Unit tests and refactoring

[charlie-gallagher]

Hi @benrwoodard, first, if you'd like to merge this into a different branch than 'master', feel free to redirect this.

Anyway, I was traveling the past few days so I took a shot at unit tests for the package using testthat. I didn't try to test everything, just those features that were easily testable without an internet connection.

This led to a lot of refactoring, but no feature changes besides the changes I submitted in the other merge request. Fortunately, I didn't catch many existing bugs, but the revised code I think is much better and easier to use as we add new features.

I still consider this a work in progress, but I think it's a good step forward.

Summary

• metric_container no longer requires you supply a vector of metric IDs. You can generate a unique list of vector IDs with the utility function create_metric_column_id.
• I extracted the formatting of the top argument into a utility function called make_explicit_top. It doesn't replace zeros, but it obeys the rules for making implicit 'top' arguments into explicit values, and recycles 'top' while following the rules. I renamed top_daterange_number to recalculate_top_arg to better describe what it does. It also now returns a named character vector (pairing dimensions and their respective 'top' values), which is good for debugging but otherwise has no effect on the results.
• I cleaned up the functionality of global_filter so that you can pass it arguments more intuitively. Only segmentId and dateRange are required, so you can focus on which values you want rather than where to put NAs like before. The type argument is handled automatically as well, to reduce errors.
• I wrote many unit tests for functions mentioned above, and others. In addition to unit tests for functions that don't require an internet connection, I'm planning "acceptance tests" that run a gamut of queries trying to trip up the package.
  • In particular, unit tests exist to differing extents for: all functions in request_format.R, make_timeframe, recalculate_top_arg, make_expicit_top, and format_URL_parameters.
• This is also a branch on #114, so it incorporates those changes as well

Requests for Comment

• aw_get_reportsuites had empty strings as defaults for two of its arguments. Why was that? I replaced it with NULL to match the other get_* functions, but I thought that there might have been a reason.
• I think aw_segments_table accepts NA values for segment IDs and then returns the metrics with no segment applied. This could be a handy feature -- maybe we should document that and return something better than we do now? I haven't experimented with this yet, though.
• aw_get_me, aw_call_data, and aw_call_api have a lot of duplicated code among them. Maybe we should pull out some of the functionality into a more general function that we can test? This also applies to a certain extent to parts of aw_workspace_report. This has nothing to do with the current PR, except that it relates to unit tests.

[benrwoodard]

I have been wanting to add unit tests to the package but obviously the api requests were hindering that. Now that you have got the development in a much more modular focus, this makes a lot of sense. Thank you again for the hard work. Let me look over this and get back with you on where I think it should live while developing it.

[charlie-gallagher]

@benrwoodard I've merged in changes from master, so that this has no conflicts any more.