XSpec model evaluation for non-continuous grids (Xspec models that call sumdem, such as apec, can crash) about sherpa HOT 5 CLOSED

As an example of how the current scheme can give surprising results, I show the model evaluation for the xskerrdisk model using a continuous grid (which has then been filtered to remove points) and the equivalent results from using a non-continuous grid: https://gist.github.com/DougBurke/d9a0074489b6d1de108e

This is one of the more obvious differences (and is down to how the model works, since it really relies on having a large grid of input values). More common is that the bins that are at the end of a contiguous sequence show differences but the other bins have the same result.

from sherpa.

Just to note that Keith Arnaud tells me that this crash is fixed in the not-yet-released next release of XSpec.

from sherpa.

Any chances we could get a patch for testing?

from sherpa.

Keith is going to be sending me the code soon. I'm also working at changing xspec_extension.hh to avoid this problem (although the simplest fix has lead to more strange bugs, so something is rotten in the state of Denmark): https://github.com/DougBurke/sherpa/tree/xspec-contiguous

from sherpa.

I have also come across another bug in the non-contiguous case, this time when the input array is in Angstroms. The error is shown in https://gist.github.com/DougBurke/b70485a9280f1b52a83e

from sherpa.