Add Web Resources usage to chrome://tracing.

third_party/WebKit/Source/core/fetch/WebCacheMemoryDumpProvider.h

+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef WebCacheMemoryDumpProvider_h
+#define WebCacheMemoryDumpProvider_h
+
+#include "core/fetch/MemoryCache.h"
+#include "platform/heap/Handle.h"
+#include "public/platform/WebCommon.h"
+#include "public/platform/WebMemoryDumpProvider.h"
+
+namespace blink {
+
+class BLINK_PLATFORM_EXPORT WebCacheMemoryDumpProvider final : public WebMemoryDumpProvider {

[haraken, 2015/10/14 00:57:41]

It seems strange to put WebXXX in core/. WebXXX needs to be in platform/ or
web/.

Can we do something like this?

- Put WebCacheMemoryDumpProvider in web/.
- Make sure that the MemoryCache is initialized in WebKit::initialize().
- Set the pointer to the MemoryCache (core/) to the WebCacheMemoryDumpProvider
(web/) in WebKit::initialize() (web/).

[ssid, 2015/10/14 05:01:22]

I think it's the name only should be changed here. There is nothing really web
about this. It's just a wrapper around MemoryCache. Should I make it
MemoryCacheDumpProvider?

I placed it here since MemoryCache has to include this file to set cache. It
cannot include a file in web/. I could probably move it, but just wondering if
it's really better.

[haraken, 2015/10/14 05:09:44]

The WebCacheMemoryDumpProvider inherits from WebMemoryDumpProvider, which is in
public/platform/. Then it looks better to put the WebCacheMemoryDumpProvider in
platform/ (not in web/, sorry).

What's an issue of putting WebCacheMemoryDumpProvider in platform/?

[ssid, 2015/10/14 06:13:17]

Platform cannot use core. So I cannot use memory cache.
 But the point of the dump provider interface is to have blink classes implement
them where ever needed and take dumps. Like in previous case memory cache itself
was implementing it.

[haraken, 2015/10/14 08:15:44]

Inheriting public/platform/ from core/ sounds a bit nasty. However, the
dependency is technically allowed and tasak@ is saying it will be ok. (To clean
up the dependency, we need the blink componentization. Since we decided not to
proceed with the componentization because of its opportunity cost, we think it's
okay to allow the dependency at the moment.)

I'll take a final look shortly.

[haraken, 2015/10/14 09:04:07]

BLINK_PLATFORM_EXPORT => CORE_EXPORT

[ssid, 2015/10/14 12:06:40]

Done.

+public:
+    static WebCacheMemoryDumpProvider* instance();

[haraken, 2015/10/14 09:04:07]

Add a comment on why this needs to be a singleton.

[ssid, 2015/10/14 12:06:40]

Done.

+    ~WebCacheMemoryDumpProvider() override;
+
+    // WebMemoryDumpProvider implementation.
+    bool onMemoryDump(WebMemoryDumpLevelOfDetail, WebProcessMemoryDump*) override;
+
+    void setMemoryCache(MemoryCache* memoryCache) { m_memoryCache = memoryCache; }

[Primiano Tucci (use gerrit), 2015/10/14 11:06:58]

is there a way to check that the thread on which setMemoryCache is called is the
same one of onMemoryDUmp (is there an equivalent of ThreadChecker in blink)?
If that is not the case this would be racy

[ssid, 2015/10/14 12:06:40]

Done.

+
+private:
+    WebCacheMemoryDumpProvider();
+
+    WeakPersistent<MemoryCache> m_memoryCache;
+};
+
+} // namespace blink
+
+#endif // WebCacheMemoryDumpProvider_h