fix: turns warnings into errors temporarily and internally

[mcanouil]

To avoid complexity of the code, we use try() to catch errors and warnings by temporarily setting options(warn = 2), which turns warnings into errors.
Then, we set back the original value of options(warn).
This is not ideal, but it works.

Fixes #110

[burrowsk]

That seems a sensible work around without doing anything too elaborate with the code.
However, I assume that if the first model call fails (for either warning or error) and the second call (with increased iterations) is initiated, that if this call produces warning messages then the call will complete? And the user will see the warning message?

[mcanouil]

Yes, only the first call of the model.

Do we want absolutely no warnings at all?

[burrowsk]

Thanks, I just wanted to make sure I understood it correctly.

My gut feeling is to allow warnings to run in order to assess the coefficients and allow a case by case decision on whether the models are good enough to use for that particular dataset. I think future users may prefer this too.

However, for the EGG projects it means trusting that the warning messages are picked up and reported back to the central team. Unless you have a strong opinion we should probably trust that warnings will be picked up and reported back.

[mcanouil]

At least, in the code, I don't catch warnings.
I could by adding a bunch of code.

Here a proposal, let's keep the current workaround, open a new issue to improve how warnings and errors are captured to be worked on a bit later.

[burrowsk]

Okay - that sounds like a good plan to me, I will approve the review now. Thanks very much!

[mcanouil]

See #113 for improvement on how warnings are handled in the linear mixed model.