1567010 - WidevineCdm plugin has crashed on Amazon

Status: RESOLVED WORKSFORME

(Core :: Audio/Video: Playback, defect, P2)

Description

[Asif Youssuff]

Attachment: Mozgression-regression report-07-12-19.txt

Report from a reddit user: https://www.reddit.com/r/firefox/comments/cby2gr/widevinecdm_plugin_has_crashed_firefox_quantum_680/

User says it is a regression (67 worked correctly).

User says "I can't watch videos from Amazon."

Comment 1

[Marcia Knous [:marcia]]

Adding bvandyk since he seems to be the developer working in this area.

Comment 2

[Ryan VanderMeulen [:RyanVM]]

What's the relationship between this bug and bug 1548845, which was a change to our test suite and not to anything we ship?

My first thought is the new Widevine update we shipped recently being a possible culprit.

Comment 3

[Asif Youssuff]

It didn't make any sense to me either, but that is what the user came up with on mozregression, Perhaps it is harder to reproduce than they thought and they came up with a false positive.

You can see their mozregression log attached to this bug - or maybe it is just the new widevine update.

Comment 4

[Ryan VanderMeulen [:RyanVM]]

Maybe QA can take another look at the regression range too. A test-only change doesn't make sense.

Comment 5

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

As mentioned above, the regressing bug seems an unlikely one to regress this. With the switch to 68 we're shipping a new CDM, so that seems like the more likely culprit.

Holding NI so I can look further into this when I have a moment, and when we hopefully have some further information.

Comment 6

[Asif Youssuff]

Attachment: User's about:support

User's about:support:

Comment 7

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

There's a couple of problems with regression finding for these CDM related bugs to be aware of:

• The CDM is not versioned based on cenrtral and is instead based on balrog rules that can change independent of releases.
  • The rules have changed recently, and I note that we currently have a rule for >=68.0 which is a bit of a footgun because I believe that it matches 68.0 but not 68.0a1.
• The MOZILLA_OFFICIAL define impacts CDM behaviour. Central builds will have this set, but various other builds, such as autoland, will not. This leads to false positives when do regression ranges due to some sites breaking (though the CDM shouldn't crash) when mozregression jumps over to autoland and similar.

Comment 8

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

Based on the discussion on the reddit thread I wonder if this is related to bug 1496607. It would be hard to know without some non-trivial debugging from our reporter, but the 32bit builds on 64bit Windows misbehaving in hard to nail down ways is consistent with that bug.

Comment 9

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

Summary of reddit debugging so far:

• User has 32bit build of Firefox on 64bit Windows. This should be fine, and is something I test when doing Widevine updates.
• Issue does happen on other premium media sites.
• Issue does happen on fresh profile.
• Issue doesn't happen on 67.0a1, 68.0a1, 69.0a1 builds from mozregression (64bit builds).
• Issue doesn't happen on newly installed 64bit Fireox.
• Doesn't happen on another Windows 10 machine owned by reporter.

Comment 10

[Asif Youssuff]

⠀

Comment 11

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

Marking as fixed worksforme based on user reports that bumping to a different Firefox version fixes the issue. Despite not starting at the same version as bug 1496607, that bug remains my best guess as to the issue.