525849 - Design UI for submitting crash reports from crashed out-of-process plugins

Status: RESOLVED DUPLICATE of bug 538910

(Toolkit :: Crash Reporting, defect)

Description

[(not currently active) Ted Mielczarek]

If we're going to ship out-of-process plugins, we'll need to suport crash reporting for the plugin process (bug 516759). However, that means we'll need to figure out what the user interface looks like if the plugin process crashes. Currently, since we only have one process, we just pop up the crash reporter when anything crashes, since it takes the entire application down.

Some possibilities might include:
1) Drawing some indicator icon in the plugin frame in place of the plugin content, showing some information when clicked.
2) An infobar indicating that a plugin has crashed--might fit well with the out-of-date plugin indicators we have now.

I'm not sure on the specifics of our OOP plugins architecture, so I don't know if we'll know the exact name of the crashing plugin or not. Also, we might consider simply restarting the plugin process if it crashes, and only notifying the user if it crashes more than once in a single session, or something like that.

Comment 1

[Benjamin Smedberg]

We will definitely have to draw something in the plugin frame: I believe Chrome has "the plugin crashed, click here to restart it" UI.

I'm pretty sure we can associate a crash with a loaded plugin... depending on circumstances we may be able to associate it with a particular page currently calling the plugin.

It's important to receive crash reports: I don't know how we want to do that; the best UI for the user is just a doorhanger and/or click-to-restart, but is that enough notification that we're going to send a crash report off?

We shouldn't auto-restart within a single page: for pages where script interacts with plugins, restarting the plugin can give odd behavior. We should restart if the user refreshes or loads a new page with the plugin.

Comment 2

[Benjamin Smedberg]

I'm wrong about what Chrome does. When a plugin crashes, the plugin frame is replaced by a frowny puzzle piece, and a doorhanger appears which says "The following plug-in has crashed: Shockwave Flash". Neither place allows you to restart the plugin in the same page: you have to hit refresh.

Comment 3

[(not currently active) Ted Mielczarek]

Ok, so how about something similar? A "broken puzzle piece" icon for the crashed plugins + a doorhanger notification that says "A plugin on this page has crasshed. [More info]". Clicking either the plugin object or the [More info] button brings up a dialog/sheet/error page/whatever that gives more detail, with a [x] Submit crash checkbox, and a "Reload page" button that will submit the crash and reload the page.

Comment 4

[Benjamin Smedberg]

This bug doesn't actually depend on hooking up the crash reporter: IPDL will inform each actor that the other process is gone and we can handle it from there independent of the crash reporter hookup.

Comment 5

[(not currently active) Ted Mielczarek]

Didn't realize bug 519541 was already filed, but let's flip the dependencies here. bug 519541 will be about the basic "plugin crashed" UI, independent of Breakpad. This bug will be about getting crash reports submitted from that UI.

Comment 6

[Josh Aas]

I'm going to hold off on landing bug 519541 until this is ready to go since otherwise it'll be unused. If someone wants me to land it though, that's fine with me.

Comment 7

[(not currently active) Ted Mielczarek]

Josh: seems like we could still land your patch and write a test for it without this part landing.

Comment 8

[Benjamin Smedberg]

Yeah, I think we should land the first part as soon as its ready, to provide a stable base for the UI work and get faster testing of this part.

Comment 9

[Benjamin Smedberg]

Dolske is going to land the plugin-crashed notification UI first, and then somebody will  hook up the crash reporting to the UI.

Comment 10

[(not currently active) Ted Mielczarek]

It sounds like Dolske is just planning on doing the entire thing in bug 538910, so let's just close this bug.