Martin Hardcastle

First one is fine — consistent with the fact that the problem doesn't appear with reproject_exact(). I'll report back on the second when it's finished. I can arrange an account...

The second code fragment also runs fine (i.e. a scaled version of the 1024x1024 appears in the lower left quadrant of the output array). I'll see if I can see...

Well, that's remarkable: image sizes of 2^15 x 2^15 demonstrate the problem with my test code above but an image of 2^15-1 on a side does not. So I guess...

Linux: Scientific Linux 7.

I haven't had a spare moment, but I haven't forgotten -- I need to get this resolved to make progress on another project (or give up and go back to...

(And yes, chunking would be very useful. Particularly if that allowed parallel mode to work in reproject_exact() -- it crashes on these big images presumably because it runs out of...

We have been using the chunking workround discussed in 2016 in production since then: see https://github.com/mhardcastle/ddf-pipeline/blob/master/utils/reproj_test.py

This is true, but in a sense ds9's fault: as ds9 does not put Stokes or frequency values into its regions even when applied to a cube, pyregion is necessarily...

I will try to test this and document it. But it might be even better if this were the default behaviour in pyregion, since ds9 regions are always spatial, and...

Documented, pull request raised.