The real value of DSLs

A few years ago, domain specific languages (DSLs) were a really hot topic and “the future”. Now, after the dust has settled there are a bunch of successful and useful examples like LINQ and Gradle but no holy grail buzz anymore.

In some of our projects we did implement some small embedded DSLs (eDSL) if we expected great benefit. Most of the time they are just working silently in the background of our solutions. But sometimes they really shine:

A recent story involving a DSL

A few days ago, I got a request from one of our customers regarding a complex “traffic light”-like system for scientific proposals in a proposal submission system. Each proposal goes through a bunch of phases where different parties have participate. While that sounds relatively simple our system has different proposal types, different roles, different states over time and so on.

All of the above leads to “Starbuck’s menu”-style complexity. And now, after several years in production the customer comes up with some special cases that are counterintuitive.

Expecting the worst I dived into the source code and was delighted to see a quite expressive DSL. Our project and thus also the DSL for the “status overview” feature leveraged the flexible syntax of Groovy (which is also used by Gradle). After a few minutes I could not only understand all the rules for the different status lights but also pin-point the cases in question.

I then moved forward and used the rule definition in our DSL to talk to our customer directly! And after a few more minutes he – as the domain expert – also understood our rule definition and the cause for the irritating display of a certain combination of proposal type, role and phase.

The fix was more or less straightforward and I simply handed that one source file to our customer for future reference.

Here is an excerpt of our status phase rules written in our own DSL:

PhaseRules {
    submission {
        green {
            proposal.isValidated()
        }
        red {
           !proposal.isValidated()
        }
        white {
            false
        }
        user {
            yellow {
                proposal.isCallPhaseIndependent())
            }
            grey {
                false
            }
        }
        scientist {
            yellow {
                false
            }
            grey {
                proposalNeedsValidationByProposer(proposal)
            }
        }
        userOffice {
            yellow {
                false
            }
            grey {
                proposalNeedsValidationByProposer(proposal)
            }
        }
    }

    check {
        green {
            proposal.isValidated() && proposal.hasFeasibilityComment()
        }
        red {
        }
    }
    // more phases, roles, states and rules omitted...
}

In that moment I was very proud of my project team, their decision to implement a DSL for this feature and an appropriate implementation.

It saved me a lot of headaches and quite some time. Furthermore it reinforced our customers trust in the application and improved the transparency and traceability of our solution.

This high level abstraction let both me and the customer ignore all the implementation details and focus on the actual functionality and behaviour correcting/improving the status display for our users.