I didn't really make any changes on this ticket. However, while working on it, I fixed a couple of problems in meas_base. The only one of substance was a failure in one place to set the result object on failure. This is the change which ended up in
DM-1067 by accident.
As far as this issue is concerned, the slight differences in SdssShape are coming from a slight difference in SdssCentroid, and that ultimately from a very small difference in the convolution image being produced by the Psf. This is still a puzzle, but I was not able to go further without a more substantial amount of time.
So there are essentially no changes for this ticket, except what Paul has already reviewed with 1067.