Procedural Generation - Additive or Subtractive?

So, I'm working on scripting a procedurally generated 2D labyrinth. I've considered both of the following options, but am curious which would be the most efficient from a resource perspective. Any insight would be tremendously helpful.

Option #1 - Generate an area, subtract backwards from an endpoint to the startpoint and wall off areas accordingly.

Option #2 - Generate the successful path, adding iterative paths for additional dimension, and build additively throughout.


In other words, in option one it would be similar to starting with a solid block of stone and carving the path from end to start, with some additional side paths to flesh it out.

The second option would be to start with a blank slate (a void), generate the solution path, build out form there, then enclose it.

It seems like the second option would allow for more uniqueness (not being generated within a particular boundary of coordinates). But the first seems like, because of having boundaries, it may keep constraints more regular and be less resource intensive.

Any thoughts?

 

Frankly, I'm not seeing the difference between the two, because in both case, you won't generate a mesh out of that; that would be very expensive for very little result.

Option 1 or 2, you'll probably end up instantiating prefabs of corridor pieces. In both case, you generate a right path, then fill the missing space with false paths. It will always end up as values in a 2D or 3D tables, therefore limited by that table's size.