I think that the error is also incorrect (np.std(dmag)) I think that it should be something like :
zp, sum = np.average(dmag, weights=1/dmagErr**2,returned=True)
sigma = np.sqrt(1.0/sum)

Assuming that the dmagErr values are correct

I'm just adopting Dominique Boutigny's proposed solution, as it seems correct, or at least vastly more correct than what we have now. I do wonder a bit whether we should be doing this average in flux units rather than magnitude units, but I think that's beyond the scope of this issue.

Robert Lupton, by "soften the errors", do you just mean "add a constant in quadrature" (with the constant coming from the config)?

Probably not exactly in the scope of this ticket, but the main problem is that the Photocal fit is not taking into account the error on the flux of the reference stars. In some region of the SDSS catalog (CFHTLS D3 for instance) they can be quite large.

This issue will address some (but not all) of the problems discussed in DM-2308.

Simon, do you have time for a small review? This fixes a big but simple problem in the Photocal algorithm, along with a number of minor fixes and cleanups. All changes on tickets/DM-2423 of pipe_tasks.

Simon is super busy. Russell Owen, could you take on this review? It's a pretty small one.

photoCal.py

The changes look helpful.

One small request: please change the final return to specify each value on its own line, as per the original code. I feel that makes it easier for a user to see exactly what named values are in the returned struct.

testPhotocal.py

The changes look helpful.

One small request: please change the final return to specify each value on its own line, as per the original code. I feel that makes it easier for a user to see exactly what named values are in the returned struct.

I find that a bit unnecessarily verbose, but I also don't care much. Will do.

Merged to master.

Yes.