Pkg.jl: Pkg should not mutate the logger

[vtjnash]

Calling Pkg.PlatformEngines.probe_platform_engines! (probably probe_symlink_creation more specifically) results in the Logging subsystem getting mutated, causing issues with subsequent attempts to log messages. This is the cause of https://github.com/JuliaLang/julia/issues/37731

[StefanKarpinski]

I was going to say that I'm about to replace that code, but realized that I'm doing a very similar thing here:

https://github.com/JuliaIO/Tar.jl/blob/77c0db7d17a86852b3eaa94176db968f4cbfabed/src/Tar.jl#L22-L37

What is the correct way to implement this that doesn't screw up the logging system?

[KristofferC]

That place in Base just shouldn't log, at least not with a level that is visible by default imo. Error or shut up.

[StefanKarpinski]

The problem is that we expect symlink to fail and don't want it to spam the user.

[KristofferC]

Yes I know. That's why I am saying that Base shouldn't log so there would be no need to disable the logger.

[StefanKarpinski]

Ok, but it does: specifically, when a symlink fails on Windows, it tells the user that one cannot create symlinks on Windows. Perhaps a better way to do this is to throw an error with an error message that includes that message. Then there wouldn't be any need for logging or printing a warning.

[KristofferC]

Yes, aka "error or shut up". ::D

It's precisely for reasons like this that (more or less) unconditional printing inside some library function is kinda dumb.

[vtjnash]

Yep "error or shut up", happy to have this thrown back onto being Base JuliaLang issues too. 👍

[StefanKarpinski]

I think we should stop disabling logging here, bump Pkg, and make CI happy again. As a separate issue, we can get rid of the warning in symlink in Base.

[tkf]

I'm not against the changes already in the pipeline. But I guess it would have not caused the problem if we were using with_logge(f, NullLogger())?