SSLClientSessionCache: Log number of times Lookup is called per Session.

net/ssl/ssl_client_session_cache.h

Index: net/ssl/ssl_client_session_cache.h
diff --git a/net/ssl/ssl_client_session_cache.h b/net/ssl/ssl_client_session_cache.h
index 4a82010badda3722a6d318ff74951016a155bfc2..20b2d899f9fa5b355dfacf6ee21e388e1eb4d01c 100644
--- a/net/ssl/ssl_client_session_cache.h
+++ b/net/ssl/ssl_client_session_cache.h
@@ -64,6 +64,14 @@ class NET_EXPORT SSLClientSessionCache : public base::MemoryCoordinatorClient {
   void DumpMemoryStats(base::trace_event::ProcessMemoryDump* pmd);
 
  private:
+  struct Entry {
+    size_t lookup_count;

[davidben, 2017/01/11 16:47:50]

If you just write size_t lookup_count = 0 here, I think you can avoid the
out-of-line constructor and destructor? Or do the clang plugins complain?

[nharper, 2017/01/11 22:31:18]

The chromium-style clang plugin complains that this struct with just two members
is complex, thus it needs an explicit out-of-line c'tor/d'tor.

+    bssl::UniquePtr<SSL_SESSION> session;
+
+    Entry();
+    ~Entry();
+  };
+
   // base::MemoryCoordinatorClient implementation:
   void OnMemoryStateChange(base::MemoryState state) override;
 
@@ -79,7 +87,7 @@ class NET_EXPORT SSLClientSessionCache : public base::MemoryCoordinatorClient {
 
   std::unique_ptr<base::Clock> clock_;
   Config config_;
-  base::HashingMRUCache<std::string, bssl::UniquePtr<SSL_SESSION>> cache_;
+  base::HashingMRUCache<std::string, std::unique_ptr<Entry>> cache_;

[davidben, 2017/01/11 16:47:50]

Does it work to do

  base::HashingMRUCache<std::string, Entry> cache_?

Saves a bunch of needless mallocs. I think the compiler will just generate a
move constructor if you don't explicitly define Entry and ~Entry.

[nharper, 2017/01/11 22:31:18]

I'd prefer to it to be base::HashingMRUCache<std::string, Entry>, but clang was
complaining about it when it was like that. For what it's worth, this used to be
a base::HashingMRUCache<std::string, std::unique_ptr<CacheEntry>> (where
CacheEntry had a bssl::UniquePtr<SSL_SESSION> member), so it's no more
indirection/mallocs than that. I'll try poking at it again, since I don't see a
reason why putting Entry in directly shouldn't work.

   size_t lookups_since_flush_;
 
   // TODO(davidben): After https://crbug.com/458365 is fixed, replace this with