The Hidden Cost of Backward Compatibility: When Deprecation Turns into Technical Debt - An Experience Report

Abstract

Context: The micro-services architectural pattern advocates for the partitioning of functionality into loosely coupled services, which should be backward compatible, to enable independent upgrades. Deprecation is commonly used as a tool to manage multiple versions of methods or services. However, deprecation carries a cost in that tests might be duplicated and might rely on services that have become deprecated over time.Objective: Using the terms of the Technical Debt metaphor, we explore the consequences of deprecation, and how it has affected the test base during seven years.Method: We take an exploratory approach, reporting on experiences found before and after servicing parts of the incurred Technical Debt. We mine code repositories and validate our findings with experienced developers.Results: We found that the growth of deprecation debt varied a lot. Some services experienced substantial growth, but most did not. Unit tests, where deprecation is visible in the developers’ tools, were much less affected than integration tests, which lack such visualization mechanisms. While servicing debt of 121 out of 285 deprecated services, we discovered that up to 29% of the spent effort could be attributed to accrued interest. However, this is an upper bound; there could be less impact, depending on whether scripting could be used to service the debt or not.Conclusion: This paper illustrates that integration tests can be viewed as a debt from the perspective of deprecated services. While the pattern was that deprecated services (debt principal) experienced no or little accrued interest, some, highly used, services experienced a lot, particularly during stressful times. Java-based tests, where deprecation is visible in the IDE, did not experience a similar pattern of increasing debt. We postulate that deprecation debt should be kept visible, either using developer tools or statistical reports.