If any of the shapes participating in the subtraction belong to a hierarchy of a parent with padding, the padding manifests in the result, which can result in unexpected behavior from the user point of view. It is not completely clear to me if this is a bug.

Taken from StackOverflow: https://stackoverflow.com/questions/51599995/why-does-the-resulting-shape-from-shape-subtract-appear-at-an-incorrect-positi

The answer there includes a reproducer code snippet and analysis. Comment out and uncomment the critical lines according to the description to reproduce all cases.

Other Shape methods are probably also affected, though I didn't test.