sort API design

[Johannes Riecken]

On Wednesday, 20 November 2019 at 09:33:25 UTC, Johannes Riecken 
wrote:
> I had looked at how to do case-insensitive string sorting with 
> Phobos and without looking at the documentation much, I went 
> with the following code and then wrongly concluded that `<` 
> does case-insensitive comparisons on strings by default:
>
> ```
> import std.algorithm;
> import std.stdio;
> import std.string;
>
> void main() {
>     auto arr0 = ["foo", "bar", "Baz"];
>     auto case_sensitive   = sort!((a, b) => a           < b
>    )(arr0);
>     auto case_insensitive = sort!((a, b) => a.toLower() < 
> b.toLower())(arr0);
>     writeln(case_sensitive);
>     writeln(case_insensitive);
> }
> ```
> Output:
> ```
> ["bar", "Baz", "foo"]
> ["bar", "Baz", "foo"]
> ```
>
> Later I learned that I was trapped by `sort` being in-place but 
> still returning a value. If this API were to be designed from 
> scratch, I wonder if it would be possible to rule out this 
> incorrect usage at compile-time while still retaining the other 
> nice features of the `sort` function like being in-place and 
> giving strong types which can be used to choose faster 
> algorithms to further transform the result, etc.? Or are there 
> any ideas in programming language research that would help with 
> such a case?

Thanks, that's some nice ideas in this thread! I should 
definitely get into the habit of using const wherever possible.
Jonathan, I think a major selling point of Phobos and D is that 
in lots of cases it can catch erroneous API usage at compile time.