The point of conflict there is that the landuse key indicates either land use or land cover, because attempts to split out a separate landcover key haven’t quite succeeded yet. This is on top of the conflict between using landuse areas to paint broad swaths of the map that are contiguous by happenstance, versus using the same key to indicate the land use of a specific site.
The broad-brush approach is entrenched in large part because renderers treat landuse as a thematic layer. If we were to rationalize landuse tagging to focus more explicitly on answering “What is it?”, then landuse=industrial would be replaced by a number of tags under man_made=* and a few other yet-to-be-coined keys. Former landuse=industrial areas would have quite different tagging than former landuse=commercial areas. This would be frustrating for traditional OSM rendering stacks as well as anyone coming from a GIS background.
On the other hand, the more granular land use areas often represent hard-won local knowledge about both the land use and the edges of that land use, which can’t easily be detected from aerial imagery using machine learning. The broad-brush land use mapping should never come at the expense of this valuable data.
I partially agree. If I map a named development that’s already covered by a broad-brush land use area, I go out of my way to cut out a piece of the existing land use area to avoid overlap.
On the other hand, if I’m mapping some even more granular land use within a development, such as a golf course, waste dump, or residential subdivision on a military base, I won’t cut it out of the military base, because some overlap is fine in these cases. Think of the smaller land use area as an exception to the larger land use area. Renderers can follow the example set by osm-carto, which automatically prioritizes rendering the smaller area over the larger areas.