The steps to producing a legacy system

The steps to producing a legacy system

The old and new legacy systems

When we think about legacy systems, we typically think of systems developed in languages like Cobol, Clipper, Pascal, Delphi, Visual Basic, connected to old databases such as Paradox, DB2 and, Firebird.

Nowadays, it's a little different in an organization with multiple languages and projects. For example, Paypal opted to go from Java to Node years ago, and Twitter switched from Ruby to Java. With these examples, we can see that in the legacy context, we are dealing with modern languages, such as Ruby and Java. However, I don't think these developers were driven to change because they favored one language over another.

Refactoring as the solution?

Refactoring is becoming more popular for a particular group of engineers who prefer the hype language over others. I'm not here to pass judgment because I'm wearing this cover 🧙🏾 at a specific point in my career. But I should emphasize that refactoring is never the easiest or best way to solve an issue. As a developer that works with a range of programming languages, You will never find a bulletproof language that works so well for frontend and backend or bff (backend for frontend), that is amazing for mobile, that is lovely and comprehensible for concurrency, that is conformable to test, and so on...

Stop thinking about frameworks and start thinking about how this language will work in your project and ask some questions; the learning curve is good for other members; and consider how other people will solve issues with your produced project. Because if you don't care, you're creating the next legacy system.

Let's get started on a list for creating a stunning legacy system.

1) Languages with limited library support

Before deciding a programming language, evaluate what you plan to develop as a project first, and then whether your stack will be supported by the programming language, for example:

I'd like to create a project involving machine learning or data science. Python, as you may know, is widely used for these reasons and has strong commercial and community support. We may be able to find additional Java or Node libraries, but you will almost certainly have to get your hands dirty translating library behaviors and providing some compatibility.

I'm not arguing that it's completely wrong to use language A or B; you can chose, but you should consider the advantages and downsides. And this decision is sound in the long run when your team chooses to move to another language since there is no support for building quickly, because nowadays you must release fast or your solution design may become deprecated.

2) Use a framework that updates slowly

Nowadays, languages support a wide range of databases, services, and integrations, but occasionally there is limited support, or the community does not generate active updates based on your requirements. That condition is common, for example: NPM, RubyGems and Hex packages without updates for months or years.

Some projects are mature and there is no requirement to update them so frequently, but there comes a specific point when the project is supported by three core committers, each of whom has their own priorities. So, in that case, you must work with these dependencies and collaborate on open-source projects to solve issues or improve security; so, before establishing a framework, list the dependencies as clearly as possible.

Therefore, if open-source efforts exceed commercial efforts, your team may switch from one framework or language to another, introducing legacy systems.

3) Don't think about concurrency or performance.

We commonly hear monolith first and keep it simple, which is a genuine and reliable technique for launching your MVP as soon as possible, but be careful to maintain things simple enough to level up when necessary. I'm not advocating putting reuse ahead of usage, but don't make things too dependent on a framework. Some abstractions with low utilization will allow you to upgrade when "concurrency" comes knocking at your door demand for more performance.

The "performance" is at the door.

4) Avoid writing tests or maintaining adequate coverage.

Coverage tests are a sensitive topic; I've heard that quantity isn't necessarily better than quality, but less coverage is always worse. You should not write code that lacks appropriate coverage; instead, you should enumerate possible cases to provide coverage; that is every developer's duty. Assume you are a developer for an airline system; is less coverage acceptable? Okay, I took it seriously. But if we get a system with no testing and a bad design we should replace it as quickly as possible.

However, these systems will occasionally live a long time if they work properly and don't interfere with the strong performance of the stack, and the team has no plans to touch these poorly built systems. A system without testing is a good approach to start a legacy system, in my opinion.

5) Write in a new language in the same way you do in previous languages.

It is important to note that approaches and patterns can be used to any language, but you should be aware of the two paradigms commonly used when developing a project, the most well-known of which is Object-Oriented Programming (OOP), and another strong paradigm known as Functional Programming (FP). While OOP emphasizes class management and reusing behaviors, FP tackles modules, functions, and immutability, thus comparing different approaches. I propose using a well-known project's design as a guide when developing your project in a new language, because it's common and comprehensible to write code for a language and then have another person look at it and remark...

This code appears to be in another language...

To summarize, writing code in a new language is challenging, especially when establishing a new project, but it is a worthwhile experience. If you did your homework and chose this challenge, try to develop small initiatives first; it's not time to rewrite all behaviors that you consider legacy.

Remember that your baby system could become legacy at the same rate that a npm package is released. 😆

6) Write code for yourself rather than for your team.

I believe this happens more frequently than it should. We should not think of of code as abstract art because abstract art is about feelings and is hard to comprehend. Don't let hype influence how you construct your application; the coding should be as straightforward as history.

When coding, try to use well-known and rock-solid methodologies such as SOLID. If you develop a project for yourself, someone will look at it months or years later and say it's too complex, and it's time to retire it and...

A new legacy was born. To replace the legacy system features, a new system will be released.

Final thoughts

In this article, I discuss things to think about while developing a new system with new behaviors or refactoring behaviors from an existing production system. Sometimes a new system is successful, as proved by metrics and team, but other times a redesign works well for a brief amount of time and a new design is required.

To recap, I am not arguing that we should not test hype languages or frameworks, but rather that when you want to bring these new techniques to your team, you should do your homework, ask questions, create proof of concept (PoC), and test metrics to avoid replacing one issue with another.

Thank you for investing your time in reading; God 🕊️ be with you and keep your kernel 🧠 updated, bye.