Skip to main content

The 'Rewrite' vs the 'Refactor'

Sometimes you will come across a codebase / application that looks in such a mess or so old that the only option is to do a full rewrite. The code is horrible and out of date, it's using an old version of a language / framework and everyone has forgotten how it works. Everyone agrees, it will be much faster to just throw it away and start again.

This seems to be how so many software projects start. Over my career I know that I've been heard to say those very things myself. It's seductive, we all love a nice green field project and the big rewrite is the easiest way to turn a boring, difficult brown field project into something everyone wants to work on.

The hard truth is that I can only think of one (smallish) rewrite that I've worked on that could be classed as success and that company is no longer around (but I don't think the two things were related). More often than not, after we've finished the project (late, normally very late) our first thoughts are 'Now, if we could just have just another 2 months we could rewrite it from scratch (again) it would be so much better". Unsurprisingly most business don't respond overly positively to those sorts of requests!

Now when we're considering a re-write I'm trying very hard to work to the following process:

  • Estimate how long the legacy application will take to refactor to the desired quality / level of maintainability. At this point, double that estimate and if the business can afford that amount of time, then you're good to go with the full rewrite (over the refactoring). Of course you might ship the first version faster than that, but we all know that it's about version 3.0 that we get the desired functionality, stability and performance - rewrites generally aren't any different.
  • When there just doesn't seem to be any way to refactor an application, you've not looked hard enough. Go back to the drawing board and try again. Maybe the UI needs to be rewritten as you're moving from a technology that is no longer supported, but the business / domain logic can probably be reused. If that needs to be rewritten, do that in a later project in smaller chunks. Sure, it may take a while but do your estimates for the refactor approach and refer back to the first point.

If you're still not convinced (you might have to live the disappointment of seeing what a v1.0 of a rewrite looks like), then the additional benefits of the refactor approach are:

  • The only functionality missing from the final application will be depreciated functionality that the business agrees is no longer needed.
  • It should be possible to ship the application incrementally through-out the refactoring. If it's not, are you sure you're not just doing a re-write?
  • Development of new features doesn't need to stop, it can continue alongside the refactoring or even get included into the refactoring process itself. Maybe changing the functionality first will make the later refactoring easier, or the refactoring can be targeted in a way to make the later change of functionality easier.

So, taking all that into account, who would ever consider a rewrite again? Now back to application I need to maintain, I'm sure it doesn't do much and I could easily replace it before I've even finished understanding what all that horrible, cluttered code in the original application actually does. I hear all the cool kids are playing with 'Rust' these days and it would be good to see what the Google Cloud platform is like for hosting…….


Popular posts from this blog

Problem installing AWS CLI

It never feels like a good start when you're trying to start out with something and the install fails with an obscure error! I was just trying to install the Amazon CLI following the instructions at and ran into the following error when running 'pip install awscli': Collecting awscli Could not find a version that satisfies the requirement awscli (from versions: ) No matching distribution found for awscli I appeared to have a correct version of Python installed (v2.7) and checking "PIP -v" indicated that 9.0.1 was installed. That all seemed to tick the required boxes but digging around a little more I did see that some people had had issues with various versions of PIP so I found / ran the following to upgrade to the latest vesion: curl -o python This installed v9.0.3 of PIP which burst into life when I re-ran 'pip install awscli' and everything seems to be ok. Like…

Mocking HttpCookieCollection in HttpRequestBase

When unit testing ASP.NET MVC2 projects the issue of injecting HttpContext is quickly encountered.  There seem to be many different ways / recommendations for mocking HttpContextBase to improve the testability of controllers and their actions.  My investigations into that will probably be a separate blog post in the near future but for now I want to cover something that had me stuck for longer than it probably should have.  That is how to mock non abstract/interfaced classes within HttpRequestBase and HttpResponseBase – namely the HttpCookieCollection class.   The code sample below illustrates how it can be used within a mocked instance of HttpRequestBase.  Cookies can be added / modified within the unit test code prior to being passed into the code being tested.   After it’s been called, using a combination of MOQ’s Verify and NUnit’s Assert it is possible to check how many times the collection is accessed (but you have to include the set up calls) and that the relevant cookies have …

Injecting HttpContextBase into an MVC Controller

It is a shame that when the ASP.NET MVC framework was released they did not think to build IoC support into the infrastructure. All the major components of the MVC engine appear to magically inherit instances of HttpContext and it’s related objects – which can cause no end of problems if you are trying to utilise Unit Testing and IoC. Reading around various articles on the subject just to get around this one problem requires the implementation of several different concepts and you are still left with a work around. The code below, along with the other links referenced in this article is my stab at resolving the issue. There’s probably nothing new here, but it does attempt to relate all the information needed to do this for Castle Windsor. The overview is that all controllers will need to inherit from a base controller, which takes an instance of HttpContext into it’s constructor. It then overrides the property HttpContext in the main controller class, supplying it’s own version…