It was during the last days of 2025 when our team lead decided to take an extra day off and missed an interesting meeting. After a recent restructure, one colleague moved out of the project, and 2 newcomers were added.
The topic seemed simple: how should we structure our new code. Architects agreed on a DDD separation with additional grouping of rest, core and repository sub-packages(it is a Spring Boot web project) for each domain. This led to an interesting discussion separated should those three layers should be.
During my ~10-ish years of professional career, the most commonly used approach has been having a DTO POJOs at the entrance/exit of the REST controllers which were transformed to an entity objects which corresponded to the structure of the data in the storage used(i.g. RDBMS). Some process exclusive POJOs might exist, but we won't mention them for sake of simplicity. The process would look something like this:
The main reason for separating the request/response objects from the ones which were used for storing/extracting data from a storage was that they might have a different structure and even if they are absolutely identical - it is a way to future-proof the code against changes.
Strictly speaking, this doesn't fully decouple the core and repository part of the code - any changes to the repository entity will directly affect the core, but in my experience you don't randomly change the entity you use without a need for that from the business logic(core) and thus I have formed the habit of only using a DTOs and entities.
There was a team decision, despite my disagreement, to fully decouple the core from the repository, so now we have something like this:
Although it seems pretty similar, now we have additional mappers and POJOs which are copies of the entities. My main concern was the overcomplication when a change is needed - there are more classes which must be changed, which, with time and project growth, will bring, in my eyes, unnecessary complication. At the meantime it might be useful in the future if there is an optimized structure for the process in the core classes which is totally different from the repository ones.
But hey - I might be wrong here and am afraid of a non-existent problem. Constructive feedback is always welcomed.
Godspeed and wish you all fantastic holidays!
Top comments (0)