BUG: quaddtype: Don't dereference pointer before knowing it is not NULL.

[WarrenWeckesser]

Closes gh-166.

[WarrenWeckesser]

Is there a test I can add that would result in given_descrs[1] being NULL?

[seberg]

You could try quaddtype_arr.astype(QuadDType) (the class), but since that is already the same dtype... it probably doesn't trigger it and it is impossible.
To be honest, I am not even sure if this is easy to trigger for any non-paramtric dtype at all (although I suspect there are paths).

Either way, I would say its hard and maybe impossible. It's long enough ago that I wrote that code that I am not 100%, but...

[WarrenWeckesser]

You could try quaddtype_arr.astype(QuadDType) (the class), but since that is already the same dtype... it probably doesn't trigger it ...

You're right, it doesn't result in given_descrs[1] being NULL.

[SwayamInSync]

Thanks @WarrenWeckesser the patch make sense, but I'm not sure either when it can be NULL. I tried even passing None but it seems a default resolve descr runs in such cases.
Nvm merging this.

[seberg]

I had a random thought just now that if you make the original array a structured dtype that contains the QuadDType and then cast it to QuadDType that may bypass some fast logic and go deeper.
But I don't think it's worth pursuing, just a bit of trivia :).