Logging framework wont work with multiple models in one executable

[MatthewHambley]

Are there any linked Issues or Pull Requests?

No response

Brief description

Because the logger is based on global scope variables it will not work if we want two modeldbs in an executable. For instance, it is not unreasonable to want each to be logging at different levels. This cannot work under the current regime.

Further details of the issue.

Global scope logging properties include logging level and standard and error output streams.

[MatthewHambley]

Two different approaches spring to mind.

• A logging class, held in modeldb
• A logging namespace tree accessed with the singleton pattern. (See Python)

Steve has suggested another option

• Between calling different models, re-initialise the framework. This is in relation to linear and non-linear models in D.A.

[stevemullerworth]

Some comments on support for applications that have two or more models or components where you may want different logging choices.

There might be other global settings such as writing to only one rank? And if we are writing to different streams we would need to define different file names.

For supporting the different log levels and streams, one could make the logger a class that is instantiated for each model and that is included in the relevant modeldb instance, but this would add code bloat to the calls and would require the logger to be passed to every subroutine that has logging.

A simple solution that would be adequate for many apps would be to reset the logger settings at the start of each timestep for each model.