======================================================================== http://mondrian.corp.google.com/file/8201059///depot/googleclient/gears/opensource/gears/localserver/ie/http_request_ie.cc?a=1 File //depot/googleclient/gears/opensource/gears/localserver/ie/http_request_ie.cc (snapshot 1) ------------------------------------ Line 567: info->stgmedData.pstm = static_cast<IStream*>(stream); On Mon Sep 15 12:17:51 2008 PDT, michaeln wrote: > Two things... > > 1) Now that we've added the ReleaseBindInfo call to the handler, I question > whether we're returning the stream properly here? > > If the caller is expected to call ReleaseBindInfo which is expected to release > the stream, that leads me to believe that we should AddRef the stream we return > here.
Don't we implicitly AddRef the stream in the next line down, "post_data_stream_ = stream", since post_data_stream_ is a CComPtr, which implicitly AddRef's and Release's? > 2) With YahooToolbar installed, our handler is NOT registered, and things are > not well behaved. Can you also run thru the unit tests with YT installed and > check for leaks/crashes in that case. I'm on IE6 / WinXP, and installing the YahooToolbar causes the localserver tests to fail quite spectacularly (the other tests, including database and workerpool, seem fine), both with and without my ReleaseBindInfo patch. And, occasionally Gears + YahooToolbar will trigger an assertion failure (that downloading_version == completed_version, on line 136 of localserver/common/update_task.cc). So, Gears + YahooToolbar + LocalServer usage = test failure and occasional crashes, but that was happening anyway, before this CL... ======================================================================== -- To respond, reply to this email or visit http://mondrian.corp.google.com/8201059
