Closed Bug 1248462 Opened 4 years ago Closed 4 years ago
Implement telemetry tracking to measure the impact of different driver blacklisting activities
We have a good architecture for blacklisting certain OS+hardware+driver combos from using different features when they are deemed to crash or prove some other security issue. See e.g. http://lxr.mozilla.org/mozilla-central/source/widget/windows/GfxInfo.cpp#831 and this wiki page https://wiki.mozilla.org/Blocklisting/Blocked_Graphics_Drivers (which is a bit out of date now though) Currently we don't have a method in place to get data about how big of an impact our blacklisting activities cause. However, at the same time, the above code has comments such as "The driver versions used here come from bug 594877. They might not be particularly relevant anymore." "We're black listing all of the devices to be safe even though we've only confirmed the issue on the G45" "OpenGL on any Intel hardware is discouraged" "This may not be needed at all" "XXX this should really check for the matching Intel piece as well. Unfortunately, we don't have the infrastructure to do that" "WebGL issues with D3D11 ANGLE on Intel. These may be fixed by an ANGLE update" which suggests that our blacklisting activities might not be the sharpest possible to only affect the exact set of hardware that exhibits the issue. Ideally, our code would be perfect and we'd be blacklisting as little as possible and nothing more, but that of course might require acquiring a lot of test hardware to reproduce all the different cases, not to mention the testing effort. Instead, what we should do is to enhance this blacklisting infrastructure to report telemetry data back to us, so that we can measure what the impact of each blacklist operation is. This will give us a good sense of which blacklist items we should care to investigate in more detail. If a single blacklist entry is blocking e.g. 5% of users from using WebGL, we might want to examine that to find the narrowest/sharpest blacklist entry possible, but if instead the blacklist entry affects only a negligible set of users, then we might not bother. Also, this infrastructure would give us good data to communicate to our partners who can use it accordingly when estimating their expected capable audience e.g. for WebGL content.
4 years ago
I think it's fair to resolve as a duplicate of bug 1259536 which I've effectively scope-creeped to do it for all features.
Status: NEW → RESOLVED
Closed: 4 years ago
Resolution: --- → DUPLICATE
Duplicate of bug: 1259536
You need to log in before you can comment on or make changes to this bug.