browser: make destruction of webContents async

[deepak1556]

Fixes #8930
Fixes #9230

Original discussion https://groups.google.com/a/chromium.org/d/msg/embedder-dev/XTv6SpambTI/V51xPz1hFgAJ

[deepak1556]

@zcbenz currently we have in api_web_contents.cc the following:

void WebContents::DidFinishLoad(content::RenderFrameHost* render_frame_host,
                                const GURL& validated_url) {
  bool is_main_frame = !render_frame_host->GetParent();
  Emit("did-frame-finish-load", is_main_frame);

  if (is_main_frame)
    Emit("did-finish-load");
}

calling webContents.destroy inside did-frame-finish-load will lead to a crash when did-finish-load is invoked with

[2853:0404/235125.714735:FATAL:wrappable.cc(24)] Check failed: !wrapper_.IsEmpty(). 
#0 0x7fe05839616e base::debug::StackTrace::StackTrace()
#1 0x7fe0583b2eab logging::LogMessage::~LogMessage()
#2 0x000000b8d723 mate::WrappableBase::GetWrapper()
#3 0x000000c782e5 mate::EventEmitter<>::GetWrapper()
#4 0x000000c7b1f9 _ZN4mate12EventEmitterIN4atom3api11WebContentsEE14EmitWithSenderIJEEEbRKN4base16BasicStringPieceISsEEPN7content11WebContentsEPN3IPC7MessageEDpRKT_
#5 0x000000c71bf4 _ZN4mate12EventEmitterIN4atom3api11WebContentsEE4EmitIJEEEbRKN4base16BasicStringPieceISsEEDpRKT_
#6 0x000000c6761a atom::api::WebContents::DidFinishLoad()
#7 0x7fe05c397999 content::WebContentsImpl::OnDidFinishLoad()
#8 0x7fe05c39777e <unknown>
#9 0x7fe05c3963fc content::WebContentsImpl::OnMessageReceived()
#10 0x7fe05c121888 content::RenderFrameHostImpl::OnMessageReceived()
#11 0x7fe057edeac5 IPC::ChannelProxy::Context::OnDispatchMessage()
#12 0x7fe058396cee base::debug::TaskAnnotator::RunTask()
#13 0x7fe0583bba1c base::MessageLoop::RunTask()
#14 0x7fe0583bbd38 base::MessageLoop::DeferOrRunPendingTask()
#15 0x7fe0583bc0fb base::MessageLoop::DoWork()
#16 0x7fe0583be4aa <unknown>
#17 0x7fe0499835a7 g_main_context_dispatch
#18 0x7fe049983810 <unknown>
#19 0x7fe0499838bc g_main_context_iteration
#20 0x7fe0583be316 base::MessagePumpGlib::Run()
#21 0x7fe0583bb7b7 base::MessageLoop::RunHandler()
#22 0x7fe0583e52b0 base::RunLoop::Run()
#23 0x7fe05c03bab8 content::BrowserMainLoop::MainMessageLoopRun()
#24 0x7fe05c03b905 content::BrowserMainLoop::RunMainMessageLoopParts()
#25 0x7fe05c03e94d <unknown>
#26 0x7fe05c03743e content::BrowserMain()
#27 0x7fe05c6db094 <unknown>
#28 0x7fe05c6d9ba0 content::ContentMain()
#29 0x000000aa569a main
#30 0x7fe047ae9511 __libc_start_main
#31 0x000000aa54e5 <unknown>

I can think of two ways to fix this:

• Make destroy async by default deepak1556/native-mate@c98d5d6 , will have to rewire cases inside the code base to follow this behavior, especially webview since we need to make sure they are destroyed before any host webContents.

• Return empty wrapper when its destroyed and check if object is available before emitting events.

What can be the best approach to fix this scenario ?

[zcbenz]

I prefer the second way, making destroy async feels scary.

[zcbenz]

This change looks good to me.

The destruction of WebContents is getting too complicated since we have too many ways to destroy a WebContents under different cases, we should find a way to simplify it in future.