+1

I still need to test this on Linux.

So to make this work I think we need to make a version of loadLibrary which is intelligent and can try to load the requested file with different suffixes and prefixes. We can avoid this for now by duplicating this logic everywhere we use class loader, but I think it makes sense to embed this functionality in class loader as we write more plugin based software that needs to work on Linux, OS X, and Windows.

@wjwwood sounds good to me.

Are you referring to the lib prefix and the platform specific extension? If yes, I guess it doesn't need to try different ones but only one which can be determined by ifdefs at compile time. Then I would suggest to implement this directly in this repo instead of copying the same code in multiple locations where this is being used.

@dirk-thomas it's a little more complicated than that because you can drop the lib on Linux and OS X if you want and you can pick the wrong extension (it is common for me to find people generate a .so file even on OS X where it should be .dylib). There's probably two functions that I could image. One just takes the "base" name and adds lib on Linux and OS X and the appropriate extension on all platforms (basically what I did in 8e9ed35). Another function might try a little harder, looking for .so on OS X if .dylib isn't found, trying without lib, etc.

I'm going to add one or both of these functions in a separate pr, are there any objections to merging this?

One thing I didn't mention yet was that I had to link pthread against the gtests. I think this might should be fixed upstream, but I don't know which package it should go in.

Beside the small comments: lgtm

We should probably also consider which changes on the ros2 branch could / should be back ported. (Not in this PR though)