parker's stuff
 

Go at Google →

Rob Pike breaks down the motivations for writing a new programming language for use at Google, namely Go. He walks through the difficulties Google was having with its go-to language, C++, and how a new language could tackle those problems. In 2007, Rob Pike, Ken Thompson, and Robert Griesemer began designing a new language to tackle the frustrations and friction in the C++, Java and Python workflows at Google. At four minutes in, he describes them:

  1. slow builds
  2. uncontrolled dependencies (#ifndef’s everywhere)
  3. each programmer using a different subset of the language
  4. poor program understanding (documentation, etc.)
  5. duplication of effort
  6. cost of updates (e.g. retrofitting an api change)
  7. version skew
  8. difficulty of automation (auto rewriters, etc.): tooling
  9. cross-language builds (inline python in a c++ program, for example)

He notes the following: Language features don’t usually address these.

I found this talk incredibly interesting not only because I have used Go and it is interesting to learn about its history and motivations, but because the motivations do in fact extend beyond Google, particularly in the duplication of effort, code readability, and dependency management aspects of programming. In the start-up world (a world Google exited many years ago), code is often written as though it were ephemeral. Due to this attitude toward the longevity of code, most start-ups have a huge problem with poorly-architected “legacy code” that slows the team down. The natural inclination for the future-minded start-up CTO is to write huge layers of abstraction into the code base, which often winds up confusing future engineers even further.

What Go does is make it easy and “feel right” to use a lot of composition. And guess what: composition scales really nicely. Imagining your system from the perspective of composition offers some real benefits over imagining it from the perspective of an inheritance tree, and it saves a lot of headaches as a result. Additionally, the concurrency support is extremely good, the portability of programs is fantastic, and the lack of arguments about code style (go fmt) is just terrific.

After hearing this talk, I’m going to pick up Go in the future as I start building new systems that don’t require a lot of type dynamicity.