fix: Prevent usage of internal UE logger during crash handling

[mujacica]

• Prevent usage of internal UE logger during crash handling
• Disable console logging for Windows crashes

[nwhite-riot]

@mujacica, @tustanivsky, can I ask what this was for? It actually makes the log file worse because Unreal doesn't flush logs, losing a number of ticks of logs (including the actual tick before the crash). I'd even opened a ticket to replicate the log flush behavior on Xbox because it improved the output: https://github.com/getsentry/sentry-xbox/issues/36.

This ticket doesn't explain why this is necessary.

[mujacica]

Hey @nwhite-riot, the reason for this is mainly trying to address #992 and #991. So in general, if anything goes wrong with UE Allocator/Stack, calling UE_LOG would cause a secondary crash of Sentry Subsystem, so we would not get the event at all. The solution that we did here was not ideal, but if you have any ideas on how to support Allocator/Overflow problems, while at the same time flushing the logs when the crash happens, I am glad to discuss.

[nwhite-riot]

Does Ivan's comment not resolve this: #992 (comment)? Specifically:

We'll check if it's feasible to pre-allocate the required amount of memory (e.g. after the SDK initialization) and ensure that the crash stack trace is properly added to logs

Does pre-allocating memory for the string allocation not resolve this issue?

To be clear, Unreal performs this same logic when it handles crashes, so I'd be surprised if the logic was fundamentally flawed.

[mujacica]

Sadly, it does not. We can pre-allocate memory in our code, but since both StackWalk and Logger itself are allocating (using Unreal allocators) internally, it doesn't solve the problem.

I spent quite some time testing this, and with poisoning memory allocator and/or causing stack overflow, a call to UE_LOG/GLog would cause secondary crashes. But as I said, maybe there is a way around it. One of the things we were discussing was sending these logs via Sentry Logs.

[nwhite-riot]

I'm assuming this is a problem because we're handling the crash on the main thread - hence why a stackoverflow blows this up. CRC handles writing these logs on a separate thread, so is there a way we can offload this to a separate thread? I realize things like the screenshots etc. require the game thread, but that doesn't mean everything has to be