3 ms·

This is the experienced developers dilemma: "To engineer or to not engineer"; engineering usually turning out to be over engineering. A more experienced develop

This is the experienced developers dilemma: "To engineer or to not engineer"; engineering usually turning out to be over engineering. A more experienced developer friend of mine would always tell me: "build for today's requirements". I try hard to fight the design/architect voices in my head that always want to imagine this made up future where we will need X or else we cannot go live. These voices usually only stall real actual work, instill fear, and serve very little to no purpose. I once had an argument with a Wall Street Java developer who was made the "lead" of one of our team projects. He decreed that every single class have an interface so that we can be generic and not tightly couple any of the components to concrete classes. I agreed that in some instances where functionality, i.e. methods that can be represented by different classes with the same method signatures made sense but not every single class needs an interface (if that is the case just go with a beautiful dynamically typed language like Python and avoid the code bloat). He got management on his side and we went off and built an overly engineered Straight Through Processing solution. It was a sheer nightmare to debug and the code bloat made me scream one day when we had a serious production issue. Even our manager (who finally had to look at the code when most of us were out on vacation once to answer some user questions) was flabbergasted at the amount of code he had to read through in order to answer the most trivial of questions. One extreme example was an interface for trade references. Our trade references were always strings with a date and some numeric value concatenated to it. The "engineer" decided that we needed an interface for this and added one interface and concrete class for our trade references. I told him that all classes needing trade references instance variables could just have a String instance variable named tradeReference or something like that and he went on to give me a design pattern lecture. We argued for nearly 20 minutes about this silly thing as he kept insisting that the future was unknown so we have to future proof the code from unforeseeable changes. When he said this I asked him to remove the Crystal ball plugin he had in Eclipse for predicting the future and get real. He got angry and we had a team call to waste yet another hour of developer time to discuss this. In the call I mentioned that our trade references scheme had not changed in 8 years and was unlikely to change... I lost the debate anyway. The ratio for most of the code base from interface to concrete class was largely 1-1 thus not justifying this code bloat approach. Experienced developers (at least I think) seem to have these crystal balls in their heads or IDEs and usually try to be clairvoyant when it comes down to building a product. We need to get out of the business of overly engineering and just do as my friend said: "build for today's requirements". It is called software for a reason: it is soft. It can change (most likely will), can be refactored, redesigned, and/or incrementally made better or more abstract to accommodate changes. I am in no way saying no design, just limit it and get to work. A successfully built product is more satisfying then the imaginations of your head and the "perfect" engineering/scaling solution that never materializes. Users will like you, you will like you, and the team will get an andrenaline boost with each and every release keeping the spirits high. Remove the Crystal ball plugin from your head/IDE and stop trying to be clairvoyant and be a developer.

rumcajz 15y ago
The rule of the thumb: If an extension point is used by only a single extension, get rid of it. It won't fit the next extension anyway so the work needed to maintain it in the meantime is a waste of time.
brazzy 15y ago
This. A hundred times this. It is truly mind-boggling how many crimes against maintainability are committed in the name of future-proofing.
kahuna 15y ago
Depends if you are engineering an enterprise system/product, or building a smaller one off project. In the case of the enterprise solution that will be around for the next 10 years, I would agree with your tech lead. Every time you make a code change for something that is out of dev budget, you face a budget overrun in the project that was interrupted. If you choose to deliver something that just works and as soon as possible, your total costs tend to balloon over the lifetime of the system. Question to ask is: how long would it take for a new dev (someone who has never seen the code) to change the tradeReference naming convention to include the asset type, or some conditional tag, lets say to conform to reporting regulations or even an expanded business mandate? Interfaces do help here, because the new guy can make a localized change, write a small unit test, and commit the code to source control before you can say "rebuilding search index". Keep in mind that you do not know if something is well architected until AFTER its been in production for at least a year and had features built into it for another year or two, and has added new members to the team, and has lost a few of the original team members in that time, I believe that most people who are able to look back and claim that they have delivered at least three large projects (>500k LOC c++ or >200k LOC java) or product releases that meet the above criteria, would agree with your tech lead.
- rosariom 15y ago
  Good points; we were building an in-house solution for an Investment management wing in a bank. To give some context on the trade reference I referred to, it was an internal tracking within the STP system used solely for tracking state and for communication between IT and the business. It was a simple date + numeric value used in our STP system for users to use in our Struts web and C# front end to check trade state through the STP flow and communicate with us if issues arose. We did have lots of interfaces where it made sense and relied highly on object composition to represent financial concepts more richly and for inject-ability via Spring and Unit testing (makes writing tests easier when you mock things out). Asset types, security identifiers, etc, were represented correctly from an OO perspective and were a part of the xsd layer/interface between us and the trading systems. To us these were read-only values we just passed through for STP. You are right about interfaces and unit testing but this is one case I highlighted of many where I think the lead was going over board. The internal trade reference was the same for 8 years and still the same to this day (which gives it another 3 years since I left for a total of 11 years). It never had more than one concrete class. It is no big deal on its own, but when combined with the other interfaces that only have one concrete class, it just bloats the system for no good reason. Design and architecture is good for the reasons you have mentioned and more but it can go over board as is the case here IMO. There were other instances of that in our code base but that would take an entire blog post to cover some of the atrocities this engineer created because of his forecasting ability.

rumcajz 15y ago

The rule of the thumb: If an extension point is used by only a single extension, get rid of it. It won't fit the next extension anyway so the work needed to maintain it in the meantime is a waste of time.

brazzy 15y ago

This. A hundred times this. It is truly mind-boggling how many crimes against maintainability are committed in the name of future-proofing.

kahuna 15y ago

Depends if you are engineering an enterprise system/product, or building a smaller one off project. In the case of the enterprise solution that will be around for the next 10 years, I would agree with your tech lead. Every time you make a code change for something that is out of dev budget, you face a budget overrun in the project that was interrupted. If you choose to deliver something that just works and as soon as possible, your total costs tend to balloon over the lifetime of the system. Question to ask is: how long would it take for a new dev (someone who has never seen the code) to change the tradeReference naming convention to include the asset type, or some conditional tag, lets say to conform to reporting regulations or even an expanded business mandate? Interfaces do help here, because the new guy can make a localized change, write a small unit test, and commit the code to source control before you can say "rebuilding search index". Keep in mind that you do not know if something is well architected until AFTER its been in production for at least a year and had features built into it for another year or two, and has added new members to the team, and has lost a few of the original team members in that time, I believe that most people who are able to look back and claim that they have delivered at least three large projects (>500k LOC c++ or >200k LOC java) or product releases that meet the above criteria, would agree with your tech lead.

rosariom 15y ago
Good points; we were building an in-house solution for an Investment management wing in a bank. To give some context on the trade reference I referred to, it was an internal tracking within the STP system used solely for tracking state and for communication between IT and the business. It was a simple date + numeric value used in our STP system for users to use in our Struts web and C# front end to check trade state through the STP flow and communicate with us if issues arose. We did have lots of interfaces where it made sense and relied highly on object composition to represent financial concepts more richly and for inject-ability via Spring and Unit testing (makes writing tests easier when you mock things out). Asset types, security identifiers, etc, were represented correctly from an OO perspective and were a part of the xsd layer/interface between us and the trading systems. To us these were read-only values we just passed through for STP. You are right about interfaces and unit testing but this is one case I highlighted of many where I think the lead was going over board. The internal trade reference was the same for 8 years and still the same to this day (which gives it another 3 years since I left for a total of 11 years). It never had more than one concrete class. It is no big deal on its own, but when combined with the other interfaces that only have one concrete class, it just bloats the system for no good reason. Design and architecture is good for the reasons you have mentioned and more but it can go over board as is the case here IMO. There were other instances of that in our code base but that would take an entire blog post to cover some of the atrocities this engineer created because of his forecasting ability.