[Veraticus]

Very insightful. Re: the generics point —

As a Go programmer I always thought the generics complaint was kind of silly in practice — complicated code should be simplified and made more concrete, not more generic.

I’m glad generics were implemented, if only to silence the chorus of people who didn’t even use Go but whined about the lack of them. Their inclusion has simplified the stdlib and led to some cool new functions. Nevertheless I think people implementing them in their own projects is basically code smell and they are a symptom of poorly thought-out code rather than excellent code.

Anyway. This was a good initial post and a good follow-up. Go is my favorite language out there right now for its clarity, power, and ease-of-use. And with loop variable capture coming (the biggest remaining foot gun in the language in my experience) the language is only getting better.

[rowanseymour]

Prior to generics we relied heavily on dynamic typing via interface{}, for example:

  BulkInsert(db *sql.DB, objs []interface{})

This unfortunately meant that any time you had []Foo.. you had to allocate a new []interface{} and copy over the items. Now a function like that can look like:

  BulkInsert[T any](db *sql.DB, objs []T)

And we're not wasting CPU cycles to copy the slice of []Foo. I'm struggling to see how that's code smell or less excellent than using []interface{} or duplicating the code for BulkInsert for every insertable type in our application.

[Veraticus]

Dynamic typing via interface{} is also huge Go code smell though, so...

Generics are definitely better for you. But I would say the overall pattern you're employing of bulk inserting different kinds of data structures with one function is the problem. Of course, I don't know your code so I'm sure you have a good reason for choosing what you did, but a BulkInsert of Any certainly made me raise my eyebrow.

[rowanseymour]

The reason is the same reason that anyone ever writes a generic function (outside of writing a library) - to keep code DRY and avoid duplication. There is no upside to maintaining a large number of identical function implementations, and no scenario in which that is preferable to using generics.

[Veraticus]

But you probably will have to differentiate at some point between what you're inserting. Why not just do that, instead of artificially combine it into one function?

Nevertheless I feel like I'm getting lost in the weeds of this example. Obviously DRY and less duplication is good. However, you have to strike a balance between being clever and being clear. And frankly I would prefer 3 lines of clear code to 1 line of hard-to-read code.

[joshuamorton]

> But you probably will have to differentiate at some point between what you're inserting. Why not just do that, instead of artificially combine it into one function?

Why prematurely optimize for differences that may not (and in practice aren't really likely to) happen?

Keep in mind that the tradeoff with generics is usually not 3 lines of clear or 1 line of hard to read, it's one line of clear code (T -> T), one line of unclear code (Interface{} -> Interface{} + casts), or n lines of complex code (concretely reimplementing the function for your particular case).

[Veraticus]

I would say that a BulkInsert(Any) is a significant premature optimization over just inserting an object as would apply specifically to that object. Because it sounds like you'd have to do weird reflection stuff on the object to determine how and what to insert where.

If you are inserting an object, you should insert it, rather than create complicated generic insert methods that morph themselves based on the object being inserted. That is idiomatic Go.

[z0r]

Writing and reading repetitive code leads to unintentional defects. This is why for loop syntax that directly iterates through a collection is less error prone than the equivalent loop built on indexed look-ups. "Clever" code, at least when it is shorter, is often clearer than "simple" code.

[Veraticus]

I entirely disagree; defects are created by complexity, not repetition.

[didntcheck]

But why is it a problem (assuming the database in this example can handle any type)? If I'm writing a function that returns the 5th element of a list, I will always write that using generics, even if I know it's only used with one type (right now). Not only is it basically free (`<T> T getFifth(List<T>)` is really not any more complicated than `Foo getFifth(List<Foo>)`), it's also a separation of concerns. The logic is just about the container, so no need to complicate it with a "red herring" of a forced type

[ilyt]

> As a Go programmer I always thought the generics complaint was kind of silly in practice — complicated code should be simplified and made more concrete, not more generic.

I guess you have never written a library. It's extremely useful there, stuff like "generic function that runs a channel thru X workers doing f() on it" is now easily possible with full type safety.

> Nevertheless I think people implementing them in their own projects is basically code smell and they are a symptom of poorly thought-out code rather than excellent code.

You can say that about literally any feature used by the incompetent.

But overall yes, they are far more useful for writing libraries than actual applications

[Veraticus]

I have written a library actually. I found interfaces perfectly sufficient for allowing applications to consume the contracts of the library without needing generics. Of course I understand that they have a use and are useful, but it's not like there weren't excellent solutions for this in Go before generics existed.

[icholy]

> Nevertheless I think people implementing them in their own projects is basically code smell and they are a symptom of poorly thought-out code rather than excellent code.

Who needs data structures right?

[Veraticus]

Uh, generics are not data structures and generics are neither the only way nor the best way to interact with data structures.

[lillywastaken]

Do you think it is somehow more elegant to have a specialised version of a data-structure for every single type that may need to be placed into it? Or just to ignore the type whatsoever and erase it via interface{}? Even Golang's designers obviously realise that generics are a better way to interact with data structures because the standard types map and array were always "generic", just special cased as such.

[lenkite]

Yeah, you must hate use the golang map type then. It is a generic data-structure.

[Mesopropithecus]

Comments like this are why I like to joke that the G in Golang stands for gaslighting.

[Veraticus]

Do you actually have anything of value to add to this discussion? Because this comment is pretty bad.

[Mesopropithecus]

I have to apologize, I misread the context of "people whining..", which was in fact about those that don't even use the language. If this was not intended to be aggressive, sorry.

Funny though that I did get triggered by it. Out of the Go community I've heard way too often "you don't really need xyz", when they mean "we're not going to support xyz, here's why, and if you disagree, we respectfully ask you to look elsewhere".

[wolfspaw]

A PL without Generics is just Terrible DX for me, let me abstract the types in my Algorithms and Data Structures Goddammit!

Now Golang is saved, with Generics it's actually an Awesome Incredible PL.

[soulbadguy]

> Nevertheless I think people implementing them in their own projects is basically code smell and they are a symptom of poorly thought-out code rather than excellent code.

I can't think of any other less confrontational way to say/ask , but beside GO which other language have you used ?

Because the above statement is so far remove from my experiences and understand of programming that i suspect we don't we really use the same day to day tools/languages in general.

[jeswin]

> power

What power?

[Veraticus]

The power of voodoo