The repackaging of source code and its unintended consequences:

This article is written with the findings from an investigation that I participated in and is still ongoing. It does not explain everything. The investigation has not concluded yet because the root cause has not been found yet. I have not heard any other similar case so far and I have had difficulty in outlining the defect but the learnings from the efforts stand clear.

A specific java code was packaged as a jar because it had a very narrow purpose and needed to be reused with applications from different sources.  This jar happened to be a fat jar which is a term to say that it contained all the dependencies needed to run its code. The jar worked well whenever the applications were run. However, it was inconvenient to copy it to each repository manually. The time-tested way of bringing in jars like these has been to use one or the other distribution repository.

The code was therefore forked which is a term used to denote another location where a copy of the code would be modified. The changes could include a rewrite and some trimmings so that the newer code was lean and mean.  The forked code would then be uploaded to a repository and consumers could then refer to it by name in the dependency and it would all work well with the elimination of the unnecessary step to bring it in any other way.

It turned out that this effort did not really work the way as intended. The reason for the failure is still under investigation. The code was supposed to be the same and it was reviewed to ensure that no defects were introduced. It had test cases that seemed to work. Yet the code was not working for users.  If all the factors that played out with the previous code were determined, the current effort would have panned out. This was the key learning and is often referred to in the industry as regression tests. These tests are written with the goal of keeping the behavior of the current code the same as that of the previous irrespective of the changes made to the current. Somewhere, there was a regression introduced and it was manifesting as a failure.

The nature of the software development is that it is always focused on functionality. However, it is the value that all the assets that guarantee the non-functional aspects of the code that determine how easy it is to use the code. The functional and non-functional aspects are both needed from any implementation but while the former has the bulk of the emphasis on the maker, it is the latter that determines the appreciation.