Description

I have been trying to write a multiprocessing function that works with headers and WCS, however I am running into an issue. I think the issue is with the method wcs.array_index_to_world, which returns something different to what I expect.

Expected behavior

No response

How to Reproduce

Here is a reduced version of the code I am running:

def example_func(header_str, dim_len):
      header = Header.fromstring(header_str,sep="\n")
      wcs = WCS(header)
      xy = wcs.array_index_to_world(0,np.arange(dim_len),0)


# multiprocessing part
dim_len = value
tasks = [(hdr.tostring(sep="\n",padding=False, endcard=True), dim_len) for hdr in hdrs]

with ProcessPoolExecutor() as executor:
        results = list(texecutor.map(example_func, *zip(*tasks)))

To use concurrent.ProcessPoolExecutor I am having to convert headers to strings (so they can be pickled) and then try rebuildling the header in the functions. The header is passed to WCS as I want to get spatial coordinates of the data.

The wcs object appears to be as it should (at least from a naiive print(wcs) inside the example function). However, after calling the method array_index_to_world I get a different result than when running the code without multiprocessing.

In single processing, xy has length 2 and xy[1] is a SkyCoord. The xy[0] is a SpectralCoord.

During multiprocessing, the variable xy is of length 3 and and the elements of the list are now Quantity.

The issue goes away if I swap concurrent.ProcessPoolExecutor for concurrent.ThreadPoolExecutor.

I don't really have a minimum working example of the issue, but happy to provide a fits file that this occurs on.

Versions

The version of astropy I am using is 6.17