These tests expect the "wrong" results when merging a relative path with a base URI consisting of just a scheme and a relative path (no leading "/"). The tests affected are 2, 6, 7, and maybe 162.
The expected results are incorrect according to RFC 3986/STD 66, but are actually better than what the algorithm in the spec produces. When RFC 3986/STD 66 was in development, Graham Klyne reported this issue and suggested that the resolution algorithm be adjusted to provide a more reasonable result. There was no argument, so I made my absoluteizeURI() function and these unit tests with the expectation that the final draft of the URI spec would contain a fix for the issue. However, when RFC 3986 was finally published, there was no change to the resolution algorithm.