Work around race condition in MakeArray for ASSERTs and document assumptions.

runtime/vm/raw_object.cc

Index: runtime/vm/raw_object.cc
===================================================================
--- runtime/vm/raw_object.cc	(revision 43879)
+++ runtime/vm/raw_object.cc	(working copy)
@@ -216,6 +216,17 @@
 #if defined(DEBUG)
   uword tags = ptr()->tags_;
   intptr_t tags_size = SizeTag::decode(tags);
+  if ((instance_size > tags_size && tags_size > 0) && (class_id == kArrayCid)) {

[Ivan Posva, 2015/02/19 22:43:56]

"if ((class_id == kArrayCid) && ...) {"

is a bit more readable to me as it tells me that we are only interested in the
actual size relation if we are looking at an array.

[koda, 2015/02/19 22:48:49]

Agreed.

Slightly less efficient, but OK since this is DEBUG mode.

+    // TODO(22501): Array::MakeArray could be in the process of shrinking
+    // the array (see comment therein). Wait a millisecond and try again.
+    int retries_remaining = 1000;  // ... but not forever.
+    do {
+      OS::Sleep(1);
+      const RawArray* raw_array = reinterpret_cast<const RawArray*>(this);
+      intptr_t array_length = Smi::Value(raw_array->ptr()->length_);
+      instance_size = Array::InstanceSize(array_length);

[Ivan Posva, 2015/02/19 22:43:56]

Might need a comment why you are only reloading the instance_size, but not the
tag size.

[koda, 2015/02/19 22:48:49]

Done.

+    } while ((instance_size > tags_size) && (--retries_remaining > 0));
+  }
   if ((instance_size != tags_size) && (tags_size != 0)) {
     FATAL3("Size mismatch: %" Pd " from class vs %" Pd " from tags %" Px "\n",
            instance_size, tags_size, tags);