Performance

[Manfred]

The guiding performance principle behind Reynard is that Ruby generally performs best when you keep the number of object allocations as low as possible. Based on this principle it keeps the parsed JSON structure in memory and used an interface that resembles Hash#dig to get to objects within the structure. Results from these operations generally keep a path (eg. ['authors', 0, 'name']) to the result instead of keeping an actual value of the result.

It would be good to assess these assumptions. We could try what happens when you store sections of the specification instead of a path. We can also see if there are hot spots in the implementation that could be removed.

[Manfred]

One thing to consider is that the dig implementation makes a clear interface for traversing between files, while consolidating a large structure of the spec in memory might not. Another consideration is that a number of Reynard instances will only access a small number of operations.