> Yes, checking for -nan seems to be the best choice we have. So we must
> replace (thing-at-point 'number) with something else in
proced-refine-test,
> proced-refine-with-update-test,
proced--assert-process-valid-cpu-refinement
> and proced--assert-process-valid-cpu-refinement-explainer. Would you
> like to provide a patch?

I was thinking more along the lines of a 'isnan' check in 'proced-<' to the
effect of making it not appear in any refinements at all, but if you think
it's too much fretting about an edge case, I'm happy to provide a patch for
the tests only (as you describe)?

> (see sysdep.c). If one of the operands is not proper, the result can be
> a NaN indeed, like -0.0e+NaN.

Could you explain what is meant by proper?