GalSim 1.5 is incompatible with the LSST Data Management stack

[danielsf]

A few weeks ago, I tried to naively update the version of GalSim being shipped with the LSST stack. This resulted in one of our GalSim-dependent packages breaking with the following errors

include/lsst/meas/extensions/shapeHSM/HsmAdapter.h:40:69: error: no matching function for call to 'galsim::ImageView<float>::ImageView(float*&, const boost::shared_ptr<float>&, const int&, const galsim::Bounds<int>&)'

include/lsst/meas/extensions/shapeHSM/HsmAdapter.h:40:69: error: no matching function for call to 'galsim::ImageView<double>::ImageView(double*&, const boost::shared_ptr<double>&, const int&, const galsim::Bounds<int>&)'

include/lsst/meas/extensions/shapeHSM/HsmAdapter.h:40:69: error: no matching function for call to 'galsim::ImageView<int>::ImageView(int*&, const boost::shared_ptr<int>&, const int&, const galsim::Bounds<int>&)'

include/lsst/meas/extensions/shapeHSM/HsmAdapter.h:40:69: error: no matching function for call to 'galsim::ImageView<int>::ImageView(int*&, const boost::shared_ptr<int>&, const int&, const galsim::Bounds<int>&)'

Technically, this is our (LSST's) problem, but none of the relevant developers have time to devote to fixing it, and that is going to interfere with DESC's ability to run ImSim with an updated GalSim on the upcoming data challenge. Would it be possible to restore the deprecated call-signatures in GalSim to preserver compatibility with the LSST stack, or is that a bad idea?

Thanks.

[rearmstr]

@danielsf actually, I was hoping to fix this problem in the next couple of days. I've added a new feature to galsim that will be merged into a tagged release soon that we want to use in LSST. I haven't looked at the problem yet, but Paul Price said he thought it was a simple fix.

[danielsf]

@rearmstr Oh. Good. Please ping me when that gets merged. There is already a branch on the LSST GalSim repo which updates GalSim (it was merged to master and then rewound when meas_extensions_shapeHSM broke)

https://github.com/lsst/galsim/tree/SIM-2550/galsim_v1.5

[rmjarvis]

I don't think we want to add back the old version, which is equivalent to the current one with step==1. But the fix in your client code is easy. Add 1 as a new third parameter. The new version doesn't require that the rows are contiguous in memory, but if they are, then step = 1 is the correct value to use.

Also, note that GalSim is officially a python package, not a C++ library, so code signatures are not guaranteed to be consistent across minor version updates. If you plan to use a lot of GalSim back-end code from C++, you may have more work to do on future versions as well.