On August 14, 2017 8:00:19 PM EDT, Jed Brown <j...@jedbrown.org> wrote: >Tobin Isaac <tis...@cc.gatech.edu> writes: >> _Complex_I has problems if one of Re(z) and Imag(z) is intended to be >Inf or -Inf. For my purposes, I wanted to be able to define boxes in >the complex plane that generalize to infinite strips. > >Sounds like maybe complex numbers aren't a good way to represent this >construct (with multiple distinct infinities).
I'm only interested in tensor-product intervals not generalized circles in $\mathbb{C} \cup {\infty}$. Using complex numbers is simple (I can use `PetscScalar lo, hi;` for all builds) and it takes advantage of ieee 754 affinely extended real number line semantics instead of coming up with my own. I can't see something better. > >I don't have a great solution other than to just cast to a struct that >you can tamper with directly. Type-punning to a union is my fallback, which by my reading of the standards is valid in all _Complex implementations this far.