Logging: test multithreaded logging

[IanButterworth]

Investigating #57376

With and without the redirect there's no crash locally.
Ideally it'd be good to test logging to the console but I don't think we can do that without noisy logs on the console..

• Tests for no crashes due to memory corruption
• Tests that no tearing occurs in log prints

[vtjnash]

You need to make the code actually correct first before adding tests for new uses cases which may be unreliable

[IanButterworth]

By "new use cases" do you mean logging from tasks running on other threads?

[vtjnash]

Yes, that has never been allowed before

[IanButterworth]

Ok then I think we need to do two things:

1. Elaborate the specific issues on Request for clarification on the thread-safety (or lack thereof) of the Julia logging system #57376 as has been requested by people (like myself) that were surprised to hear that logging from threads has always been forbidden and is highly dangerous (even though these test don't shake the bugs out)
2. Update the docs because as far as I can tell, it doesn't say anywhere that logging from threads is forbidden.

Hopefully we can avoid 2) by just fixing the stuff in 1) first, but I am unaware of the issues, and I believe from slack that you have a private list of issues @vtjnash so it'd be good to make that public.

I'll happily look into fixing this, I just am unfamiliar with the issue(s).

[simsurace]

Yes, that has never been allowed before

Could you confirm that this multi-threaded hello world program is in your opinion a user error:

Threads.@spawn @info "Hello from thread $(Threads.threadid())"

[vtjnash]

Yes, only functions that declare themselves thread-safe may be used

[simsurace]

Since the point of the logging macros is to abstract away the mechanics of how the logs are processed, doesn't an otherwise thread-safe function becoming unsafe by virtue of some logging statements violate expectations for this abstraction?