Replace nop regex loops with empty (#118079)

When loop bodies end up containing zero-width assertions and the loop has a min bound of 0, the whole loop can be removed, as the zero-width assertion may or may not apply.

Author: stephentoub

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexNode.cs

@@ -894,7 +894,9 @@ private RegexNode ReduceLoops()
             // If the Loop or Lazyloop now only has one child node and its a Set, One, or Notone,
             // reduce to just Setloop/lazy, Oneloop/lazy, or Notoneloop/lazy.  The parser will
             // generally have only produced the latter, but other reductions could have exposed
-            // this.
+            // this. We can also reduce or eliminate certain loops that are nops, e.g.
+            // a loop with a minimum of 0 that wraps a zero-width assertion is either asserting something
+            // or not, and is thus useless.
             if (u.ChildCount() == 1)
             {
                 RegexNode child = u.Child(0);
@@ -906,6 +908,17 @@ private RegexNode ReduceLoops()
                         child.MakeRep(u.Kind == RegexNodeKind.Lazyloop ? RegexNodeKind.Onelazy : RegexNodeKind.Oneloop, u.M, u.N);
                         u = child;
                         break;
+
+                    case RegexNodeKind.Empty:
+                    case RegexNodeKind.PositiveLookaround or RegexNodeKind.NegativeLookaround or
+                         RegexNodeKind.Beginning or RegexNodeKind.Start or
+                         RegexNodeKind.Bol or RegexNodeKind.Eol or
+                         RegexNodeKind.End or RegexNodeKind.EndZ or
+                         RegexNodeKind.Boundary or RegexNodeKind.ECMABoundary or
+                         RegexNodeKind.NonBoundary or RegexNodeKind.NonECMABoundary
+                         when u.M == 0:
+                        u = new RegexNode(RegexNodeKind.Empty, Options);
+                        break;
                 }
             }
 

src/libraries/System.Text.RegularExpressions/tests/UnitTests/RegexFindOptimizationsTests.cs

@@ -43,9 +43,8 @@ public class RegexFindOptimizationsTests
         [InlineData(@"(?=\b)^abc", 0, (int)FindNextStartingPositionMode.LeadingAnchor_LeftToRight_Beginning)]
         [InlineData(@"(?=\b)(?=^.*$)abc", 0, (int)FindNextStartingPositionMode.LeadingAnchor_LeftToRight_Beginning)]
         [InlineData(@"(?=\b)(?=\B)^abc", 0, (int)FindNextStartingPositionMode.LeadingAnchor_LeftToRight_Beginning)]
-        // The next two could be improved slightly to be LeadingString_LeftToRight.
-        [InlineData(@"(?=^.*def)?abc", 0, (int)FindNextStartingPositionMode.FixedDistanceChar_LeftToRight)] 
-        [InlineData(@"(?=^)?(?=^)abc", 0, (int)FindNextStartingPositionMode.FixedDistanceChar_LeftToRight)]
+        [InlineData(@"(?=^.*def)abc", 0, (int)FindNextStartingPositionMode.LeadingAnchor_LeftToRight_Beginning)] 
+        [InlineData(@"(?=^)(?=^)abc", 0, (int)FindNextStartingPositionMode.LeadingAnchor_LeftToRight_Beginning)]
 
         [InlineData(@"^", (int)RegexOptions.RightToLeft, (int)FindNextStartingPositionMode.LeadingAnchor_RightToLeft_Beginning)]
         [InlineData(@"hello^", (int)RegexOptions.RightToLeft, (int)FindNextStartingPositionMode.LeadingAnchor_RightToLeft_Beginning)]

src/libraries/System.Text.RegularExpressions/tests/UnitTests/RegexReductionTests.cs

@@ -244,6 +244,19 @@ public class RegexReductionTests
         // Large loop patterns
         [InlineData("a*a*a*a*a*a*a*b*b*?a+a*", "a*b*b*?a+")]
         [InlineData("a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?a?aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "a{0,30}aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")]
+        // Nop loops
+        [InlineData("(?:)*", "")]
+        [InlineData("a(?=abc)*b", "ab")]
+        [InlineData("a(?<=abc)*b", "ab")]
+        [InlineData("a(?<!abc)*b", "ab")]
+        [InlineData("a$*b", "ab")]
+        [InlineData("a^?b", "ab")]
+        [InlineData(@"a\b*b", "ab")]
+        [InlineData(@"a\B*b", "ab")]
+        [InlineData(@"a\z?b", "ab")]
+        [InlineData(@"a\Z?b", "ab")]
+        [InlineData(@"a\A?b", "ab")]
+        [InlineData(@"a\G?b", "ab")]
         // Group elimination
         [InlineData("(?:(?:(?:(?:(?:(?:a*))))))", "a*")]
         // Nested loops
@@ -542,6 +555,11 @@ public void PatternsReduceIdentically(string actual, string expected)
         [InlineData("a*(?(xyz)bcd)", "(?>a*)(?(xyz)bcd)")]
         // Different prefixes on alternation branches
         [InlineData("^abcd|$abce", "^abcd|^abce")]
+        // Zero-width assertions in non-removable loops
+        [InlineData("a(?=abc)+b", "ab")]
+        [InlineData("a(?<=abc)+b", "ab")]
+        [InlineData("a(?<!abc){1,2}b", "ab")]
+        [InlineData("a${3,}b", "ab")]
         // Anchors
         [InlineData(@"\b\B", "\b")]
         [InlineData(@"^$", "^")]