I’ve occassionaly tried combining multiple bump maps by plugging one into the normal input of the other. Sometimes it works. Is this something I should not be doing? What is it really used for?
No, you should chain them together. The normal input is if you want to mix a bump map and a normal map. To use two bump map you could just mix them with a mixrgb node and then use one bump node, but you could use two bump nodes with the output of one of them (normal output) goes into the normal input of the second bump node.
You can connect another Bump or Normal Map node.
You can chain. It’s different than adding:
Two triangle waves. Those two waves, added together (displace modifiers in global Z). Those two waves, chained together (displace modifiers in normal). A plane, copying the normals of the last one on projected face interpolated.
Any time you want, you can do a displacement test with a bump map to see what it’s doing. Bump map is just like a flat subdivided plane, copying the normals of a mesh with a displacement map.
In general, I’d say just adding two bump maps is less prone to problems. See how the chained ones create discontinuities? Note also that if you chain, the order that you chain matters.
I’m often chaining to have big bumps go to both substrate and topcoat, and small bumps only go to the substrate where the topcoat has none, reduced, or separate bumps. But as bandages say, it is wise to test the bumps by temporarily try it out as a microdisplacement to verify directions are okay and scales are okay. Using bump strenght 1, bump distance should be set to the same value as microdisplacement scale, and the produced normals should be the same. Curvature on bumps can be tougher to control though. The best look on bumps is generally obtained if you have a continuous change in curvature rather than sharp changes or long straights of angles.