I benchmarked some go code this week between use of an `atomic.Value` and a `sync.Mutex`. Use is vastly biased towards reads (almost no writes, and thus no need for exclusive locking), so I expected the atomic.Value, being a teeny tiny wrapper around CAS primitives, would be faster.
And it's not. Single-goroutine use of atomic.Value is actually slower than use of the exclusive sync.Mutex in my benchmarks. Why? Compare-and-swap should be just about the fastest synchronization primitive possible. So, what else is wasting time?
As far as I can tell, it's the interface conversions that atomic.Value (and I, as the consumer) have to do. The sync.Mutex sits next to the var instead, so the var still has its own type info that way, and so though the lock is heavier, there's no casting. (Mind, this is not a huge speed difference -- this is faffing around at the 10s/100s of nanosecs here. I'm far enough inside a hot loop it matters; many applications won't be. But still: it just startles me atomic.Value is slower here. I'm going to go out on an opinionated limb here, but: It shouldn't be.)
I'm still going to use the atomic.Value, because it's algorithmically the right choice: as the structure it's covering grows, the amount of cost the exclusive locking would cause will increase, and with atomic.Value it won't. But it's... it just feels Odd to look at that benchmark.
Perhaps one of the design decisions was to make this cost clear to the developer. It's also a sign that Go doesn't want to be so high-level as to make low-level stuff secondary.
The fact that it's slow on the CPU side is not great, but I don't think it's a major problem in the eyes of the core dev team: as far as I can tell, they do not believe that generics support as commonly envisioned is in line with go's mission.
Go is the anti-lisp / anti-unix of languages; designed to not give a team of junior devs enough rope to hang themselves. If we'd like a generics proposal that would get buy-in, I think we need to adhere to that dream while we imagine working generics and propose something; I bet it would be accepted.
Moreover, the idea that generics allows junior devs enough rope to "hang themselves" is preposterous. There are more than one type of generic system out there, we don't need to replicate C++ templates here.
Let's not fool ourselves here. There's a difference between "we just haven't figured it out yet" and "generics doesn't fit into the vision of Go". The former attitude is perfectly acceptable. The latter is really annoying and misleading because Go does use generics -- internally.
I think there's some misconception here, Russ Cox (one of the core Go team) posted this on HN a few weeks ago https://news.ycombinator.com/item?id=9622417
I've followed the go team's discussion of generics for a long time now. They've always claimed they're open to proposals that fit the philosophy of the language, but they're seemingly not in any hurry to address the issue. And that's fine. But I won't pretend it isn't frustrating every time I want to build something with a non-trivial data structure. Sometimes a map or a slice just doesn't cut it.
When I needed a generic data structure in C, I would do code generation (but only after I'd have a use for it in more than one place). The number of times I ever needed to do this for anything other than an array or hashtable is (not really surprisingly) vanishingly small.
So, coming from the perspective of Go is a better C (that avoids the mistakes of C++), this choice makes sense to me. I don't necessarily think it's the right choice -- I think generally generics are useful, but tend to use generic functions in my C++ code more than generic types or containers.
Might be worth noting that I work in game development, where using a library data structure isn't really considered for most things, because a special purpose one is likely able to be higher performance, due to assumptions it's able to make, operations it doesn't need to implement, etc. I guess I can see the situation where not being able to find an off the shelf generic data structure is frustrating, but I think it might be a bit overblown.
Lock-free concurrent data structures always struck me as the killer use case for generics as far as data structures are concerned: reasoning about atomics (especially when you have to annotate with relaxed/acquire/release/sequentially consistent ordering) is really tricky and one small mistake will lead to incredibly difficult-to-reproduce bugs. You really don't want to rewrite those over and over. Even something like a j.u.c-style concurrent hash map with striped locking needs careful thought around growth and reallocation lest you deadlock; it's so much easier to just have someone else write the data structure.
The idea of cutting/pasting code seems pretty backwards to me compared to just using one of these libraries:
http://java-performance.info/large-hashmap-overview-jdk-fast...
Exactly.
It might be a bit difficult to predict what the input to the compiler actually looks like after CPP has done its job. So we could restrict the kinds of things that CPP can do to make it more predictable. Let's say CPP can only substitute symbols in places where types are asked for...
I don't think Go is really about catering to hordes of cheap junior devs or rebelling against Lisp or Unix, etc. It's (in my interpretation) a way to use channels and goroutines without overwhelming people with academic syntax. As a tool, "the right job" is small and focused on throughput and concurrency - not wiring up webs of generic abstractions. Obviously some people are also finding the language general enough to do the latter, but not without a bit of frustration.
If my understanding is reasonable, people shouldn't be asking "why doesn't Go have generics," but rather "should I be using Go for this? Does this program even have one 'chan' in it?"
Rob Pike disagrees with you.
So is Scala. Meaningless.
> .. anti-unix
Actually, the issue is precisely that Rob and team are entirely too comfortable on _nix, so there is a (typical for Go) duality where (L) language is syntactically light and semantically small, (R) *nix is your IDE and piping source through source level transformers are assumed to be 2nd nature to the user.
[p.s.]: (L) is motivated by making another work-day tool for the broadest range of developers and experience levels, and (R) possibly emerged as a de-facto approach given the zippy compiler and source level parsers built into the Std. lib, and comfort at that level is a virtue of a much smaller set of developers that map into (L).
Strange, I thought Go was all about the unix philosophy ?
> It's possible to use these ideas to construct something analogous to type-safe Unix pipes.
> designed to not give a team of junior devs enough rope to hang themselves
yeah, I thought Google only hires PhDs ... if a noob from a code bootcamp can understand generics within 6 weeks, certainly a guy from a prestigious university hand picked by google can ... please stop being insulting.
I find this to be an utterly silly goal.
It is also not one language that one can use on the interview exercises.
So given Rob Pike's statement about an easy language for newly hires, one has to wonder where they are coming from.
More like they can't add generics without breaking the backward compatibility of their mediocre type system.
Reference? Proof?
(Genuinely interested.)
My problem with the Go team's line of reasoning is that they always cite Java (erasure gives problems) and C++ (too long compile times), but there are many languages that have implemented parametric polymorphism nicely ages ago (C#, ML, Haskell).
Sometimes I wonder if it is the lack of a person like Philip Wadler on the Go team who can hash out how to best fit generics into Go's type system.
Sounding more like a patronising, holier than thou language to me.
And given that Java implemented generics (sort of) and the world didn't collapse indicates to me junior developers could handle it.
I suspect that the OP might not have used Go for very long if he didn't discover any of these solutions.
I personally don't use them because the emacs version never worked quite right for me and at that point I'd rather just deal with Go directly and not have an additional annoyance.
I agree that it is a shame that these options aren't built-in and many newbies would have to discover them on their own.
With C, which is custom-data-structure land (and besides pointer use is encouraged) you can get by.
But Go purports to be safe and modern and uses a GC.
Therefore, I think they should make Go suitable to be used as an intermediate language (the compiler would become a back-end). Having Go as a compiler back-end could spawn a lot of interesting new languages.
The JVM can support goroutines (similar to Quasar Fibers) and the various GC implementations are significantly more advanced than what Go has. But I suspect Go is popular mostly for its simplicity than features.
Definitely. I think there are a couple of reasons:
* It's familiar to anyone who knows C-like languages.
* The tooling is really good for such a young language (go get, go fmt, goimports, pprof, etc.).
* Java's tooling is still lightyears ahead, but a significant chunk of programmers dislikes the JVM for various reasons.
* It integrates well with the UNIX/C environment.
* Simplicity (someone mentioned in a thread a couple of days ago: Go seems to look what C++ does, and then does exactly the opposite).
Having written so much Go code, there were situations where generics would have been useful, but they were of a very different kind. Generics and typed unions would be very helpful for writing compilers (what I do). But even then it's a matter of making code clearer and and more type safe, rather than saving time/not writing data structures.
Other than compilers I have never felt the need of generics in Go, and I have never felt the need to write data structures containing interface{}. interface{} is very useful as a parameter type, to reflect on it, but it's almost always a mistake to use it inside a data structures (there are exceptions, but people usually abuse it).
There's no question that some form of generics would be useful to some people in some context, but its necessity is overblown.
Yes you want a generic implementation of a data structure. Because it's dry. Now personally I stopped arguing with Gophers about that. This kind of arrogant speech is exactly why I dropped Go. Let me decide what I want and what I need , and how to implement it . If the language doesn't allow that then I definitely dont need that language.