[sql] Consider fresh databases suitable for memory-mapped I/O.

sync/syncable/directory_backing_store.cc

Index: sync/syncable/directory_backing_store.cc
diff --git a/sync/syncable/directory_backing_store.cc b/sync/syncable/directory_backing_store.cc
index 55c99d38264ac5be42a54fbfd7dcf369da6fb74e..aad79bb257e068a127a748beab558605ff83c341 100644
--- a/sync/syncable/directory_backing_store.cc
+++ b/sync/syncable/directory_backing_store.cc
@@ -1728,6 +1728,10 @@ void DirectoryBackingStore::ResetAndCreateConnection() {
   db_->set_cache_size(32);
   db_->set_page_size(database_page_size_);
 
+  // TODO(shess): The current mitigation for http://crbug.com/537742 stores
+  // state in the meta table, which this database does not use.
+  db_->set_mmap_disabled();
+
   // TODO(shess): Sync corruption tests interact poorly with mmap, disable for
   // now.  http://crbug.com/533682
   db_->set_mmap_disabled();

[rmcilroy, 2015/12/17 10:03:22]

You've already disabled it here - just add the TODO and avoid calling
mmap_disabled twice?

[Scott Hess - ex-Googler, 2015/12/18 00:00:32]

Derp.  There's another CL in progress which was going to enable it, and I was
just copy/paste programming...