On teams of enough engineers, especially as those engineers rotate in and out of maintenance roles, richer abstractions tend to obfuscate intent and hurt maintainability more than help it.

For me it's mainly due to the lack of generics and helper tools that naturally spawn from generics. Go's solution to having less LOC inside a function is _only_ to create helper methods/funcs. These helpers are bound to specific data structures and often are only applicable to that one use case in that one method that they came from. Your ability to make that helper function apply to more code depends heavily on what data structures are being used.

This form of non-generic bloat was far worse for me than the classic examples of `Fooi32()` and `FooString()`. The latter I don't mind at all.

Rust aids this process to me by, well, having generics. Converting one data structure into another becomes a breeze, and converting a Slice of those data structures is just as easy. I'll be really interested to see how Go feels once generics make it in - though I suspect I'll stick with more explicit memory management.

Speaking of memory management, it would be nice if Go had a way to intelligently indicate which structures are safe for concurrent use and which aren't. I've found in Rust that the explicit nature of the concurrent safety is not something I knew I wanted in Go. I know Go will not likely ever match Rust's safety, but if it at least indicated that something was an thread-unsafe implementation that would be golden.

[1] edit: Oh, and there is a disclaimer in this that the projects I'm involved in seem to heavily rely on managing several data structures in repeated ways. Aka, very generic friendly.. so not having generics made my projects extra painful. This is not likely the norm.

This has been my experience as well. People just parrot what the golang authors say about it being a language suites for "large scale development", whatever that means, without providing anything to back it up. Just because they said it was must be it's true? In practical experience, it's a verbose, clunky, and awkward language to use for any non-trivial code bases.

Rust is still too young to take advantage of this, but in due time it will.

I actually find Rust's standard library great. I think my only complaint so far is the lack of time handling. Luckily community packages (like Chrono, Diesel, Tokio, Hyper, Itertools) have implemented great solutions to a few missing details.

Hell, Iterators alone make my life so much better.

Then the team behind Rust after many research found that you can actually guaranty memory safety if you can track how data are used by each thread. To do this you need to track how each value is used and borrowed by each thread to insure memory safety. And so the borrow checker was born. It gives you the best of both world: speed And safety while adding only a little bit of time to the program to compile.

If Rust is slow to compile it’s not because of the borrow checker it only adds a little bit to the bottom line, but it’s because of Generics and other things. Which is also something that Go is lacking and doesn’t have yet.

... and yet experienced gophers will criticize newcomer's code for using channels too much and they will tell you you need to use traditional stupid shared memory approach instead of channels in many cases.

The maintainability of the code is mostly determined not by how good the underlying language is but how good is that DSL. To take an example that's been widely discussed in HN, let's think about machine learning - the look, structure and maintainability of your code is highly influenced by the choice of your framework/"DSL" e.g. pytorch vs tensorflow1.x vs straight numpy vs Caffe will have major differences despite being in the "same language".

And in that regard, the features of the core language matter only in regard of whether they facilitate making a good DSL. Of course, some languages fail because their choices result in the "core DSL" i.e. standard libraries that everybody uses being fragmented and confusing; Go has it quite good IMHO for the standard library; but for every specific domain it also matters if the DSL for that domain is good, whether it's an ORM layer or a OpenGL interface or a deep learning framework or an physics simulation library - whether the language abstractions mean that these DSL's (often being interfaces/wrappers to some third-party code written in another language) are going to be good and convenient, or whether the language structures limit the DSL so it becomes unwieldy to use.

Every time I've seen this, it's been a disaster. Custom DSLs for business logic are, 100% of the time, a red flag of a development process gone off the rails.

(We may work in different industries.)

Go prevents you from doing this at a very low level. This is a huge advantage.

And it's worth noting that many (all?) of these particular systems were actually initially made as custom DSLs for internal use within a single company - Rails for 37signals/Basecamp, Pandas for AQR Capital Management, Django for Lawrence Journal-World, Tensorflow by Google Brain, etc. They were the exact thing that you describe - a custom DSL invented for the business logic that the particular company needed repeatedly. And they were not a disaster but the right thing to do in their case - mostly because they had the ability to make a somewhat good DSL though in all those cases it was an 'okayish' DSL initially and became good only through years of substantial changes. Yes, for every such case there are many bad internal DSLs. "Not Invented Here" is a common problem. You need to ensure that you're using good DSLs, and good DSLs generally gain wide adoption within that domain instead of every company building their own shoddy DSL.

I've seen a bunch of nightmarish financial systems with horrible DSLs. However, even in that case the DSL problem was that this DSL was horrible not that a DSL shouldn't have been used - any fix or rewrite would generally require replacing it with a better DSL/framework/whole-encompassing-structural-API/whatever, instead of sticking to the core programming language (which would just result in another emergent "DSL" of some core libraries/classes, which would be lousy compared to something you intentionally design to be usable).

The differences between programming languages are smaller than the differences between DLS's implemented in the same language. Code in Rails (Ruby) and Django (Python) has more in common than code for Django and Pandas. Writing Tensorflow code in Python is somewhat similar to writing TensorflowJS code in Javascript, but very dissimilar to writing Pytorch code that does the same thing.

They're not really independent libraries, they're frameworks that suggest, influence and sometimes even mandate most of the infrastructure around your code and the structure of your code in a much more stringent manner than the core language does. There's a fundamental difference between making a few calls to a black box library versus having the content and style of most of your code being dominated by the traits of that "library" - the large frameworks overwhelm the small(ish) core languages. So I believe it's justifiable to consider the frameworks I mentioned as DSLs.

So you're right, for programming in the large it's very important how well the language supports creating good reusable APIs and data structures. I think that's why so many people make such a big deal about generics in Go, though personally I think it's a poor example and the complaints are tiresome. I think I'd focus more on whether the language has things like macros and lambdas/continuations that let non-API idioms be expressed consistently and conveniently, as though they're part of the language (even though I think creating a true DSL is usually a bad idea). Not sure Go meets this standard TBH.

Well, nothing I'm aware of beats the LISP family of languages as far as "making a good DSL" goes, but none of those are particularly popular :(

But that’s another topic.

   (defclass foo (superfoo1 superfoo2)
      ((some-slot :type bar)))

   (defmethod baz ((some-foo foo) (some-bar bar))
     ...)

   (defmethod foobar ((some-foo foo))
     (baz foo (make-bar)))

Some tool that can point where in a 1M lines of code project those classes are being used, and if they would need to change.

That's a good thing. But then comes the git merge (or any other automatic code altering tool). How do you know something hasn't been messed up in the process ? (unless you wrote unit test for every argument of every function)

Lisp has compilers.

Some do also some type checking, here SBCL 1.5.2:

  (declaim (ftype (function (fixnum) fixnum) foo))
  (defun foo (a)
    (+ a 1))

  (declaim (ftype (function (fixnum) string) bar))
  (defun bar (a)
    (if (zerop a)
        "zero"
        "not-zero"))

  (defun baz (a)
    (foo (bar a)))

  * (compile-file "/tmp/test.lisp")
  ; in: DEFUN BAZ                                                                                                                                                                      
  ;     (FOO (BAR A))                                                                                                                                                                  
  ;                                                                                                                                                                                    
  ; note: deleting unreachable code                                                                                                                                                    
  ;                                                                                                                                                                                    
  ; caught WARNING:                                                                                                                                                                    
  ;   Derived type of (BAR A) is                                                                                                                                                       
  ;     (VALUES STRING &REST T),                                                                                                                                                       
  ;   conflicting with its asserted type                                                                                                                                               
  ;     FIXNUM.

I had convinced myself I was reading Clisp, instead of CLOS.

Common Lisp has an object system with classes since the early 90s. In fact it was the first standardized object-oriented language.

Common Lisp also has optional type declarations. Not very sophisticated, but they are there.

Check out the SBCL implementation, which has a compiler which does some static type checking/inferencing.

http://www.sbcl.org

But one usually can use types/classes at runtime, too.

That's what some Go proponents say, not necessarily reality, and there's no study about this, just hearsay.

I find Go hurts readability for larger programs because it prevents many structural abstractions that make code more clear and concise (as opposed to mere fancy "clever" code, which it still allows), and because it forces too much boilerplace (error handling, no generics, etc).

Every large projects contains re-implementations of the same things (from data structures to algorithms, to parallel processing handling) that could have been done once and re-used forever with generics and a little more power.

Not maybe because of the lack of generics?

If I recall correctly one of the arguments for lack of generic was that they would make more obvious code, hence easier to follow and read. Actually it seems like the lack of generic just make more obvious code, hence more difficult to read.

The other argument was compilation speed, that honestly I cannot complain about.

I don't believe that is a reason ever articulated by the core team, but a post-hoc justification by apologists. The reason the core team gave was that they had not settled on a design that would fit well with their other goals for the language, and it simply didn't make the short list for 1.0.

Compared to python or Java? Well written python basically resembles Go. And most Java I've come across tries to spread logic across class boundaries so you never get to have all the code in front of you at one time.

I attribute it more to a static type system that lacks features I've grown use to in modern languages. Most of my experience in Java is 1.6 and prior so its possible code bases have improved since then.

Could you give a couple of examples how this is made clearer in other languages?

Edit: And the intrusive data structures commonly used in C are way more complicated than the equivalent generic code (in my opinion).

https://lwn.net/Articles/184495/

A red-black tree implemented in the same simple style by an OpenBSD developer a couple of decades later was adopted almost overnight by all BSDs: https://man.openbsd.org/tree#EXAMPLES

Unlike the Linux version, the BSD version is as type-safe as a C++ template, permits a node to exist in as many collections as you want (one member field per collection), and like a template the BSD version can inline the comparison function. You even have the option of making all the routines extern or static, which you can't even do with C++ templates.

What the Linux version has going for it is that it doesn't need to rely on the preprocessor so the implementation looks a little neater and more like a C++ template--no columns of backslashes as required with preprocessor-based generation. But that's a horrible tradeoff. queue.h and tree.h have basically been frozen for decades.[1] After initial curiosity wanes (day #2 of many thousands), nobody reads the headers because the interface is simple and never changes.

[1] Various systems have tweaked things here and there (e.g. to add parameterized trace points), but the original APIs never changed and code has always remained portable at the source level. The only big change over all that time was most systems (BSDs, macOS) eventually dropped CIRCLEQ after someone (I think a FreeBSD developer) posted benchmarks showing it sucked relative to TAILQ. I actually preferred CIRCLEQ; the entire header is small enough that for awhile I just copied it around everywhere.

- Download one that uses interface{} / void pointers - okay, but not very efficient or type-safe.

- Download one that's intrusive - okay, but pretty convoluted.

- Write my own - a pain.

- Code generation - generics but worse.

(This isn't sarcastic.)

But if you are working on a project where the details of the tree do not matter (so you can use a library), you will probably use C++, not C. I still have a hard time understanding what niche Go fills, it's too high level to compete with C or Rust, but here we are comparing its expressiveness with C.

The overhead of interface{} isn't as high as people sometimes think. You have the overhead of boxing the value, but the actual type checking is just a single comparison and jump.

This is way too much in a tight loop. I want my structure to fit in the cache and I want to have as little tests as possible.

If that is a hard requirement, then the answer is just that Go isn't a suitable language. AIUI the current generics proposal in Go, it won't fix that problem for you either.

Where I disagree with the programming community at large is how often that's actually a hard requirement. A lot of programmers grossly overestimate their performance requirements on the microscale [1], and so concerns about "premature optimization" remain alive and well decades later. However, there are undeniably still a non-trivial number of cases where that really is a hard requirement, in which case the answer is, don't use Go.

(Preferably, figure that out before you start. As a fairly heavy Go user, I actually kinda cringe when reading one of the several "We sped up Go by 10x on our particular big thing and got to barely-acceptable performance" articles, rather than consider it something to be celebrated. I mean, cool that you sped it up that much and all, but honestly the correct answer was to notice in the design phase that you were doing something you probably shouldn't have done in Go, modulo concerns about easy prototyping and such.)

[1]: I qualify with "microscale" because I also think a lot of programmers underestimate the performance requirements on the macro scale. But that's actually a very different issue. There's a lot of very sluggish and slow code out there in the wild that is architecturally wrong; at every individual step the code is reasonably optimized and fast, the problem is having the wrong steps.

But yes, if you're writing performance critical code (which you generally won't be), then you'll need to use code generation to avoid boxing.

This kind of discussion is pointless without some kind of real world reference point. The vast majority of Go projects will never run into a situation where the overhead of using interface{} is a performance bottleneck. If you're working on a project where you consistently have to think about the cache performance of your code, then C++ or Rust would probably be better choices. Go isn't meant to be all things to all people.

I have different kinds of constraints, mainly propositional clauses (a or b or c) and pseudo-boolean constraints (2a + b + c >= 2). Long story short, a propositional clause is a special kind of pseudo-boolean constraint, so I can process my clauses as if they were regular pseudo-boolean constraints, but this is slower than using dedicated code.

Lots of problems only deal with propositional clauses, so they take longer to solve than they should, because I cannot say "this is a problem with only propositional clauses", vs "this is a problem with all kinds of clauses". I could use interfaces, and tried, but this is way slower, to the point it doesn't make sense to use them.

And I can't use C++ or rust for this problem: there are already tons of very good c++ SAT solvers, but my goal was to provide a SAT solver I can use in my go programs without resorting to cgo black magic.

[1]https://github.com/crillab/gophersat

Yes, fair enough, I think that is a good example of the kind of code that Go is not particularly suited to (if performance is critical).

The cognitive overhead certainly is. Computers exist to help us do things faster and with fewer mistakes. This is a prime example of using a computer to help us make more mistakes, worse.

There is no doubt that not having generics makes some code worse than it would be with generics. But I think the problem is often overstated.

I honestly wonder if go isn’t cursed to follow the same road.

Blaming just generics for that does not work imho; in Java people were writing really insanely over complex code close after the first release; I had an ex C++ guy before 2000 writing Java code that used every design pattern in The Book, everywhere he could and it ended up with 1000s of files and those strange long names that became the standard for Java a few years later.

But yeah, I share your concern. It would be nice to have a better way to write container types in Go, but generics always seem to end up getting abused.

You can build any abstraction you want in C.

This is disingenuous. Show me Pi types in C.

I took a quick look at what these are. Couldn't they implemented with macros like (this builds a type "Utf8String" which "depends" on the value of the "ENCODING" parameter to figure out what set of functions to call for the guts of the functions that handle the type, e.g. RawStringUtf8Length):

    enum Encoding { Ascii, Utf8, Utf16 };
    
    #define TYPENAME Utf8String
    #define ENCODING Utf8
    #define CHARTYPE char
    #include "stringdef.ph"

    typedef struct { size_t len; CHARTYPE* text; } TYPENAME;
    
    #define MAKEFUNC(name) TYPENAME##name
    #define RAWSTRINGFUNC(name) RawString##ENCODING
    
    void MAKEFUNC(Create)(TYPENAME* str);
    void MAKEFUNC(Destroy)(TYPENAME* str);
    void MAKEFUNC(CreateFrom)(TYPENAME* str, const CHARTYPE* text);
    
    /* this goes in a single C file, single header library style */
    #ifdef IMPLEMENTATION
    
    void MAKEFUNC(Create)(TYPENAME* str) { ... }
    void MAKEFUNC(Destroy)(TYPENAME* str) { ... }
    void MAKEFUNC(CreateFrom)(TYPENAME* str, const CHARTYPE* text) {
        str->len = RAWSTRINGFUNC(Length)(text);
        str->text = RAWSTRINGFUNC(Duplicate)(text);
    }
    
    #endif
    
    #undef MAKEFUNC
    
    /* cleanup input parameters so includers wont need to do it */
    #undef CHARTYPE
    #undef ENCODING
    #undef TYPENAME

    Utf8String u8s;
    Utf8StringCreateFrom(&u8s, "foo...");

Interpreter, but close enough :)

https://github.com/codr7/g-fu/tree/master/v1

Embedding a higher level language that's controlled from a more primitive host language allows solving many otherwise tricky problems without the struggle.