My previous question was in the context of overloading opEquals. Here it is again in a simpler form.

void foo(const(char)[] s)
{}

void foo(const(wchar)[] s)
{}

void foo(const(dchar)[] s)
{}

void main()
{
    foo("hello");
}


Compilation error:

deneme.d(10024): Error: function deneme.foo called with argument types:
	((string))
matches both:
	deneme.foo(const(char)[] s)
and:
	deneme.foo(const(dchar)[] s)


Adding overloads for string, wstring, and dstring does not help. Same problem...

Replacing "hello" with either of "hello"c, "hello"w, or "hello"d fixes the issue.

This is a bug, right? I've been assuming that unqualified string literals were immutable char arrays, but the behavior is different between "hello" vs. "hello"c.

Am I missing something?

Thank you,
Ali

On Sun, 11 Apr 2010 15:33:09 -0400, Ali Çehreli <acehreli@yahoo.com> wrote:

> This is a bug, right? I've been assuming that unqualified string literals were immutable char arrays, but the behavior is different between "hello" vs. "hello"c.
>
> Am I missing something?

"hello" is typed as a string *only* if you are using at a string.  If you are using it as a wstring or a dstring, then it is typed that way.  You can even use it as a const(char) * and it becomes an ASCII C-style string with a zero terminator!

This way, you can do things like this without casts, conversions, or suffixes:

dstring ds = "hello";

Unfortunately, this leads to the problem, what version of foo to call when supplied with just a literal?  It can call all three!

I don't like the implementation -- give an error -- but it's not an unreasonable choice.  I'd file a bug and see what happens, perhaps Walter can change it.  I'd recommend assuming the type that occurs when using auto.

-Steve

Steven Schveighoffer wrote:
> On Sun, 11 Apr 2010 15:33:09 -0400, Ali Çehreli <acehreli@yahoo.com> wrote:
>
>> This is a bug, right? I've been assuming that unqualified string
>> literals were immutable char arrays, but the behavior is different
>> between "hello" vs. "hello"c.
>>
>> Am I missing something?
>
> "hello" is typed as a string *only* if you are using at a string.  If
> you are using it as a wstring or a dstring, then it is typed that way.
> You can even use it as a const(char) * and it becomes an ASCII C-style
> string with a zero terminator!

I did not know that. :)

Could you please share some guidelines about chosing the type of the D-string to use at the interface...

Should applications stick to one of these types and require callers to convert explicitly if needed? Should the common type be dstring? On the other hand, string seems to be a better choice because classes have the common toString member functions that return string.

In my case, I have a set of classes that represent alphabet letters and alphabet strings. The motivation is to provide logical sorting and capitalization. (To me, even for the English alphabet, â should be sorted between a and b.)

Since I want these types to be used as seamlessly as possibly, I wanted to provide opEquals overloads for char[], wchar[], and dchar[]. Should I not bother doing that? In fact, I really shouldn't due to this compiler error.

Should my classes only interface with dchar and dstring?

Thank you,
Ali

On Mon, 12 Apr 2010 00:40:44 -0400, Ali Çehreli <acehreli@yahoo.com> wrote:

> Steven Schveighoffer wrote:
>  > On Sun, 11 Apr 2010 15:33:09 -0400, Ali Çehreli <acehreli@yahoo.com> wrote:
>  >
>  >> This is a bug, right? I've been assuming that unqualified string
>  >> literals were immutable char arrays, but the behavior is different
>  >> between "hello" vs. "hello"c.
>  >>
>  >> Am I missing something?
>  >
>  > "hello" is typed as a string *only* if you are using at a string.  If
>  > you are using it as a wstring or a dstring, then it is typed that way.
>  > You can even use it as a const(char) * and it becomes an ASCII C-style
>  > string with a zero terminator!
>
> I did not know that. :)
>
> Could you please share some guidelines about chosing the type of the D-string to use at the interface...
>
> Should applications stick to one of these types and require callers to convert explicitly if needed? Should the common type be dstring? On the other hand, string seems to be a better choice because classes have the common toString member functions that return string.
>
> In my case, I have a set of classes that represent alphabet letters and alphabet strings. The motivation is to provide logical sorting and capitalization. (To me, even for the English alphabet, â should be sorted between a and b.)
>
> Since I want these types to be used as seamlessly as possibly, I wanted to provide opEquals overloads for char[], wchar[], and dchar[]. Should I not bother doing that? In fact, I really shouldn't due to this compiler error.
>
> Should my classes only interface with dchar and dstring?

If you had to choose, I'd suggest choosing string.  The reason is simple:

auto s = "hello";
yourFunc(s);

Because the compiler must choose a type when declaring s, it chooses string.

I think perhaps the compiler should use the same rules when calling an overloaded function with a literal.  You should file an enhancement request with bugzilla, see what Walter thinks.  If he doesn't like the idea, he usually shoots it down pretty quick :)

-Steve