Currently, users choose whether to make PSF-Matched or default (Non-PSF-Matched) Temp Exps and Coadds by flipping the boolean config: CoaddBaseConfig.doPsfMatch. To create both, users must run makeCoaddTempExp.py twice and assembleCoadd.py twice. This warps the calexps (a computationally expensive operation) twice. In this RFC, I propose to change the behavior and interface of MakeCoaddTempExpTask to create both types of coadds simultaneously by default, and optionally one or the other per a command-line config. By default MakeCoaddTempExpTask will hold two TempExps in memory, each calexp will be read in, warped, placed into image #1, then matched and placed into image #2.

See github https://github.com/lsst/pipe_tasks/compare/tickets/DM-8491 for illustration. (This is not a PR)
You may ask, why make the warps and psf-matched warps per calexp in simultaneously instead of in serial? One can imagine making a complete set of non-matched temp exps, and then the psf-matching these temp exps. This would be nice because a user could change their mind about the FWHM of the model PSF and re-match the warps for half the cost, but this isn’t feasible at the moment because: (1) Placement of calexps on the patch can cut off too much of the original calexp to retain a sufficient number of kernel candidates for accurate PSF-Matching (2) PSFs in the warps are discontinuous at chip boundaries. It is technically feasible to store enough information about the PSF models in the warps and make PSF models with known (not-rectangular) discontinuities, but this complex functionality does not yet exist in the stack.

Comments welcome on:

1. Config names and default behavior for makeCoaddTempExpTask and assembleCoaddTask:
   • Default behavior change: make both psfMatched and non-Psf-Matched Coadds by default and use two new configs:
     • CoaddBaseConfig.makeOnlyPsfMatched=False
     • CoaddBaseConfig.makeOnlyDirect=False
       With these configs, current default behavior of just creating non-matched coadds would be replicated with setting CoaddBaseConfig.makeOnlyDirect=True.
2. Name of new data products:
   • deepCoadd_tempExpPsfMatched
   • deepCoaddPsfMatched
3. New interface of WarpAndPsfMatch.run(self, exposure, wcs, modelPsf=None, maxBBox=None, destBBox=None) which currently returns a regular temp exp if modelPsf is None and a matched temp Exp if not None. If we need WarpAndPsfMatch to keep its original responsibilities, I propose returning a Struct from WarpAndPsfMatch with a warped and matched calexp called “exposurePsfMatched” and a warped exposure called “exposure.” However, WarpAndPsfMatch is so minimal I don’t know why it needs to be its own task… No tasks other than makeCoaddTempExp depend on it, and warping and psfMatching could easily be called directly from makeCoaddTempExpTask.
4. How will this affect supertasks or pipe_drivers?

Paul Price, you mentioned you had plans to refactor assembleCoadd and makeCoaddTempExp soon. I’d like to hear what you were planning to make sure these changes fit within your vision.