the difference is that the final node (the composite in this case) is actually “viewable”. You can connect it to the viewer, and see what it is really going to output. Like at any other point of the tree.It behaves just like any other node.
One example of how it would work if blender worked the same way:
You load an image that’s 7000x5000px to the compositing tree and connect that directly to the viewer. Then it would show that to you, full format (as if it were a regular image viewer…bonus points if it were an EXR, it would give the option to choose which layer to see somewhere in the UI).
But if you connect that into the composite and feed it to the viewer instead (that’s what’s really lacking IMO), it would show you the same image, but now it would be cropped to the actual output of your composition (which would be taken from the render properties).
And that’s what, to me, seems the weirdest thing. That, just at the very end of the tree, there’s a crucial point in the composition that potencially makes big changes to the final image, and you cannot see what, or if, it does without rendering it. And what @izo is doing with the passepartout and such is very nice, seriously kudos to him, but still to me feels like a workaround for a lack of basic usability of one of the main compositing nodes, which is being able to view the composite output as if it were a regular node…because it’s there in the tree.