Coverage remained the same at 74.951% when pulling 71fb3d0 on rt-107326 into 19d0dfe on master.

Thanks for adding the test cases! The Travis build didn't seem to have run the cdf test. It should have failed given the new test. Closing the PR without merging for now as it'll break the master build unless the test is fixed or ignored.

I'm going to reopen since this is still a live issue, and this being open indicates that.

Coverage remained the same at 74.951% when pulling 71fb3d0 on rt-107326 into 19d0dfe on master.

Most immediately, this is no longer an issue for this repo since the CDF functions are now in main PDL.

Going slightly deeper, this is an example of PDL working as designed; the inputs for functions including gsl_cdf_hypergeometric_P which take unsigned values are having the inputs coerced to unsigned values. For -1 this gets changed to 65535, which obviously is not what the user expects and is quite unhelpful.

For reasons of backward compatibility, we can't change the behaviour for negative inputs on conversion to unsigned. Therefore, unfortunately there's nothing to be done in this case. Sorry to have taken so long to dig into this!