Gregory Jefferis

Worked around by 69d3f2c736c3f42c1575bf0a459911e947ddf8d3 which simply skips test on CRAN.

Note that the windows and mac errors/warning are rgl related. So the digest mismatch is apparently due to a change in r-devel / r-patched.

A (possible) alternative is to set the `rgl.useNULL` option on package startup if we are in non-interactive mode.

But that would still not solve the issue that rgl is frequently not available in some macosx configs.

This solves the rgl load issue, but still leaves rgl as a dependency at build time, which can cause problems.

Is the origin defined in terms of pointno or index into vertices?

origin is defined in terms of pointno. Not sure how this compares with ```R as.directed.usingroot

This actually uncovers several issues. First `all.equal.im3d` was not checking attributes at all and second it was not checking the dim attribute by default. See #211 and #212.

Thanks so much @dokato. Can you remind me during the week to get the test data back up on flybrain? I didn't check yet if this is the only issue...

Hmm, well it does check it **if both objects have dim attribute**. Guess it should always check if one of them has a dim attribute.