Fix using SemanticLogger#log(severity, message, progname)

[rqz13]

SemanticLogger provides interface which is fully compatible with Logger interface
It is possible to use SemanticLogger#add and SemanticLogger#log instance methods on SemanticLogger as it would be just plain Logger
Datadog instrumentation doesn't expect that first argument in method #log might be not SemanticLogger::Log but severity level like Logger::WARN as result fails with error NoMethodError: undefined method 'named_tags' for 2:Integer.

Originally met issue in Rails project, here is specs:

• rails 6.1.7.3
• rails_semantic_logger 4.12.0
• semantic_logger 4.13.0
• ddtrace 1.4.2

When have configuration for datadog:

Datadog.configure do |conf|
  conf.env = Rails.env
  conf.service = 'my_service'
  conf.tracing.instrument :rails
  conf.tracing.sampling.default_rate = 1.0
end

When have enabled :rails' instrumentation, ':semantic_logger added right away as there is SemanticLogger on project.
Real case scenario is that on project there might be some external gems that may allow to provide logger as part of it's configuration, but may work with logger as it's plain ruby's Logger and use #add or #log methods.
In case of using #log it causes failure because instrumentation prepended and there is no safeguard on first argument log

Example to reproduce having rails with setup datadog to instrument rails and semantic_logger installed:

SemanticLogger['MyService'].log(Logger::WARN, 'some warn')

[TonyCTHsu]

👋 @rqz13 Thanks for submitting this PR!

To be honest, I am not very familiar with SemanticLogger, it might take me some time to learn more about the details and plays around with it before giving proper feedback and code review.

[rqz13]

@TonyCTHsu thanks for looking into PR.

I'll try explain logic and data flow in more details, hope it would be helpful.
SemanticLogger guarantees compatimility with Logger interface, so it also provides implementation of those methods, add method and log method definitions.

The change I'm introducing based on check of arguments that are passed to log method. It might be SemanticLogger::Log instance when called with "traditional" methods like logger.info('some info') or it might be severity passed like for ruby's Logger logger.log(Logger::INFO, 'some info').
Because of instrumentation prepended into SemanticLogger this check is not performed and cause error, so I've add guard check in instrumentation.
When severity passed as log argument with the guard check it would call execution of SemanticLoggers original log method and also would to the check
Then flow would lead us into add method which would call log_internal
In log_internal method all the args would be processed and SemanticLogger::Log entity would be created and passed again to log method

Now we're back in our instrumentation method but now log argument would be instance of SemanticLogger::Log so the guard check would be skipped in that case and we can inject some extra tags into log.named_tags without failures.

[TonyCTHsu]

👋 @rqz13 , Thanks for the detail elaboration! I am able to reproduce the error and understand the problem! 👍