Merge lp://qastaging/~dooferlad/linaro-image-tools/copy-files-support into lp://qastaging/linaro-image-tools/11.11

Line #359 and on is a bit cryptic - instead of using unobvious numeric indexes targetting at the unneeded for tests feature of being able to change "foo" and "bar" in one place, it would be better to use "foo" and "bar" literally as needed - the test would become more clear and actually easier to maintain and extend.

More importantly, this has a regression wrt to original test which I wrote - which check that even if we have 2 different packages with the same layout (i.e. files with the same names at the same paths, but with different content), then exactly the file from expected package is used. This actually hit me due to bug in file expand code, and of course, it may hit again.

With new implementation, the problem is in line 266, which writes the same mock content to files on the same package, so it's not possible to distinguish 2 such files from different packages. I suggest letting user pass the content explicitly, then it can be easily specified as:

package_files = {
"foo": [("usr/lib/u-boot/omap4_panda/u-boot.img", "mock uboot #1"), ("usr/share/doc/u-boot-linaro-omap4-panda/copyright", "mock copyright #1")]
...
}

Yea, the "foo", "bar" thing has been happening a lot in other tests in the suite and I was sticking with the style. I agree that it is easier to read with the strings used instead of a reference, but I was trying not to repeat myself. I don't have a strong opinion about it, so I am happy to change it.

Actually each mock file has the originating package and path written to it, so I can add more checks. I thought I had mirrored your original test - sorry if I missed something. Since the logic is there, I might as well use it for all the files though, so I will do that.

Thanks for the review.

Oops, spoke too soon. I am testing the origin of all the files (I just have some duplicated tests, which is pointless). Will take out the duplicates and use explicit package names.

Ok, I got confused with workflow of code (again, how much easier it would be to just look at *exact* packages being used), it appears that content will be unique already if package names are unique.