Fragility versus rigidity trade-off


Recently I’ve been involved in a lot of discussions about dynamic versus static type systems1. It is very difficult to have constructive discussions about this subject because it touches personal identification with tools, lack of knowledge about the programming language design space, differences in vocabulary, group thinking, emotions running amuk2

In the end people don’t pay much attention to other people’s points and when the time for conclusion arrive a variant of “there are trade-offs, languages are tools and you should pick the one matching your problem” and saying goodbye.

What leaves me deeply unsatisfied is listening to such sensible final remarks but not having spent most of the time discussing the trade-offs! I refuse thinking those remarks are polite but empty sentences not really meant and I blame it just lack of focus.

Leading by example, let’s focus on one important trade-off we face when we move through the dynamic-static axis: rigidity versus fragility. There are many definitions of the software -ilities (fragility, rigidity, viscosity, opacity…) but let’s quote the ones in an article about design principles3:

Rigidity is the tendency for software to be difficult to change, even in simple ways. Every change causes a cascade of subsequent changes in dependent modules.

[…]

Fragility is the tendency of the software to break in many places every time it is changed. Often the breakage occurs in areas that have no conceptual relationship with the area that was changed.

To keep things concrete, we can use an imaginary codebase with the same characteristics as the typical system you can find in real life. Adding features to such codebase is not just writing more code since you need to make changes to the existing design to accommodate for the new code.

Those changes are more costly as the codebase grows (more than linearly) and, critically, can become astronomical due to poor code quality.

On one hand, if you real-life tech-indebted system is written in a statically typed language, you will face significant rigidity. The reason is that modifying a tightly coupled module can mean modifying many, many files in a so called shotgun surgery because dependencies are explicit and checked at compile time. Depending on the environment you won’t be able to run tests unless everything compiles again, making it even more hideous. You can alleviate you pain with tools like the mikado method and the stuff in the Refactoring book.

On the other hand, the same codebase could be written in a dynamically typed language. The same far-from-ideal design will pose significant fragility. Thanks to duck typing, the cross references producing rigidity in the previous case won’t stop you from modifying the code but you will need to check all of them anyway. You should pay special attention to all the mocked interactions, reflective uses and double-check code not covered by tests4. As a result, innocent changes can and do break the code in many places sometimes far removed from the original change. However, you can delay updating the whole design (at the expense of inconsistency).

Assuming all other things are equal5, how to pick programming languages based on this trade-off?

  • System size. As the system grow in size and complexity dependencies will be more remote. Here rigidity is a win as it signal when to invest in scaling the design.
  • System critically. Prefer fragility for non-critical, throwaway systems aimed at demonstrating concepts.
  • System longevity. The more fungible the codebase the more fragility you can tolerate.

Of course, there is much more to discuss within the umbrella of the static versus dynamic debate.

  1. We had a podcast episode, a debate at JobAndTalent, coffee discussions and some mailing list threads… 

  2. We developers are passionate about our tools in an exceptional way. Can you imagine a street sweeper spending the weekend configuring its new broom? 

  3. Design Principles and Design Patterns, Robert C. Martin, 2000. 

  4. The dynamic languages community emphasizes unit testing and TDD but practitioners sin. 

  5. This is not usually the case because many other important aspects will be different when you compare static language X with dynamic language Y.