https://bugs.winehq.org/show_bug.cgi?id=41930
--- Comment #27 from Bryan Varner bryan@varnernet.com --- Ok! So the mesa headers are out of date with the implementation.
Looking into the actual osmesa.c source - it looks like it's saying it'll wrap the the actual buffer created to CHAN_BITS size with one that reports / functions as the type specified.
Perhaps I should do some more debugging here -- I'm wondering if osmesa is returning a GL_FALSE (resulting in the black terrain) or if there's something else amiss here.