QA/Platform/Graphics

Consult this page for more information.

Documentation

• Team Page
• Device Inventory (moved to wiki)
• PCI Device ID List
• nVidia Codenames
• Graphics driver blocklisting
• Draft guides

Understanding the Problem Space

First order of business for my transition to the Graphics team is to understand the problem space so I can understand the immediate needs of the team and make the best impact I can in the shortest amount of time.

What are the key problems/challenges facing the Graphics team in terms of quality?

• discrepancy in environments between testers and release users
• discoverability of bugs pre-release
• ?...

Where can QA add value/support to the Graphics team?

• improving pre-release discoverability of bugs
• closing the gap between tester and release systems
• helping with bug triage, particularly with bugs hiding in general components
• representation in crashkill
• improving code coverage and/or identifying gaps in code coverage
• identifying ways to improve participation in the graphics team (events, projects, One & Done, etc)
• documentation of tools, testing processes, etc
• building out the lab in Toronto
• continuing to drive Betabreakers testing every 6 weeks
• verifying bug fixes (what does this look like)?
• profiling areas of risk (eg. troublesome configs)
• conducting root cause analysis for regressions
• understanding problems outside of our control (eg. driver resets)
• feature testing and upcoming priorities (e10s, Windows 10, El Capitain, Android, B2G, etc)

What does QA need to know to be effective?

• key components of an actionable Graphics bug
• fundamentals/technologies that should be learned
• how to distinguish a graphics crash from a non-graphics crash with a graphics signature
• meetings, mailing lists, bugzilla components to watch, blogs, IRC channels to join, etc
• who is each member of the team (incl. contributors) and what do they do
• where does graphics code reside in the tree?
• what role does Unified Telemetry in graphics quality?
• what are the prefs to enable/disable different functionalities?
• we need a database of known-troublesome hardware/driver configurations to inform testing, hardware acquisitions, and blocklisting

Participation

• Sanity checking via One & Done
• Meetups to connect testers/users with devs
• Testdays to teach people about graphics testing
• Documentation and translation of documentation
• Engaging on community spaces (Discourse, Reddit, Facebook, Twitter, etc)

Telemetry

• COMPOSITE_TIME: time in CompositorParent::CompositeToTarget dispatching draw calls and calling SwapBuffers, but not texture upload (ie. complete composition)

Projects

Project	Deliverable	Status
Sanity Checking	Betabreakers	On-going
	Crowd Testing	Drafted materials in peer-review
	Toronto Graphics Lab	Testing has indefinitely ceased
Stability	Device Driver Crashes	Policy development in progress
	Top Crashes	Policy development in progress
	Explosive Crashes	Policy development in progress
Bug Triage	Bugs to be Investigated	Policy development in progress
	Untriaged Incoming Bugs	Policy development in progress
Metrics	WebGL Success Rate Tests	Blocked on Socorro APIs
	Determining a baseline for Graphics Quality	Not yet started

Sanity Checking

The purpose of this project is to identify pre-release graphics bugs in Firefox on desktop platforms (Boot2Gecko is tracked separately).

Betabreakers

Betabreakers provides sanity testing of Firefox Developer Edition on a wide-array of systems.

• [DONE] Firefox 38: MSE stress test
• [DONE] Firefox 39: beta sanity check
• [DONE] Firefox 40: WebGL with e10s
• [DONE] Firefox 41: WARP disabled on Windows 7 and earlier
• [DONE] Firefox 42: Windows 10 AMD/NVidia stability
• [ON TRACK] Firefox 43: Typical sanity check plus bug cleanup
• Firefox 44: to be determined

Bugzilla query error

Array ( [type] => error [message] => http-bad-status [params] => Array ( [0] => 406 [1] => Not Acceptable ) ) 1

Crowd Testing

One & Done provides a task which enables volunteers to run a sanity test and report results to the team.

• Task in One & Done
• Form of Results

Toronto Graphics Lab

NOTE: Activity has stopped as of September 1, 2015 while we determine how best to resource this going forward.

The Graphics team hosts an ad-hoc lab of systems in Mozilla's Toronto office for testing and debugging.

Test (results)

• Start with a new profile (see above)
• Resize the window so it's at least 1024x768
• Go to nytimes.com, scroll up and down for ~15 seconds
• Open another tab and go to pinterest.com, scroll up and down for ~15 seconds.
• Switch back and forth between the two tabs three times
• In nytimes.com tab, go to youtube.com and play this video, this video, and some other random video for 15 seconds
  • Note: If a video plays with Flash find another video to test. To check, right mouse click on the video and choose “Stats for nerds”. The video is valid if the mime type is video/mp4 or video/webm.
• Go to cnn.com, scroll for ~15 seconds then find a video and play it
• Update your graphics driver to the latest available version
  • Note: don't restart Firefox or your computer if it can be avoided. If Firefox crashes, report the crash, restart Firefox and continue on to step 9.
• Repeat the test from step 3 through 7.
  • Note: if Firefox crashes, report the crash, restart Firefox and repeat steps 3 through 7.
• Quit Firefox after all steps have been completed and submit your test results.

Stability

The purpose of this project is to identify the most concerning graphics crashes and escalate them to developers.

Understanding the Problem

How do we identify a graphics crash?

• by signature: gfx, layers, D2D, D3D, ?...
• by topmost filename: gfx, ?...
• by driver (DLL, version, ?...)
• by device/vendor ID?...
• ?...

How do we prioritize graphics crashes?

• Overall topcrashes in release > beta > aurora > nightly
• Gfx crashes in release > beta > aurora > nightly
• Explosive crashes in release > beta > aurora > nightly

What tools do we have at our disposal to investigate crashes?

• Bughunter for investigating crashes correlated to a URL
• KaiRo's reports for identifying crashes that are new or escalating quickly
• Socorro for getting detailed information about crash reports

What information is needed to make a crash actionable by developers?

• Correlations to particular hardware, driver, add-on, 3rd-party software, or library
• ?...

Device Driver Crashes

Queries

• Top crashes by device
• Top crashes by driver
• AMD: amd*.dll, ati*.dll
• Intel: igd*.dll
• NVIDIA: nv*.dll

Top Crashes

Process

1. Load the Firefox crash-data dashboard
   The first thing you'll see is a graph showing the crash rate for each of the branches.
   The second thing you'll see is a list of reports for each of the branches.
2. Click the Top Crashers link for the version you want to look at
   From here you'll see a large table of crash signatures ranked by the number of reports in the last week.
   Typically you'll want to start by looking at Release (right-most) as that's the highest priority.
3. Starting at the top of the table, scroll down until you find a graphics-related crash
   ** NEED A PRIMER ON IDENTIFYING A GRAPHICS CRASH **

Bugs

Bugzilla query error

Array ( [type] => error [message] => http-bad-status [params] => Array ( [0] => 406 [1] => Not Acceptable ) ) 1

Explosive Crashes

• Explosiveness Reports: use this to find and report new/explosive crashes (MDN Docs)

Cold Topcrashes

The following tracks resolving cold topcrashes, ie. crashes which have not been updated in over 30 days. In a nutshell we need to check which of these crashes still happen, trying to bring new life into crashes which still exist and closing those which do not.

Bugzilla query error

Array ( [type] => error [message] => http-bad-status [params] => Array ( [0] => 406 [1] => Not Acceptable ) ) 1

All topcrashes

Bug Triage

The goal of bug triage is to ensure every bug report has been reviewed and prioritized in a reasonable amount of time. This responsibility currently rests with engineers on Platform Engineering's Graphics team, managed through a rotating weekly duty. While QA is not directly involved in this process we do have a role to play. First, we serve to assist with investigations as requested. Second, we serve as a safety-net to make sure bugs aren't slipping through the cracks. The following documents each of these activities.

Bugs to be Investigated

Note: this section currently tracks bugs reported in the last year.
The following issues have gone cold as they need steps to reproduce, a regression window, or a minimized testcase. If the bug has not been updated in more than six months, mark the bug as RESOLVED INCOMPLETE and ask the reporter to reopen the bug if they can still see the issue in the latest Firefox version. Otherwise, attempt to reproduce the bug on your system and provide any details you can about your testing if you're able to reproduce the issue. If you're not able to reproduce the bug, ask the bug reporter for more information and flag them with the needinfo? flag.

Bugzilla query error

Array ( [type] => error [message] => http-bad-status [params] => Array ( [0] => 406 [1] => Not Acceptable ) ) 1

Untriaged Bugs

The following bug reports were missed from the rotating developer triage duty. Each of the bugs below needs to be triaged to make sure they have sufficient information to proceed with investigation and to be appropriately prioritized. The bug report should have the [gfx-noted] whiteboard tag added once triaged. If you're not sure please use the needinfo? flag to ask :ashughes for advice.

Bugzilla query error

Array ( [type] => error [message] => http-bad-status [params] => Array ( [0] => 406 [1] => Not Acceptable ) ) 1

Get Involved

Want to help the Graphics team? Here are some ways you can get involved.

• Run a sanity test
• Add your details to our inventory