For this test, I think build of each leaf_image* should be adjusted instead of changing this expected result, because those existing NVRs are not actual NVRs. For example, leaf-image-4 could be changed to leaf-image-4-1.0-1 and etc.

Another blank line is required.

Would it be easy to understand with just x and y?

Through this method, name image_id and parent_id are only used to locate an image in the batch to_rebuild. Maybe simple name x and y would be more straightforward to represent the intention. Especially for the line nvr_to_coordinates[nvr] = [image_id, parent_id].

There is no problem with using image_id and parent_id. I just would like to share my thought on the naming. :)

Remove this continue?

Why there could be a s2i-base-1-1 in LightBlue which is built based on different NVRs s2i-core-1-1 and s2i-core-1-2?

How about move this outside top-level for-loop? To make call to sort len(nv_to_nvrs) times.

Yeah, it should be different s2i-base version. I will fix it, but it does not make the test work differently.

I have tried to use x and y now, but I find the image_id/parent_id more descriptive when it comes to this line:

                # And in case this image is not the the leaf image, also replace
                # the ["parent"] record for the child image to point to the image
                # with highest NVR.
                if parent_id != 0:
                    to_rebuild[image_id][parent_id - 1]["parent"] = nvr_to_image[latest_nvr]

Although I agree x and y are more consistent with the "coordinates" name I chose...

+1

Commit d4bca471 fixes this pull-request