This looks good to me and solves the issue.

Slightly related question, why do we write out these zero value nuclides at all. Would another approach be to not write out a nuclide if the value is zero. Are we reading in h5 cross sections that are not needed?

@shimwell That's a very good point. Perhaps this is something we should intercept at the time add_nuclide is called. That is, if you call mat.add_nuclide(nuc, num) and num is either 0 or a subnormal, you'll get a warning and the nuclide won't get added. In any case, I propose we handle that as a separate PR.

Would removal of these up front in add_nuclide potentially mess with depletion composition expectations (even if the composition components are neglibigle). I'm thinking of generating Material objects from depletion results as an example. I wonder if it might instead be better to catch this at the time we write the XML and display a single warning then?

@pshriwise yeah, there are possibly situations where it could cause weirdness, so I'd like us to think through them more (and as part of a separate issue/PR). The change here is strictly to ensure that OpenMC doesn't crash but otherwise doesn't meaningfully change current behavior.