Add one-char string table for faster String.charAt to the VM.

runtime/vm/symbols.cc

Index: runtime/vm/symbols.cc
===================================================================
--- runtime/vm/symbols.cc	(revision 14902)
+++ runtime/vm/symbols.cc	(working copy)
@@ -28,8 +28,8 @@
 };
 
 
-const char* Symbols::Name(intptr_t symbol) {
-  ASSERT((symbol > kIllegal) && (symbol < kMaxId));
+const char* Symbols::Name(SymbolId symbol) {

[Ivan Posva, 2012/11/15 09:07:10]

I find it unfortunate that we cannot access the name over the full range of
valid SymbolIds.

[Florian Schneider, 2012/11/15 21:54:42]

Yes, but not sure if the names are actually needed for those >= kMaxPredefined.

+  ASSERT((symbol > kIllegal) && (symbol < kMaxPredefinedId));

[Ivan Posva, 2012/11/15 09:07:10]

I find it unfortunate that we cannot access the names over the full SymbolId
range.

   return names[symbol];
 }
 
@@ -44,17 +44,22 @@
   // Turn off population of symbols in the VM symbol table, so that we
   // don't find these symbols while doing a Symbols::New(...).
   // Create all predefined symbols.
-  ASSERT((sizeof(names) / sizeof(const char*)) == Symbols::kMaxId);
+  ASSERT((sizeof(names) / sizeof(const char*)) == Symbols::kMaxPredefinedId);
   const Array& symbol_table =
       Array::Handle(isolate->object_store()->symbol_table());
   dart::String& str = String::Handle();
 
-  for (intptr_t i = 1; i < Symbols::kMaxId; i++) {
+  for (intptr_t i = 1; i < Symbols::kMaxPredefinedId; i++) {

[Ivan Posva, 2012/11/15 09:07:10]

if the single char strings would be part of the names table, then you would not
need to have special handling here.

     str = OneByteString::New(names[i], Heap::kOld);
     Add(symbol_table, str);

[Ivan Posva, 2012/11/15 09:07:10]

The naked Add into the symbol table seems rather suspicious here. I know this is
not in your code, but we should verify that the symbol table never grows as we
keep it cached throughout this loop.

Fixed with https://chromiumcodereview.appspot.com/11308032/

     predefined_[i] = str.raw();
   }
   Object::RegisterSingletonClassNames();
+
+  for (uint32_t c = 0; c <= kMaxOneByteCharCode; c++) {
+    ASSERT(kMaxPredefinedId + c < kMaxId);
+    predefined_[kMaxPredefinedId + c] = New(&c, 1);
+  }
 }
 
 
@@ -118,7 +123,6 @@
 template<typename T>
 RawString* Symbols::New(const T* characters, intptr_t len) {
   Isolate* isolate = Isolate::Current();
-  ASSERT(isolate != Dart::vm_isolate());

[Ivan Posva, 2012/11/15 09:07:10]

If single char was handled in names, you would not need to remove this assertion
either.

[Florian Schneider, 2012/11/15 21:54:42]

It is a bit unfortunate, but I also want to avoid duplicates between the
predefined symbols and all one-char strings. It should not hurt to invoke
Symbols::New inside the vm-isolate, except that the code below will try twice to
find any non-existing string before adding it.

   String& symbol = String::Handle(isolate, String::null());
   Array& symbol_table = Array::Handle(isolate, Array::null());
 
@@ -201,6 +205,14 @@
 }
 
 
+RawString* Symbols::FromCharCode(uint32_t char_code) {
+  if (char_code > kMaxOneByteCharCode) {
+    return New(&char_code, 1);
+  }
+  return predefined_[kMaxPredefinedId + char_code];
+}
+
+
 void Symbols::GrowSymbolTable(const Array& symbol_table) {
   // TODO(iposva): Avoid exponential growth.
   intptr_t table_size = symbol_table.Length() - 1;