Albert Y. Shih

It's not being "incorrectly" triggered, per se. `reproject_and_coadd()` tries to determine the extent of each reprojected map in the output WCS by transforming just the corners of each original map...

I'm going to re-label this issue pending discussion to decide on whether this should be fixed internally or externally to SunPy.

There is a set of FITS files with SOHO orbit information, but with only 10-minute resolution: ftp://sohoftp.nascom.nasa.gov/pub/data/ancillary/orbit/predictive/

As an aside, MDI FITS headers *do* include observer location, although SunPy does not currently interpret this information correctly (see #3040).

I guess that's what we get for not actually having tests for #1750 (see #1774)...

A [recent mailing-list thread](https://groups.google.com/d/msg/sunpy/BqFosoHdYZ0/caDMv3jeAQAJ) relates to this issue. For completeness, I have also edited the initial post to link last year's mailing-list thread that prompted this issue.

So, our position should probably be to change nothing and continue to let SunPy to defer to Astropy's behavior on memory mapping. Note that: - Astropy has a configuration operation...

The latter uses FITS keywords such as BZERO et al. that specify how the stored values need to be modified to get to real values. Thus, it's not possible to...

The above hack should not be necessary now that module-level `__getattr__()` is supported in Python 3.7+

For past attempts, see #1736 and my comment there: _Originally posted by @ayshih in https://github.com/sunpy/sunpy/issues/1736#issuecomment-930185325_ > No progress, largely due to lack of interest. Separate from this PR, I drafted...