Call RegisterCrashKeys for chrome.exe in static build.

chrome/app/chrome_crash_reporter_client.cc

Index: chrome/app/chrome_crash_reporter_client.cc
diff --git a/chrome/app/chrome_crash_reporter_client.cc b/chrome/app/chrome_crash_reporter_client.cc
index 25769f6cba1c476d807fcfad9a4b638df5a021e7..71d99bf676f5aaf9858a589b5b3c417c0a3a3f24 100644
--- a/chrome/app/chrome_crash_reporter_client.cc
+++ b/chrome/app/chrome_crash_reporter_client.cc
@@ -321,9 +321,11 @@ bool ChromeCrashReporterClient::GetCrashDumpLocation(
 }
 
 size_t ChromeCrashReporterClient::RegisterCrashKeys() {
-  // Note: This is not called on Windows because Breakpad is initialized in the
-  // EXE module, but code that uses crash keys is in the DLL module.
-  // RegisterChromeCrashKeys() will be called after the DLL is loaded.
+  // Note: On Windows this only affects the EXE. A separate invocation from
+  // child_process_logging_win.cc registers crash keys for Chrome.dll.
+  // Also note that, on Windows, this is not called in a component build, as in
+  // that case a single copy of 'base' is shared by the EXE and the various
+  // DLLs, and that copy is configured by child_process_logging_win.cc.
   return crash_keys::RegisterChromeCrashKeys();

[grt (UTC plus 2), 2015/03/17 02:59:31]

for safety's sake, should this be
#if defined(COMPONENT_BUILD)
  CHECK(false);
#else
  return crash_keys::RegisterChromeCrashKeys();
#endif

[erikwright (departed), 2015/03/17 14:40:03]

I think a NOTREACHED is more appropriate?

The guaranteed outcome of double-registration is a CHECK in
base::debug::InitCrashKeys.

The outcome of non-registration is a DCHECK in debug and missing crash key
values in release.

So I don't see any reason to hard-terminate the process in a release build here,
but I agree that a NOTREACHED is a good way to catch an error.

 }