• try to deal with non-estimable coefficients in `model': if the correspondings columns of `linfct' are zero anyway, we silently proceed. If not, we stop and ask for a correctly specified model.
• try to detect interactions with numeric covariates. We need to choose a specific value here but it seems hard to add something appropriate to the user interface (I don't want to touch `mcp'). For the time being, we fire a warning and the user can modify `linfct' manually.
• add mcp( = "Means"), contribution by Richard M. Heiberger
• add Rich to DESCRIPTION file