Looks like the initial empirical rule "mixin must not depend
on another mixin" is too restrictive for practical purposes
given enough image targets multiplied by enough platforms;
let's declare it obsolete and see what follows.