Conversation

ML-style generics are less complicated than Go interfaces. There is no reason for Go not to have generics.

Sep 4, 2018

Seeing the draft, I'm not sure it's practical to retrofit Generics into the lang without making it completely unreadable. The first rule of language design should be: “You got 4 pairs of brackets: (), {}, [] and <>. Use them wisely." A lesson Rust failed to learn, too, though.

Graydon Hoare

@graydon_pub

Sep 4, 2018

I tried; I was overruled.

soc

@oxnrtr

Sep 4, 2018

I believe that Rust 2 will be the point in history where it shows whether Rust is able to only grow new features and will go the way of C++, or if its current designers are able to acknowledge & correct mistakes (like abusing <> for generics) and manage to reduce some complexity.

Patrick Walton

@pcwalton

Sep 4, 2018

Blame me for using <> for generics. I still think it was the right decision among a bunch of bad choices.

soc

@oxnrtr

Sep 4, 2018

It was a decision made with decades of precedent of <> working out poorly, and with dozens of languages as evidence which tried to work around the innate problems and failed. But so is language design ... two steps forward, three steps back. ¯\_(ツ)_/¯

Stephan Tolksdorf

@s_tolksdorf

Sep 4, 2018

Could you elaborate on the problems you're having with <> being used for generics? In C++ it sometimes hurts the readability of expressions specified for non-type template parameters, but other than I can't think of any significant issues.

soc

@oxnrtr

Sep 4, 2018

- Poor readability compared to []. - Almost impossible to parse without sacrificing consistency in other parts of the language. - Leads to inconsistent usage of all the language's brackets. - Has lead to confusing syntax in every lang: <> vs. ::<>, new <A>B<C>(), D.<E>f() etc.

Stephan Tolksdorf

@s_tolksdorf

Sep 4, 2018

It's not clear to me that [] for template parameters would be more readable. Since brackets are also used for array types, indexing operators and lambda capture lists, that syntax wouldn't be conflict-free either, at least in C++. And in the eyes of a C++ programmer '<' being ...

Brendan Zabarauskas

@brendanzab

Sep 5, 2018

Scala uses different notation for array construction and indexing. Seems to work out fine! Here's the ambiguity issue, in Rust at least: github.com/varkor/rust/bl

Replying to

But yeah, shifting things around in the same way for Rust would have other knock on effects, so I won't pretend it's a simple fix or anything. For example Scala's array construction requires variadic arguments, which in turn complicates overloading.

12:51 AM · Sep 5, 2018Twitter Web Client

Replying to

I'm not really seeing this complexity here. You could just do the same thing as vec! – or realize that varargs are annoying to implement for compiler authors, but a necessary part of a language and add them proper, instead of making people go the macro route to emulate varargs.