Fix logger refactoring regression: manager logging add

[wxiaoguang]

The rotate option is optional.

Test result:

[silverwind]

IIRC, there is a lint rule for such missed second arguments on type casts, we should probably use it.

[wxiaoguang]

I do not think we need a lint for this case.

Eg: here is a real use case: for a field t any , the caller1 knows its type must be T1, caller2 knows its type must be T2, so the code could be written as : x := t.(T1) directly for caller1. If there is any bug, the runtime panic tells the developers that "you passed incorrect type here, please fix it".

But if it's written as: x, _ := t.(T1) , it loses the strict type check, the caller1 just gets a default T1, which is unexpected.

If it's written as x, ok := t.(T1), then you need to write another if !ok { panic } to show the error.

So, x := t.(T1) is good enough for many cases.

[silverwind]

If there is any bug, the runtime panic tells the developers that "you passed incorrect type here, please fix it".

Runtime panics are bad because the condition for the panic may not be hit during testing.

If it's written as x, ok := t.(T1), then you need to write another if !ok { panic } to show the error.

I think forcing this style would be okay. Go is (overly) explicit with it's error handling, just not in this case apparently, but it could be enforced with the linter rule.

[silverwind]

For reference, this is the lint option I'm talking about:

linters-settings:
  errcheck:
    # Report about not checking of errors in type assertions: `a := b.(MyStruct)`.
    # Such cases aren't reported by default.
    # Default: false
    check-type-assertions: true

[lunny]

For reference, this is the lint option I'm talking about:

linters-settings:
  errcheck:
    # Report about not checking of errors in type assertions: `a := b.(MyStruct)`.
    # Such cases aren't reported by default.
    # Default: false
    check-type-assertions: true

Maybe another PR.