It is possible to construct a scenario where two sibling encrypted cells could have the same prefix, but due to the wide allowance in naming conventions, could in fact be different.

I've been trying to think of an example of this, but struggling to. I know that the logical netlist may have hierarchical names where foo/bar/baz and foo/"bar/baz" can possibly both exist and be differentiated in the logical netlist (untested) but it's even more unclear how they can both exist in the flattened physical netlist.

If you've got an example handy please send it my way!

If you've got an example handy please send it my way!

Suppose you want to black box the instance foo/bar/baz and you have a cell called foo/"bar/baz"/LUT5. By looking at the physical name only, there is no way to know that the LUT5 is not part of foo/bar/baz because of the name aliasing and would get unplaced. With the logical netlist, we do have a chance of detecting this.

Suppose you want to black box the instance foo/bar/baz and you have a cell called foo/"bar/baz"/LUT5. By looking at the physical name only, there is no way to know that the LUT5 is not part of foo/bar/baz because of the name aliasing and would get unplaced. With the logical netlist, we do have a chance of detecting this.

I think I'm following. Consider two groups of leaf cells, each under a legitimate logical hierarchy:

foo/bar/baz/a
foo/bar/baz/b
foo/bar/baz/c

foo/"bar/baz"/d
foo/"bar/baz"/e
foo/"bar/baz"/f

Now these logical leaf cells lead to unique physical cell names. But I wonder what happens if the cell names collide? (i.e. {d,e,f} -> {a,b,c}). I'll have to test to see at which point Vivado starts refusing.