> Hmmm. It'll be difficult to review such patch without access to a platform > exhibiting such behavior ... IIRC IBM offers free access to open-source > devs, I wonder if that would be a way.
I don't have access to such platform either, and I don't know too much about this business. I left this patch out for now. It is something that should affect all float operations not only the geometric types anyway. Anybody who knows and cares about these platforms can pick this up. We can fix -0 issues in localized way, like you have done, if there would be any more.
