You can clone with
HTTPS or Subversion.
On my computer this leaks memory on numpy 1.7.0 but not 1.6.2:
In : a = np.ones((500,500))
In : for i in range(10000000): a.diagonal()
Functions that call a.diagonal, such as np.diag() and np.trace(), will also leak.
Known problem, should be fixed in 1.7.1. Can you give it a try?
Sorry about that. Yep, no leak in numpy 1.7.1.