookarounds = new ArrayList<PikeVM>();
       // write it out!
       expr.writeCode(this);
     }
@@ -66,10 +66,10 @@ class Compiler implements PikeVMOpcodes {
     public PikeVM toVM() {
       CharacterMatcher[] classes = new CharacterMatcher[this.classes.size()];
       this.classes.toArray(classes);
-      PikeVM[] lookaheads = new PikeVM[this.lookaheads.size()];
+      PikeVM[] lookarounds = new PikeVM[this.lookarounds.size()];
-      this.lookaheads.toArray(lookaheads);
+      this.lookarounds.toArray(lookarounds);
       return new PikeVM(program, findPreambleSize, groupCount, classes,
-        lookaheads);
+        lookarounds);
     }
 
     public int addClass(CharacterMatcher characterClass) {
@@ -81,12 +81,12 @@ class Compiler implements PikeVMOpcodes {
       return result;
     }
 
-    public int addLookahead(PikeVM lookahead) {
+    public int addLookaround(PikeVM lookaround) {
       if (program == null) {
         return -1;
       }
-      int result = lookaheads.size();
+      int result = lookarounds.size();
-      lookaheads.add(lookahead);
+      lookarounds.add(lookaround);
       return result;
     }
   }
@@ -226,14 +226,22 @@ class Compiler implements PikeVMOpcodes {
     }
   }
 
-  private class Lookahead extends Expression {
+  private class Lookaround extends Expression {
     private final Group group = new Group(false, null);
+    private final boolean forward;
+
+    public Lookaround(boolean forward) {
+      this.forward = forward;
+    }
 
     @Override
     protected void writeCode(Output output) {
       PikeVM vm = new Output(group).toVM();
-      output.add(LOOKAHEAD);
+      if (!forward) {
-      output.add(output.addLookahead(vm));
+        vm.reverse();
+      }
+      output.add(forward ? LOOKAHEAD : LOOKBEHIND);
+      output.add(output.addLookaround(vm));
     }
   }
 
@@ -302,15 +310,28 @@ class Compiler implements PikeVMOpcodes {
               + regex);
           }
           c = array[index];
+          boolean lookAhead = true;
+          if (c == '<') {
+            if (++ index >= array.length) {
+              throw new RuntimeException("Short pattern @" + index + ": "
+                + regex);
+            }
+            lookAhead = false;
+            c = array[index];
+            if (c != '=' && c != '!') {
+              throw new IllegalArgumentException("Named groups not supported @"
+                + index + ": " + regex);
+            }
+          }
           switch (c) {
           case ':':
             capturing = false;
             break;
           case '=': {
             capturing = false;
-            Lookahead lookahead = new Lookahead();
+            Lookaround lookaround = new Lookaround(lookAhead);
-            current.push(lookahead);
+            current.push(lookaround);
-            groups.push(lookahead.group);
+            groups.push(lookaround.group);
             continue;
           }
           default:

test/regex/PikeVM.java

@@ -27,21 +27,21 @@ class PikeVM implements PikeVMOpcodes {
    */
   private final int findPrefixLength;
   private final CharacterMatcher[] classes;
-  private final PikeVM[] lookaheads;
+  private final PikeVM[] lookarounds;
 
   public interface Result {
     void set(int[] start, int[] end);
   }
 
   protected PikeVM(int[] program, int findPrefixLength, int groupCount,
-    CharacterMatcher[] classes, PikeVM[] lookaheads)
+    CharacterMatcher[] classes, PikeVM[] lookarounds)
   {
     this.program = program;
     this.findPrefixLength = findPrefixLength;
     this.groupCount = groupCount;
     offsetsCount = 2 * groupCount + 2;
     this.classes = classes;
-    this.lookaheads = lookaheads;
+    this.lookarounds = lookarounds;
   }
 
   /**
@@ -293,13 +293,14 @@ class PikeVM implements PikeVMOpcodes {
     ThreadQueue queued = new ThreadQueue(startPC);
 
     boolean foundMatch = false;
-    for (int i = start; i <= end; ++i) {
+    int step = end > start ? +1 : -1;
+    for (int i = start; i != end + step; i += step) {
       if (queued.isEmpty()) {
         // no threads left
         return foundMatch;
       }
 
-      char c = i < end ? characters[i] : 0;
+      char c = i != end ? characters[i] : 0;
       int pc = -1;
       for (;;) {
         pc = current.next(pc);
@@ -312,7 +313,7 @@ class PikeVM implements PikeVMOpcodes {
 
         // pc == program.length is a match!
         if (pc == program.length) {
-          if (anchorEnd && i < end) {
+          if (anchorEnd && i != end) {
             continue;
           }
           if (result == null) {
@@ -346,11 +347,17 @@ class PikeVM implements PikeVMOpcodes {
           }
           break;
         case LOOKAHEAD:
-          if (lookaheads[program[pc + 1]].matches(characters,
+          if (lookarounds[program[pc + 1]].matches(characters,
               i, characters.length, true, false, null)) {
             current.queueImmediately(pc, pc + 2, false);
           }
           break;
+        case LOOKBEHIND:
+          if (lookarounds[program[pc + 1]].matches(characters,
+              i - 1, -1, true, false, null)) {
+            current.queueImmediately(pc, pc + 2, false);
+          }
+          break;
         /* immediate opcodes, i.e. thread continues within the same step */
         case SAVE_OFFSET:
           if (result != null) {

test/regex/PikeVMOpcodes.java

@@ -25,6 +25,7 @@ interface PikeVMOpcodes {
   final static int CHARACTER_CLASS = -20;
 
   final static int LOOKAHEAD = -30;
+  final static int LOOKBEHIND = -31;
 
   final static int SAVE_OFFSET = -40;