I wonder whether this will work on all platforms - for instance Windows or 32-bit platforms? What is the usual way to deal with this @embray @mdboom?

Well, I didn't think about it :( Still have to learn to think in the bigger picture.
I've tested it only locally with python2.7 and 3.3, but should have checked on our obsolete server (64bit) with python 2.6.4, which gives the original format.

Sorry for the noise, now I doubt it's the right approach.

@bsipocz - it's a corner case, so not obvious - and in fact, it may not actually be an issue. I just seem to remember things like this causing issues in very specific instances in the past.

We could just use an ellipsis instead, i.e.:

array([1, 4]...)

Sorry I should have mentioned sooner (I was doing this on Friday though and ran out the door before I had a chance) but I actually already fixed this in a branch where I was fixing Windows test failures.

There's a simple workaround for this particular case to just use print instead--then Numpy won't add the dtype to the output either way.

Meanwhile, I'm working on a separate PR (won't be ready for 0.4 though) that adds support for environment markers to the .. doctest-skip directive (for example one can write .. doctest-skip:: platform == 'win32')

So just so we're clear, @embray, you have a branch/soon to be issued PR that addresses this, or should @bsipocz update this now?

Either way really. If this is updated and merged first then it doesn't matter.

Though if so I would just rebase this so that there's a just a single commit using the workaround I suggested, if at all.

Done.

Thanks @bsipocz!

This affects v0.3.x too so I changed the milestone.