64 bit IRAF

[karlglazebrook]

Re https://noirlab.edu/science/news/announcements/sci20062

I just wanted to let you know that is IS possible to run IRAF natively on M1 Macs:

https://twitter.com/olestreicher/status/1392091865411989511?lang=en

I downloaded this on mine, and it worked fine.

STSDAS doesn't work, I don't know if the Gemini part requires this or other 32 bit stuff but you might want to look in to it.

You should also consider a Docker solution.

[jehturner]

Hi Karl,

Yes, we're aware that Ole's "IRAF community" version or IRAF has an arm64 port. Unfortunately Gemini IRAF does have 32-bit dependencies that have not been ported, including STSDAS, some of our own code and the version of TABLES that passes our testing. It was decided some years ago not to port those things because of the significant and uncertain effort involved and the pressing need to concentrate our limited resources on moving to Python. Obviously the situation on Apple has deteriorated since then, but our plan is still to support IRAF there using the VM and add instrument modes to DRAGONS as quickly as feasible.

We have tried running IRAF with Docker, but there were 2 major problems with it -- one was Docker's overlay filesystem using inode numbers that are too large for 32-bit IRAF's file pointers and the other was Docker triggering corruption of IRAF's memory management on M1. That second thing wasn't terribly well understood, but it's likely to be a rabbit hole and there are reports suggesting that it could be a much more general problem with Docker's support for running Intel binaries on ARM, which they explicitly don't plan to fix. In any case, Docker is mostly serving as a wrapper for qemu when trying to run Astroconda IRAF on M1 and we have had much more success using qemu directly -- which is what we hope to publish before long -- so there was no point in expending a lot of effort on trying to make Docker work. In either case, of course, there is a significant slow-down due to emulation, depending on what processing you're doing.

We are, however, collaborating with Ole's related effort to make PyRAF work better on Python 3, so I hope we may finally be able to ditch Python 2 this year (that's not official for now).

Thanks,

James.