Trainer logger is None when given a DummyLogger instance

[MetaHG]

Hi!

I am encoutering an issue with pytorch_lightning.loggers.logger import DummyLogger.

I have implemented some PL Module where I sometimes call my logger like this:

self.logger.experiment.log(...)

Such calls are typically used when logging "media" objects with a WandbLogger for example.

When doing code adjustements and debugging, it is sometimes useful to disable the logger. (Note: I know that there is a parameter called fast_dev_run matching this purpose partially). To disable the logger, I tried the two following Trainer instanciation:

Trainer(logger=False, ...) # First solution

Trainer(logger=pytorch_lightning.loggers.logger.DummyLogger(), ...) # Second solution

However, in both cases, my code crashes when reaching statements such as

self.logger.experiment.log(...)

where the following error is reported

AttributeError: 'NoneType' object has no attribute 'experiment'

This means that self.logger is None, even when setting the logger to a DummyLogger instance.

I don't understand why the PL Module self.logger is None when explicitly setting it to a DummyLogger and I think that it should not be the case. What am I doing wrong? Am I missing something?

Please let me know if this looks more like a bug and if I should open an issue instead.