Data Driven Model for Merge Tools

This is just an idea to make adding new merge tools simpler. A lot of the merge tools function quite similarly and their implementations in ApprovalTests each require the creation of at least on class per merge tool. However, given multiple versions, names, or operating systems, the number of classes required increases. This amounts to quite a bit of boilerplate when adding new merge tools. By defining a struct for storing all of the necessary data, construction of one-off classes can be avoided, and this struct can provide all of the necessary details required to find and call a diff tool from within ApprovalTests.

An example struct as follows hopefully encapsulates most of the variance between merge tools.

#include <string>
#include <vector>

enum Context {
  Background = 0;
  Foreground
};

struct DiffTool {
  std::string name;
  std::vector<std::string> search_paths;
  std::vector<std::string> names;
  std::string arguments;
  Context = Context::Background;
  unsigned int priority = 0;
};

By defining a standard search procedure for the search_paths and names, each tool can be found using the same process. The search_paths and names members would likely need to be defined in a way to only include certain values dependent on the OS.

I'm not sure how viable this is, but I wanted to get the idea out there. It would be nice to only have to define most merge tools in one place.

Note: Multiple versions of the same tool would still require special handling.

Start by reviewing the existing merge-tool implementations in ApprovalTests.cpp and how they locate tools across Linux, Mac, and Windows. Compare the repeated data and special handling for multiple versions, then determine whether a shared data-driven design can cover the documented cases without losing platform-specific behavior.