object with 'dtype' property crashes numpy

[mwiebe]

The following code crashes numpy (latest master):

import numpy as np
class crashnumpy(object):
    @property
    def dtype(self):
        return self
np.array([1], dtype=crashnumpy())

[njsmith]

I guess this is triggering an infinite recursion in C code?

In principle we can fix such things, but it takes a bunch of work on a
case-by-case basis, and people who set out to break things have many many
other options using numpy. (We provide free access to arbitrary memory
locations, after all.) Is there some reason this bug is a problem, or is it
just a curiosity?

On Tue, Aug 13, 2013 at 11:36 PM, Mark notifications@github.com wrote:

The following code crashes numpy (latest master):

import numpy as np
class crashnumpy(object):
@Property
def dtype(self):
return self
np.array([1], dtype=crashnumpy())

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/3614
.

[mwiebe]

This came up when I accidentally passed a dynd type to numpy. The dynd type includes dimensions, and its dtype attribute returns the type with these dimensions stripped away. If the type has no dimensions, then it returns itself, leading to a behavior similar to the crashnumpy class.

[OmerJog]

was this fixed?

[seberg]

No, gh-13003 includes this fix, but I was travelling and on a small conference now, so I did not finish it up. I guess we will probably have it fixed soon. Does this affect you much?

[seberg]

Oh, forgot to close this one. It raises a maximum recursion depth error right now. There is the next step of deprecating parts of it, but there was even another issue about it IIRC.