Productivity vs Performance

Productivity vs Performance When I was writing software in college, there was more emphasis on program execution speed than on time spent implementing it. In startups and most work environments, the reverse tends to be true. It took me a while to figure this out, and for the first few years of programming, I would often introduce optimizations that were not necessary, or make code uglier than it needed to be for the sake of performance. I’m not talking about premature optimization, I’m talking about poor design decisions stemming from assumption that performance trumps legibility.

I’ve spent a lot of time refactoring poorly written code in Grafpad – code that wasn’t necessarily bad to begin with, but quickly outgrew its initial purpose as more special cases were introduced to it. What gave me even more grief, however, were the special cases I imposed on myself, in an attempt to preserve bandwidth, CPU and memory usage. For example, each shape point in Grafpad consists of 3 items: x-coordinate, y-coordinate, and curvature flag. In original version of Grafpad, shapes that I knew couldn’t have curvatures omitted that curvature flag. As a further optimization, I wrote faster versions of multiple algorithms (edge detection, intersection, bounding box computations) which didn’t have to deal with curved lines. I later ended up regretting that, having realized that I did twice as much work to handle a case that may have been fast enough anyway. I wasted time I could have invested elsewhere, I introduced special case logic that didn’t need to be there.

Another example is the logic I created for transmitting data to server and back. I didn’t like that JSON.stringify included a lot of irrelevant information, which I wouldn’t need since I knew exactly what kind of object I’m sending over. My packing method transmitted only the values themselves, and I was unpacking them correctly by following the same order of operations. Once again, I performed a bunch of work that JSON.stringify could have handled for me, and ended up with a more fragile solution that depended on pack/unpack logic on both front-end and back-end to be identical.

I’m not saying the work I did was pointless, it simply wasn’t the kind of work I needed to do at an early-stage startup. By the time these kinds of optimizations become relevant, the product should already have multiple users and a team of developers who have time to do these optimizations. An early-stage startup should concentrate on getting the product out the door and fixing bugs that affect users, performance issues rarely matter at that stage. And with proper use of polymorphism, those optimizations will be easy to add later on in cases where they do matter.