This could perhaps break someone's test, but it seems unlikely. I think generally we have chosen not to parse the arguments in the TestLogger, so that the test has a true precise record of what was called (for example, we also ignore the minlevel). But it sounds like #41625 has support for changing this particular behavior.

I think generally we have chosen not to parse the arguments in the TestLogger, so that the test has a true precise record of what was called

Yes, I think this change is a bit questionable in that respect — perhaps a neater strategy would be to enhance the pattern matching in @test_logs rather than modify which log events are recorded. This (edit: ie, the current PR) is easier though...

perhaps a neater strategy would be to enhance the pattern matching in @test_logs rather than modify which log events are recorded.

What would you suggest?

Alternatively, I wonder how much one can really access the log events recorded via the public API- perhaps exposing a richer interface would let users test things like maxlog without needing special handling here. I’ve found @test_logs a bit difficult to use compared to Test.collect_test_logs but I don’t think that’s part of the public interface.

for example, we also ignore the minlevel

@vtjnash that's not true, right?

julia> Test.collect_test_logs() do
       @debug "hi"
       end
(Test.LogRecord[], nothing)

Ah, apparently I was wrong