-
David Tseng authored
In the attached bug, speech is triggered in a startup, fresh imaged environment. To replicate this, `rm -rf /home/chronos/*`, then trigger some tts after boot. The browser process hangs, e.g. gdb attach <browser_pid> ... bt ... #5 0x000058e02342da8f in TtsExtensionEngine::GetVoices(content::BrowserContext*, std::__1::vector<content::VoiceData, std::__1::allocator<content::VoiceData> >*) () #6 0x000058e021612615 in content::TtsControllerImpl::SpeakNow(std::__1::unique_ptr<content::TtsUtterance, std::__1::default_delete<content::TtsUtterance> >) () #7 0x000058e021612e8f in content::TtsControllerImpl::SpeakNextUtterance() () #8 0x000058e01ff769d4 in content::TtsControllerImpl::VoicesChanged() () #9 0x000058e02342eeb4 in ExtensionTtsEngineUpdateVoicesFunction::Run() () #10 0x000058e02185ebe2 in ExtensionFunction::RunWithValidation() () #11 0x000058e021860c5d in extensions::ExtensionFunctionDispatcher::DispatchWithCallbackInternal(ExtensionHostMsg_Request_Params const&, content::RenderFrameHost*, int, base::RepeatingCallback<void (ExtensionFunction::ResponseType, base::ListValue const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)> const&) () #12 0x000058e021860773 in extensions::ExtensionFunctionDispatcher::Dispatch(ExtensionHostMsg_Request_Params const&, content::RenderFrameHost*, int) () #13 0x000058e0218818ea in bool IPC::MessageT<ExtensionHostMsg_Request_Meta, std::__1::tuple<ExtensionHostMsg_Request_Params>, void>::Dispatch<extensions::ExtensionWebContentsObserver, extensions::ExtensionWebContentsObserver, content::RenderFrameHost, void (extensions::ExtensionWebContentsObserver::*)(content::RenderFrameHost*, ExtensionHostMsg_Request_Params const&)>(IPC::Message const*, extensions::ExtensionWebContentsObserver*, extensions::ExtensionWebContentsObserver*, content::RenderFrameHost*, void (extensions::ExtensionWebContentsObserver::*)(content::RenderFrameHost*, ExtensionHostMsg_Request_Params const&)) () The hang: TtsControllerImpl::SpeakNextUtterance -> TtsControllerImpl::SpeakNow -> TtsControllerImpl::OnSpeakFinished -> enqueues the utterance back onto |utterance_list_| on failure. which occurs all while(... !utterance_list_.empty()), leading to a browser hang. Solution: The above utterance fails immediately because the voice is not yet ready. We fix this by never returning the voice in the first place, in TtsControllerImpl::GetVoices. Therefore, we never get to the point where an utterance matches against a not yet initialized voice, leading to an attempt to speak, then to an immediate failure, next to a re-enqueue, and finally a hang. Bug: 1161107 Change-Id: I2d88c0c6e86bf79ad9ac5dc5240825848daa40ae Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2642746 Commit-Queue: David Tseng <dtseng@chromium.org> Reviewed-by:
Dominic Mazzoni <dmazzoni@chromium.org> Cr-Commit-Position: refs/heads/master@{#845861}
b34ba0a1
