I can also confirm that the included patch does "workaround" the issue.

A big pain point here for me is in determining what exit handler is causing this.

GDB has been it's usual great help in determining information about the exact segfault, but the problem code itself is rather elusive since it's ran so much earlier and just causes this issue downstream later in execution. Unfortunately the extensions I'm dealing with also happen to take a lifetime to compile, exacerbated by the fact that they are dependent upon one another.

Can we please apply this to 3.x's tree?