scipy.linalg.qr with pivoting=True should return P as (N,N) permutation array instead of (N,) array

I am a bit surprised why scipy.linalg.qr with pivoting=True returns P as an array of shape (N,) instead of a proper two dimensional array of shape (N,N) just like scipy.linalg.lu does. This way I could quickly test if A = P.T@Q@R just like I can test A=P@L@U. Is there any reason for this inconsistency?

[ilayn]

I’d argue in this case that LU is the odd one since it is a waste of n**2 array just to provide sorting keys. I see how this can be unintuitive when you have plenty of memory but nevertheless having an array of n instead of n**2 is efficiency wise should be prefferred. I agree nevertheless we have a historical inconsistency across linalg functions.

[ilayn]

I am still considering this since most software has the option Outputform as a switch between 1d perm vs 2d full permutation array. It is not a very difficult thing to implement but what we will definitely not do is to choose one that would break code.