1884947 - Crash reports are not per-profile

Status: NEW

(Toolkit :: Crash Reporting, defect)

Description

[Alex Franchuk [:afranchuk]]

Crash reports are all stored in an application-specific directory, but not a profile-specific directory. The result is that profile settings about how to handle crash reports are not always respected. If a crash occurs under one profile which has crash report submission disabled, but then another profile is loaded which allows submission, the report will be submitted, potentially against the user's intention by using separate profiles.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The severity field is not set for this bug.
:gsvelto, could you have a look please?

For more information, please visit BugBot documentation.

Comment 2

[Gian-Carlo Pascutto [:gcp]]

It's unfortunately also possible to crash before we had a chance to select the correct profile :-)

Comment 3

[Alex Franchuk [:afranchuk]]

Right! For other reasons such as ease of access for the crash reporter client, I think that crash reports being stored in one place is likely still okay, however we need to record whether (if available) the originating profile intended for the report to be submitted or not. I'm not sure what the behavior should be if a profile was not yet loaded, however if we default to the conservative "don't send" stance then we'd probably miss out on a class of bugs. That case would probably be very likely to be characterized by a main process crash though, so the crash reporter client would allow the user to make a choice. Notably, if the profile isn't loaded then a "always send reports" option won't be available, so the user might be confused by seeing the crash reporter client when they have already configured reports to always be sent. It would be nice to add some additional text in the client for startup crashes like this: "Firefox crashed before it could load your profile settings."

Comment 4

[Gabriele Svelto [:gsvelto]]

We could add an annotation for the current profile, and it could be "Unknown" for early startup crashes where we don't have one. We'd need to check how this interacts with automation though, since we don't always use a fake profile there.