Comment by cs702 | Hacker News Reader

Comment by cs702 | Hacker News Reader

While the title is linkbait-ish, the article makes interesting points with code examples, and the author is not 'religious' -- he concludes by saying he will continue to use Go.

Like most other languages, Go is neither great nor bad; it's just another tool designed by people to solve a specific set of problems -- in Go's case, people who in the past have had to develop very, very large, complicated systems with lots of concurrency. For projects like that, Go would be a great choice; for other projects, it might not be.

In some circumstances even Ruby can be a better choice than Go for high concurrency. I recently read a post on HN in which the author (someone with an Erlang background) lays the reasons why he chose Ruby for a highly concurrent application. When I first saw the link, my immediate thought was, Ruby!!?? But then I read the post and the reasons were all very sensible and practical-minded, so in that case Ruby was arguably a much better choice than Go, Erlang, Scala, Rust, and the like.

Here's the post: https://news.ycombinator.com/item?id=10394450 (direct link: https://github.com/rustyio/super-imap#why-ruby) -- I really liked the rational, practical, 'non-religious' tone.

Suggestion: don't call Go or any other popular language "poorly designed" with a short blog post. Instead, say "it doesn't fit my current needs."

> don't call Go or any other popular language "poorly designed" with a short blog post. Instead, say "it doesn't fit my current needs."

that's not what the author wanted to say though. very, very often someone says "this is bad" when they actually mean "this doesn't fit me". in this scenario, I think the author made a very good case for "it's poorly designed".

RodericDay:

Language design decisions by superstars like Robert Griesemer, Rob Pike, and Ken Thompson (the creators of Go) that may seem inconvenient or annoying when working on small to mid-size projects could very well make life much easier when working on giant projects which take years and do not fit on any single person's head.

I'm reminded of this insight by Lawrence Kesteloot of Dreamworks: "What’s particularly hard is having technical discussions with someone who hasn’t broken through as many walls as you have. Breaking through these walls means making different trade-offs, and specifically it means making a decision that seems to make less sense in the short term but will help later. This is a hard argument to make—the short term advantages are immediately demonstrable, but I can’t convince anyone that a year from now someone may make an innocent change that breaks this code."[1]

Unless you personally have led Internet-scale mission-critical software projects at a place like Google, as the authors of Go have, I don't think you're in a position to judge their decisions too harshly. Neither am I, BTW.

--

[1] http://www.teamten.com/lawrence/writings/norris-numbers.html

I keep seeing this argument all the time. Lets put a counter-argument-by-authority: Fran Allen [1] thinks that the C was a huge step backwards in language design [2], and there is no reason to think that Go didn't repeat the same pattern.

Interestingly, the article you quoted mentions functional programming and immutable data as the step to go from 200K to 2M lines. Go is fundamentally incapable of functional programming* and its builtins allow pervasive mutable state.

[1]: https://en.wikipedia.org/wiki/Frances_E._Allen

[2]: http://www.amazon.com/Coders-Work-Reflections-Craft-Programm...

* Its impossible to support FP without generics. Even the most basic higher order functions require type variables.

The guarantees offered by functional programming and immutable data are great, but they mostly evaporate when you move from a single process application to a multiple processes system, distributed on many machines and communicating through IPC.

Are you aware of a 2M lines code base that fully relies on FP and immutable data?

Erlang's very existence is a pretty solid refutation of the claims you're putting forth. :P

I don't think so. An Erlang process has no mutable state. But it can communicate with another Erlang process to set/get some state. This is how a lot of mutable structures are implemented in Erlang.

I beg to differ: the guarantees functional programming and data immutability offer are exactly the sorts of guarantees needed for scalable parallelisation.

LOC is not a useful metric for the utility of a program, in fact the reverse is true in terms of maintainability and places for bugs to hide.

This chap claims to write very robust programs, entirely functionally that are worth billions in their operation.

http://logicaltypes.blogspot.co.uk/2015/08/pure-functional-p...

I agree that "the guarantees functional programming and data immutability offer" are very useful for "scalable parallelisation".

But this is not what I was answering to.

I was answering to the idea that FP and immutability are useful when you scale from 200K to 2M lines of code:

> the article you quoted mentions functional programming and immutable data as the step to go from 200K to 2M lines

In my opinion, this argument is irrelevant.

The argument being that non functional programming has fundamental limits to its scalability, seems like a very important question as the complexity of code begins to grow exponentially.

Large imperative codebases are inherently fragile, hence the need for test driven development - good batteries of tests offer demonstrable reliability.

Functional programming offers provable reliability.

Provability is not just for scaling complexity but also for much simpler software that absolutely must not fail.

Imperative code becomes hard to reason about before 200K LOC is reached.

The counterargument is that very large codebases tend to be composed of multiple services. Each service lives in its own process and maybe in its own machine, and communicate with other services through IPC. At that point, you've lost the benefit of functional purity and immutability, because each service can maintain some "hidden" state.

You're right, above some threshold, it becomes hard to reason about imperative code with mutations. Your preferred solution is to adopt functional programming and immutability. Another solution is to decompose your system in multiple communicating services. Google's codebase is a very well known example of the latter approach.

This is false. Well, it may be true in Erlang, but its false in Haskell, where there is always clear separation between equations and effects/state no matter the source (IO to other services or local state). Not to mention that there are tools such as STM that let you deal with state in a much safer and less error-prone way.

Separating the state between multiple communicating services is basically the same strategy as OOP (objects that communicate via messages and encapsulate state) but with stronger encapsulation requirements (access only via API, you can't just "reuse" methods willy-nilly all over the place but you have to come up with a sensible library or a third service, etc etc).

It works, yes but it works about just as much as OO does. Which is... not very much. And it also comes with its own tradeoffs (good luck getting atomic changes over multiple microservices)

Do they, now? Or do they only evaporate in Erlang?

Since most FP codebases do the same with a few orders of magnitude less code, no, I'm not aware. Though I imagine that if they had to copy their code for every single instance of their generic typeclasses, they would quickly amass many millions of lines.

Reply on news.ycombinator.com