The above architecture is quite complex, as the inputs supplied can lie anywhere on the image plane, as can the output, and they can be subsections of the 'complete' image. Not all hosts work in this way, generally it is only the more advance compositing systems working on large resolution images.
Some other systems allow for images to be anywhere on the image plane, but always pass around full RoD images, never tiles.
The simplest systems, don't have any of of the above complexity. The RoDs, RoIs, images and project sizes in such systems are exactly the same, always. Often these are editting, as opposed to compositing, systems.
Similarly, some plugin effects cannot handle sub RoD images, or even images not rooted at the origin.
The OFX architecture is meant to support all of them. Assuming a plugin supports the most general architecture, it will trivially run on hosts with simpler architectures. However, if a plugin does not support tiled, or arbitarily positioned images, they may not run cleanly on hosts that expect them to do so.
To this end, two properties are provided that flag the capabilities of a plugin or host...
A plug-in should flag these appropriately, so that hosts know how to deal with the effect. A host can either choose to refuse to load a plugin, or, preferentially, pad images with an appropriate amount of black/transparent pixels to enable them to work.
is redundant for plugins that do not support tiled rendering, as the plugin is asking that it be given the full Region of Definition of all its inputs. A host may have difficulty doing this (for example with an input that is attached to an effect that can create infinite images such as a random noise generator), if so, it should clamp images to some a size in some manner.
The RoD/RoI actions are potentially redundant on simpler hosts. For example fixed frame size hosts. If a host has no need to call these actions, it simply should not.