@seberg needed your opinion here
For errors like these (mentioned in numpy/numpy-quaddtype#15 )

>>> from numpy_quaddtype import *
>>> QuadPrecision(1).imag
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[2], line 1
----> 1 QuadPrecision(1).imag

RuntimeError: Using legacy SETITEM with NULL array object is only supported for basic NumPy DTypes.

Will it be fine to approach these by setting the getters as tp_getset property (done in this PR)? or Ideally it should be integrated within Numpy's calling mechanism?

This makes sense, but for array support we need something more. I am interested in that too (soon), though. I suspect an ArrayMethod or even ufunc makes sense, but for ufunc need to add view support there (should be possible, but not sure it's worth it).

Nice, yeah make sense, will be happy to extend this in NumPy whenever you feel right!

Merging this in

Do we have tests for this?

One thing that might help for ensuring new code has tests is to look at integrating coverage testing. That way you would have gotten a test failure here because the coverage went down.

One thing that might help for ensuring new code has tests is to look at integrating coverage testing. That way you would have gotten a test failure here because the coverage went down.

I know coverage.py does something like this (never used it though) is that what you recommending?

To get coverage of native code you need to use a native coverage tool like gcov or llvm-cov. But yeah, something like that. There's also codecov.io, which has some fancier features to display the results and add diagnostics to PRs.