I used perl and PDL heavily, before moving to Python and numpy. Both have annoying issues, and oddly, their warts are complementary. Particularly, the core API in PDL is miles better than numpy's. Before I could tolerate actually using numpy, I had to write a library to patch away numpy's warts, by effectively writing a PDL compatibility layer. Check it out:

https://github.com/dkogan/numpysane/

Now the core numpy has usable broadcasting, concatenation and basic linear algebra. Kudos to the PDL team for the excellent core design.