1588530 - [meta] Crash reporting Rust rewrite

Status: NEW

(Toolkit :: Crash Reporting, task)

Description

[Gabriele Svelto [:gsvelto]]

We'll use this meta-bug to track all activities related to the crash reporter rewrite. Here's links to the design documents with all the nitty-gritty details:

• Issues with the existing crash reporter
• Crash reporter rewrite high-level overview
• Crash reporter rewrite implementation proposal

Comment 1

[David Teller [:Yoric] - still alive but not very active]

Do the Linux libc limitations of the current crashreporter (which I don't quite understand yet) carry over to the Rust reporter?

Comment 2

[Gabriele Svelto [:gsvelto]]

(In reply to David Teller [:Yoric] (please use "needinfo") from comment #1)

Do the Linux libc limitations of the current crashreporter (which I don't quite understand yet) carry over to the Rust reporter?

Generally speaking no. Most of the limitations regarding libc on Linux (and functions doing allocations on macOS & Windows) stem from the fact that they run within the exception handler and so they might trigger a recursive exception which we obviously don't want. Right now when the main process crashes we have to generate the minidump from within the crashed process which causes the exception handler to be large and complex. The Rust rewrite will do all minidump generation out-of-process so the exception handler will be a lot smaller and thus will only require to do a minimum amount of external calls to do its job.

You can find more details on this problem (and other ones too) in the documents in comment 0.

Comment 3

[David Teller [:Yoric] - still alive but not very active]

Most of the limitations regarding libc on Linux (and functions doing allocations on macOS & Windows) stem from the fact that they run within the exception handler and so they might trigger a recursive exception which we obviously don't want.

Yeah, that's what I figured at first. And then, digging up in the code, it started mentioning sandboxing, at which point I stopped.

You can find more details on this problem (and other ones too) in the documents in comment 0.

You're right, I should have read this first. Thanks, will catch up on this!