[BUG]: When nonsensical kernels like f770w_to_f770w or f2100w_to_gauss0p35 are missing, psf_matching crashes and deletes the whole house

[akleroy]

Crashes out on poorly formed band combinations during the convolution post-processing steps. @rchwn has a fix I think - either way we should add a small amount of error checking.

[akleroy]

@rchwn When you are back online it'd be good to get this in. I have temporarily hacked my version near line 170 in psf_matching_step so that it does not raise an error and returns True if it cannot find the relevant kernel during psf matching.

Since failure in the step triggers removal of the band directory, this is a pretty severe issue. I think the step must only ever have been run with resolutions >~ 1" that can be reached by all bands.

[akleroy]

@thomaswilliamsastro - sorry annoyingly poking on this

I think the issue is addressed by just going to around line 167 of psf_matching_step and switching the FileNotFoundError to instead just print a warning and return True (i.e., that the step was successful [enough]).

Missing kernels are going to happen unless we use target PSFs that EVERY band can reach and the current state where things crash out and return False leads to a catastrophic failure at this step where everything gets wiped.

This seems really easy - I can work on a PR for it next week but maybe it's possible to just tweak this on your end?

[akleroy]

So @rchwn the proposed fix still crashes in the case that no convolutions are done because the attempt to write psf_matching_step_complete.txt fails - no directory has been created and there's no .txt file.

[akleroy]

The test here is: does it run and complete when one of the bands has NO viable kernels and/or no input data.