@SpecLad please look at it. I think it is ok to merge the fix after release.

You're trading initialization order problems for thread safety problems. I'm not so sure this is a good tradeoff.

There are other global MatOp objects besides g_MatOp_Initializer. Wouldn't they need to be fixed in the same way?

Roman,

I have assigned the PR to you.

@SpecLad, the patch is harmless. The MatOp_Initializer class does not have any mutable members (mutable in generic, not C++ sense), so there are no thead-safety problems. I think, we should merge the patch. If something else needs to be done with other MatOp objects, that can be done in further PR. I do not see any problem with solving the problems step by step.

@SpecLad Does this make sense?

Yeah, I guess it's okay.

Wow, so fast, thanks!

Yeah, only took, like, what, 3 months? 😃