Hacker News Reader

Comment by phyrex | Hacker News Reader

Frankly, this article has the opposite effect of convincing me. The author starts with a perfectly reasonable data model and munges it for some questionable space savings (that only apply in the dumbest of databases that don't do any sort of compression), for the price of a lot of boilerplate, and significantly less insight into the data if you look at it in serialized format (either directly in the database or on the wire). Maybe someone can explain to me how this is supposed to be better?

View on HN

Bringing Java's premature over-engineered over-abstracted practices to Go.

Java has Enums, so why would you bother with all of this?

You aren't wrong. I think it depends on what your needs are. As well as what the scenarios are.

If you are serializing this data a lot, then having ints vs strings is a significant advantage.

Again, it depends on your use case. If you don't care about serialization size, then I agree, it's extra effort that you may not need.

One of Go's strengths is that it should make it easier to swing between the optimized abstraction and the less-optimized abstraction. If you start with that data represented as strings and then find yourself backed into a performance corner and need to change the representation, Go types and type-based compile-time method selection can make it a bit easier to make that change.

Even in this era of fast computers and preposterously large storage, one sometimes hits a situation where the way to get the improvements you need is to start number-representing your strings.

Is this compared to a good language or (say) server-side JavaScript because the languages I tend to use allow me to do that and Go looks looks very Algol-68-y from that perspective as opposed to some hyper abstract wonder-language.

What languages do you tend to use?

I work for the D foundation, so mainly D for things I get to choose.

That's a great point. Anyone who has spent significant time with Go knows how easy it is to do a large refactor (or even a minor one) due to it being a types language.

And yes, you may not start with ints, but you could easily add code for the marshal/unmarshal later on to convert those strings to ints for serialization purposes. And it wouldn't require a change to any of your other code, not to mention that if you stored these values in a database already you don't need to perform a migration either.

Go has many shortcomings that make it difficult and extremely annoying to do refactorings. Things like no constructors where types are declared as follows

    A {
        Field1: field1,
        Field2, field2,
    }

Now adding a new field to A would require ensuring that all code paths initialize Field3, otherwise you're going to have silent errors at runtime. This has been solved ages ago in Java and C# and similar languages by means of constructors.

A constructor is just a function though... If I add a new field to a Java class and fail to add its initialization to the constructor, I have the exact same problem because Java initializes the field to its default value when the constructor is called.

You are correct that every situation where the struct in Go is initialized "bare" would need to be addressed if a field is added, but Go considers this a feature, not a bug (and, conversely, considers "bare initialization" of structs at dozens of places in your code to be bad practice if that struct could ever grow new fields). If you're bare-initializing structs, you're comfortable using them in a "loosey-goosey" context where zero-initialized fields are permitted (or, better, useful... https://www.youtube.com/watch?v=PAAkCSZUG1c&t=6m25s). In Go, the issue of required structure is addressed by wrapping the struct in an interface and then providing a function in the package that can create an instance of the interface. Used in that way, you get something very similar to a Java class (though Go doesn't force you into the "everything is a class" paradigm that Java demands).

> but Go considers this a feature, not a bug

I see this dismissive philosophy in golang a lot (I'm not saying you're doing it, mind you). The entire language is full of arbitrariness. I've seen enums being dismissed by the golang team just because, without providing any meaningful arguments. Same with how nullability was not addressed in the language, contrary to how proper modern languages have tacked the issue. The excuse? "that's how the underlying machine operates". Quite meaningless and dismissive really.

Time and time again, it's been shown that the goal is to have a simplistic (not simple) language that also makes it easy to write the compiler for. This breaks down at larger scales because reality is complicated.

Usually you would create a constructor without that parameter which sets the field to something backwards-compatible.

This perspective ignores the fact that there is no shortage of typed languages that don’t have all the—in 2021—inexcusable downsides of golang.

There is unfortunately a shortage, still in 2021, of languages that have a null set of inexcusable downsides.

I will trade - unhappily - ADTs for value types, automatic memory management, some language-level concurrency support, a compiler that builds our largest project in under two minutes, and a community large enough I can spend my time training new hires on fundamentals and business problems and not tool onboarding.

.NET Native, OCaml, Eiffel, Common Lisp, D, Nim just for starters.

Value types are almost there in Java, but without deliberately trashing the GC you can get away without them, and it has all the other qualities (and will have ADTs as well, sealed classes are already in preview)

That boiler plate usually can be generated for you. See Go's stringer which does all the magic for you. All you have to do is define the type and constants, and a go generate directive.

Code generators don't solve the problem of the language being too verbose/boiler plate. Instead of expressing a concept in a simple fashion, you have to generate 100s of lines of code that needs to be maintained across versions of golang and your data model.

And install stringer in every dev env, and debug all the boilerplate anyway when things go wrong.

Code generation is always the last resort.

I'm not saying it's the solution. I'm just pointing what is idiomatic go. Certainly there are arguments for and against code generation.

Where I work we are a go house, and the number of tools I've seen in our pre-commit configurations and in our development environment setup scripts is pretty intense. So is the number of questions around said tooling. There's a lot more friction than I'd expect. I can't even use lack of experience as an excuse for these questions because they're really talented people. Some of them have worked on the product for years, and have literally moulded the product's tooling into what it is today.

would you have a sample code? sounds interesting

Example from the docs:

    // pill.go
    package painkiller
    
    //go:generate stringer -type=Pill
    type Pill int

    const (
      Placebo Pill = iota
      Aspirin
      Ibuprofen
      Paracetamol
      Acetaminophen = Paracetamol
    )

Then running `go generate` will yield:

    $ cat pill_string.go
    // Code generated by "stringer -type=Pill"; DO NOT EDIT.
    
    package painkiller
    
    import "strconv"
    
    func _() {
     // An "invalid array index" compiler error signifies that the constant values have changed.
     // Re-run the stringer command to generate them again.
     var x [1]struct{}
     _ = x[Placebo-0]
     _ = x[Aspirin-1]
     _ = x[Ibuprofen-2]
     _ = x[Paracetamol-3]
    }
    
    const _Pill_name = "PlaceboAspirinIbuprofenParacetamol"
    
    var _Pill_index = [...]uint8{0, 7, 14, 23, 34}
    
    func (i Pill) String() string {
     if i < 0 || i >= Pill(len(_Pill_index)-1) {
      return "Pill(" + strconv.FormatInt(int64(i), 10) + ")"
     }
     return _Pill_name[_Pill_index[i]:_Pill_index[i+1]]
    }

Note: indentation is messing up because of Hacker News, so ignore the indentation.

I think this is mostly a toy example. In the real world the Genre would have it's own data, books can belong to multiple Genres, and you'd need the concept of SubGenres. So you'd have:

Book

Genre

BookToGenre

GenreToGenre

Reply on news.ycombinator.com