build_sf_inset_layers() decides whether or not to add an inset layer based onis.null(inset). This is wrong, because layers are supposed to default toinset = NA so that it's controlled by the coord. This doesn't work though since the layer helper is trying to make the decision at plot configuration time, whereas the inset config in the coord won't be known until too late, at build time. The result is that the default creates an incorrect duplicate copy of the sf layer (seecidm-ph/ggautomap#5).

It seems like the proper fix would be to emit both layers in the layer helper when using theauto mode, and have that secondary layer drawn nothing in the case that there is no inset config in the coord.