From the article: This is like finding a new alloy of steel and then building a bridge out of it to see if its any good. If Go doesn’t have generics, and you know you need generics, why the fuck did you bother with Go? These details are widely, exhaustively documented; they are not surprises.
Well, I took a good hard look at Go, and I said, "There's a lot of stuff here that's really, really sweet, but error handling is going to require 600 gross conditional statements, and I'm going to have to rewrite common algorithms over and over and over because there are no generics."
But lots of people in Go community kept saying, "Yes, it looks a little unusual, and it doesn't have all the features you'd expect, but you need to give it a fair chance, and try writing a real project or two in it. Don't be a blub programmer, afraid to try out strange new abstractions."
So I wrote a couple of medium-sized programs in Go, including a server which broke simple substitution ciphers (to help perform OCR on subtitles). This involved a lot of tweaking of algorithms and optimizing. This was maybe my second or third medium-sized Go program.
And maybe about the third time I had to tear apart the server and rewrite it, and said, "This whole experience is actual incredibly painful. Why does this code keep hurting me every time I need to redesign something?" So I paid closer attention, and I said, "You know, my code is buried under error-handling conditionals and lots of stupid little subroutines that should be in standard library—but which can't be there, because there are no generics."
Seriously, if Go had Erlang OTP-style error-handling and some ruthlessly-simplified generics, I would use it all the time. The type system rocks, the basic data structures are excellent, and the standard library is good. It's clearly well-optimized for writing certain types of programs. It's just I don't write those programs very often.
It turns out what I want is a lot closer to Elixir: Erlang concurrency semantics, decent Unicode string handling, some light metaprogramming and a Ruby-like syntax. Too bad Elixir is still so young and obscure.
I'm with you here. Golang has a lot going for it, and it isn't shy about its shortcomings, but the story is "Try it, you'll like it." After a year, I still find myself bogging down coding things which should be easy and expanding my SLOC because of warts in the language.
As someone who learned C in the pre-ANSI language days, in the pre-POSIX library days… I have high hopes for Go 2.0, whatever that ends up being.
If you know Go doesn't have what you need, then why did you use it?
Sometimes, a language isn't a good fit for what you're building, and you said that yourself, you don't write those programs very often.
And some other times, a language just isn't for you, it doesn't match the way you think about programming.
For example, I tried to learn Scala a lot and it just didn't work, so I stopped trying because there are better uses of my time than saying: "I wish language X had feature Y" or "I wish language A was more like language B"
If Elixir has what you need, then by all means use it. Nobody knew Python before Google started using it, didn't stop them from doing great things with it.
I think he answered why he used it pretty clearly in paragraph 3: because people in the Go community told him he had to use it for a bit to give it a fair shake.
Another way of saying it: he couldn't have known it didn't have what he needed until he tried it. So he tried it. Which I think is great. There are lots of thing we think we need that we don't, and lots of things that we never knew we needed until we tried them.
A language with everything would be a nightmare. To program efficiently you need a balance between capability and mental footprint. If I have to write large amounts of code to accomplish a task, that is bad for me and the eventual reader. If I use some feature so esoteric the read has to stop and research, or worse, misunderstands that is also a problem.
The Go promise is that they don't have some things, but you won't miss them. Some are deliberate omissions, some are deferred until they decide if they are required and can sanely be implemented. How those impact you is going to depend on what problems you are solving.
The danger is to end up wanting language casserole with bits and pieces from everywhere. It might be what you want, but by the time you union that with everyone else it is too big to know.
You don't go in saying "I need exceptions. I will use Go!" You start down the road saying, "I used to use exceptions, but in Go I will explicitly check all the errors." Only with experience can you look back and see what the cost was.
>I think of these languages, these tools, as something akin to violins and cellos. They’ve been around, virtually unchanged, for several hundreds years. They are an absolutely gigantic pain in the ass to play- they haven’t got any frets, they have tiny fingerboards designed to be played by scrawny, malnourished 15th century musicians. They’ve got tuning pegs designed for strings made out of animal guts and which rely on friction alone to stay in tune. They’re freaking baroque. And yet, they get played a lot. You might argue that their design actually encourages mistakes.
I wish people wouldn't make analogies about things they don't understand. Frets do not make string instruments automatically better, in fact they limit you in terms of intonation and vibrato. The fingerboards are exactly the size they need to be, and would be less ergonomic if they were simply made wider (every attempt at an ergonomic violin I've seen has involved changing the body shape to make it easier to hold, not widening the neck. Widening the neck substantially on a bass would just be completely unhelpful) and a lot harder to bow. Friction pegs were certainly designed for gut strings, and that's why you'll find geared tuners on many instruments that use steel strings (and sometimes even on ones that use guts).
I just find it funny that he doesn't like it when people who don't understand Go try to criticize it, and then criticizes something he doesn't understand in an effort to strengthen his argument.
Actually, the author is rather well informed on the history and construction of stringed instruments. He simplified his loosely related analogy so he wouldn't have to write huge footnotes of apologia and exceptions in his mapping between domains.
In general, attacking the perceived inapplicability of an analogy is weak sauce. In this case, you allowed your pedantry to lead you to completely miss the point the author was making, both in his analogy and his thesis.
I don't think you read my comment in full. I said nothing about how effective his analogy was, just that it was obviously misinformed, to a laughable degree.
Of course I read it in full. You write a lot about how his comments on violins are wrong (which I say is irrelevant) and then say "I wish people wouldn't use analogies about things they don't understand" (which I say is also irrelevant, because it's effective).
You then misrepresent what he doesn't like, which I ignored because one can only pick on so many annoyances in one post, then you suggest he was criticizing violins (which he wasn't) in an effort to strengthen his argument (which he wasn't, and he isn't arguing anyway he's describing his viewpoint).
I'm going to go with my original comment before I changed it for something more polite.
Your comments are like the Go language: weakly typed and useless.
Hey, you're right, being intellectually lazy is effective!
The problem with Go it that it is fundamentally non-exciting. It brings very little to the table, when compared to C.
On the top of my head:
- GC (which some people find to be actually a cons)
- Some more modern ammenities for debugging.
- Coroutines (this is really where go shines)
On many other fronts, Go is actually a regression from C, because it favors being "safe" over being expressive. Goodbye macro, pointer arithmetic, etc.
Omitting exceptions is a moronic decision. In a program that does its fair share of I/O, this means that error checking will end up taking a huge % of the code written (easily 30%). It will naturally push people to either eschew error checking, or to write huge methods to dilute the error checking code.
So in the end, I'll use C with a coroutine library. It'll feel (and be) more hackish, but I'll be more productive and my code will be easier to understand.
Now, that's my opinion. Maybe other people have different sensibilities and Go is just fine for them. But I think it's pretty obvious that Go is not right for a lot of people, and nothing's wrong with them. It's unfair to pretend they haven't a point. Go is not for everyone.
They didn't. Go has panic/recover. It isn't idiomatic to expose a panic in a library, but if you know you're going to be doing a lot of error handling, then you can utilize panic/recover to make error handling much terser. I do it frequently when writing parsers[1] or web applications.
> So here we find ourselves in a world with various better choices than C and C++, yet they still get used for a pretty significant chunk of all the coding that goes on in the world. Why is this?
Isn't the emergence of relatively many system programming languages in recent years partly a reaction to there indeed not being many viable choices? I mean, I guess you have all the languages that are C-level (or Go-level) that were simply out-competed by C in the mainstream software world for whatever reasons, but people don't seem willing to go and program in languages like Ada and seeing if it's a better fit (or at least I haven't heard much of it). It seems more like, "every other decently mainstream language has mandatory garbage collection → therefore C/C++".
Well, I took a good hard look at Go, and I said, "There's a lot of stuff here that's really, really sweet, but error handling is going to require 600 gross conditional statements, and I'm going to have to rewrite common algorithms over and over and over because there are no generics."
But lots of people in Go community kept saying, "Yes, it looks a little unusual, and it doesn't have all the features you'd expect, but you need to give it a fair chance, and try writing a real project or two in it. Don't be a blub programmer, afraid to try out strange new abstractions."
So I wrote a couple of medium-sized programs in Go, including a server which broke simple substitution ciphers (to help perform OCR on subtitles). This involved a lot of tweaking of algorithms and optimizing. This was maybe my second or third medium-sized Go program.
And maybe about the third time I had to tear apart the server and rewrite it, and said, "This whole experience is actual incredibly painful. Why does this code keep hurting me every time I need to redesign something?" So I paid closer attention, and I said, "You know, my code is buried under error-handling conditionals and lots of stupid little subroutines that should be in standard library—but which can't be there, because there are no generics."
Seriously, if Go had Erlang OTP-style error-handling and some ruthlessly-simplified generics, I would use it all the time. The type system rocks, the basic data structures are excellent, and the standard library is good. It's clearly well-optimized for writing certain types of programs. It's just I don't write those programs very often.
It turns out what I want is a lot closer to Elixir: Erlang concurrency semantics, decent Unicode string handling, some light metaprogramming and a Ruby-like syntax. Too bad Elixir is still so young and obscure.