[regexp] Check capture_count before using experimental engine

Sometimes the parser throws away redundant parts of the AST while
parsing.  For example, the regexp /(?:(?=(f)o))?f/ is (almost)
equivalent to just /f/ because the optional block (...)? is zero-length.
The parser notices this and returns the same tree as for /f/.  However,
there is a capture inside the (...)? block (which is never recorded
because the quantifier containing it can only match zero-width, which is
considered failure), so in this case it doesn't suffice to check that
the regexp AST doesn't contain captures.

Cq-Include-Trybots: luci.v8.try:v8_linux64_fyi_rel_ng
Bug: v8:10765
Change-Id: I6145849d95b3522a397eadd2bae63d1d8e880f28
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2397896Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Martin Bidlingmaier <mbid@google.com>
Cr-Commit-Position: refs/heads/master@{#69733}

src/regexp/experimental/experimental-compiler.cc

@@ -18,8 +18,9 @@ constexpr uc32 kMaxSupportedCodepoint = 0xFFFFu;
 class CanBeHandledVisitor final : private RegExpVisitor {
   // Visitor to implement `ExperimentalRegExp::CanBeHandled`.
  public:
-  static bool Check(RegExpTree* node, JSRegExp::Flags flags, Zone* zone) {
-    if (!AreSuitableFlags(flags)) {
+  static bool Check(RegExpTree* node, JSRegExp::Flags flags, int capture_count,
+                    Zone* zone) {
+    if (!AreSuitableFlags(flags) || capture_count > 0) {
       return false;
     }
     CanBeHandledVisitor visitor(zone);
@@ -145,9 +146,9 @@ class CanBeHandledVisitor final : private RegExpVisitor {
 
 bool ExperimentalRegExpCompiler::CanBeHandled(RegExpTree* tree,
                                               JSRegExp::Flags flags,
-                                              Zone* zone) {
+                                              int capture_count, Zone* zone) {
   DCHECK(FLAG_enable_experimental_regexp_engine);
-  return CanBeHandledVisitor::Check(tree, flags, zone);
+  return CanBeHandledVisitor::Check(tree, flags, capture_count, zone);
 }
 
 namespace {

src/regexp/experimental/experimental-compiler.h

@@ -19,7 +19,8 @@ class ExperimentalRegExpCompiler final : public AllStatic {
   // but see the definition.
   // TODO(mbid,v8:10765): Currently more things are not handled, e.g. some
   // quantifiers and unicode.
-  static bool CanBeHandled(RegExpTree* tree, JSRegExp::Flags flags, Zone* zone);
+  static bool CanBeHandled(RegExpTree* tree, JSRegExp::Flags flags,
+                           int capture_count, Zone* zone);
   // Compile regexp into a bytecode program.  The regexp must be handlable by
   // the experimental engine; see`CanBeHandled`.  The program is returned as a
   // ZoneList backed by the same Zone that is used in the RegExpTree argument.

src/regexp/experimental/experimental.cc

@@ -14,8 +14,9 @@ namespace v8 {
 namespace internal {
 
 bool ExperimentalRegExp::CanBeHandled(RegExpTree* tree, JSRegExp::Flags flags,
-                                      Zone* zone) {
-  return ExperimentalRegExpCompiler::CanBeHandled(tree, flags, zone);
+                                      int capture_count, Zone* zone) {
+  return ExperimentalRegExpCompiler::CanBeHandled(tree, flags, capture_count,
+                                                  zone);
 }
 
 void ExperimentalRegExp::Initialize(Isolate* isolate, Handle<JSRegExp> re,

src/regexp/experimental/experimental.h

@@ -19,7 +19,8 @@ class ExperimentalRegExp final : public AllStatic {
   // TODO(mbid, v8:10765): This walks the RegExpTree, but it could also be
   // checked on the fly in the parser.  Not done currently because walking the
   // AST again is more flexible and less error prone (but less performant).
-  static bool CanBeHandled(RegExpTree* tree, JSRegExp::Flags flags, Zone* zone);
+  static bool CanBeHandled(RegExpTree* tree, JSRegExp::Flags flags,
+                           int capture_count, Zone* zone);
   static void Initialize(Isolate* isolate, Handle<JSRegExp> re,
                          Handle<String> pattern, JSRegExp::Flags flags,
                          int capture_count);

src/regexp/regexp.cc

@@ -175,7 +175,8 @@ MaybeHandle<Object> RegExp::Compile(Isolate* isolate, Handle<JSRegExp> re,
   bool has_been_compiled = false;
 
   if (FLAG_enable_experimental_regexp_engine &&
-      ExperimentalRegExp::CanBeHandled(parse_result.tree, flags, &zone)) {
+      ExperimentalRegExp::CanBeHandled(parse_result.tree, flags,
+                                       parse_result.capture_count, &zone)) {
     ExperimentalRegExp::Initialize(isolate, re, pattern, flags,
                                    parse_result.capture_count);
     has_been_compiled = true;