This stems from a call to getPsf().computeShape() without specifying a position, so the default and highly dangerous position is used, see RFC-796 (I believe this is happening near the edge of the coadd footprint, so there are many “template” images that will have very few actual good science pixels and whatever is being returned by the default position is an area where there is no real data). It seems we have gotten lucky so far with the full RC2/DC2 runs, but this subset has revealed that this is an issue that warrants fixing.
There are a few offenders in this file:
here and here (and, with the deprecation warning added on
DM-31535, a recent processing run will reveal just how many such cases we have with the message:
FutureWarning: Default position argument overload is deprecated and will be removed in version 24.0. Please explicitly specify a position.)