Add documentationComment for Class to Kernel. Parse it. Resynthesize in Analyzer.

runtime/vm/kernel_binary_flowgraph.h

Index: runtime/vm/kernel_binary_flowgraph.h
diff --git a/runtime/vm/kernel_binary_flowgraph.h b/runtime/vm/kernel_binary_flowgraph.h
index f3d899be177a888837cdb0c3138af5dbbd92fd0e..e67047f6a9d538565bde302bb587ba8e4b3c76d7 100644
--- a/runtime/vm/kernel_binary_flowgraph.h
+++ b/runtime/vm/kernel_binary_flowgraph.h
@@ -1212,6 +1212,7 @@ class ClassHelper {
     kIsAbstract,
     kNameIndex,
     kSourceUriIndex,
+    kDocumentationCommentIndex,
     kAnnotations,
     kTypeParameters,
     kSuperClass,
@@ -1263,6 +1264,9 @@ class ClassHelper {
         builder_->current_script_id_ = source_uri_index_;
         builder_->record_token_position(position_);
         if (++next_read_ == field) return;
+      case kDocumentationCommentIndex:
+        documentation_comment_index_ = builder_->ReadStringReference();

[Paul Berry, 2017/07/14 22:03:08]

Would it be better for the VM to just skip documentation comments, rather than
store them in a field that is never referenced?  They don't affect semantics so
it seems silly to bloat the VM's internal code representation with them.

[Siggi Cherem (dart-lang), 2017/07/14 22:12:43]

I share your concerns too, I like the idea of just skipping the bits.

[scheglov, 2017/07/15 00:41:57]

I agree, we can just skip it.

+        if (++next_read_ == field) return;
       case kAnnotations: {
         annotation_count_ = builder_->ReadListLength();  // read list length.
         for (intptr_t i = 0; i < annotation_count_; ++i) {
@@ -1336,6 +1340,7 @@ class ClassHelper {
   bool is_abstract_;
   StringIndex name_index_;
   intptr_t source_uri_index_;
+  StringIndex documentation_comment_index_;
   intptr_t annotation_count_;
 
  private: