I'm confused about this:

import std.conv;

void main()
{
    enum s = "`1`.to!int;";
    enum c = __traits(compiles, mixin("{auto a = new "~s~";}")); // line 1
    mixin("auto a = "~s~";");                                    // line 2
}

This does not compile, giving errors about instantiating std.conv.to!(int).to!(string). If I switch the order of lines 1 and 2 it compiles. Is this a bug or something I don't understand about __traits(compiles)?

On Tuesday, 5 March 2013 at 07:53:15 UTC, cal wrote:
> I'm confused about this:
>
> import std.conv;
>
> void main()
> {
>     enum s = "`1`.to!int;";
>     enum c = __traits(compiles, mixin("{auto a = new "~s~";}")); // line 1
>     mixin("auto a = "~s~";");
>  // line 2
> }
>
> This does not compile, giving errors about instantiating std.conv.to!(int).to!(string). If I switch the order of lines 1 and 2 it compiles. Is this a bug or something I don't understand about __traits(compiles)?

You can't test declarations inside of __traits(compiles), only expressions. It's in the docs: http://dlang.org/traits.html#compiles

On Tuesday, 5 March 2013 at 08:04:12 UTC, Andrej Mitrovic wrote:
> You can't test declarations inside of __traits(compiles), only expressions. It's in the docs: http://dlang.org/traits.html#compiles

So why does this work:

import std.conv;

void main()
{
    enum s = "`1`.to!int;";
    enum c = __traits(compiles, mixin("{auto x = "~s~";}")); // true
}

On Tuesday, 5 March 2013 at 08:04:12 UTC, Andrej Mitrovic wrote:
> On Tuesday, 5 March 2013 at 07:53:15 UTC, cal wrote:
>> I'm confused about this:
>>
>> import std.conv;
>>
>> void main()
>> {
>>    enum s = "`1`.to!int;";
>>    enum c = __traits(compiles, mixin("{auto a = new "~s~";}")); // line 1
>>    mixin("auto a = "~s~";");
>> // line 2
>> }
>>
>> This does not compile, giving errors about instantiating std.conv.to!(int).to!(string). If I switch the order of lines 1 and 2 it compiles. Is this a bug or something I don't understand about __traits(compiles)?
>
> You can't test declarations inside of __traits(compiles), only expressions. It's in the docs: http://dlang.org/traits.html#compiles

The errormessage is rather cryptic though.. Makes it look like there is something with conv rather than compiles.

On Tuesday, 5 March 2013 at 08:09:37 UTC, cal wrote:
> On Tuesday, 5 March 2013 at 08:04:12 UTC, Andrej Mitrovic wrote:
>> You can't test declarations inside of __traits(compiles), only expressions. It's in the docs: http://dlang.org/traits.html#compiles
>
> So why does this work:
>
> import std.conv;
>
> void main()
> {
>     enum s = "`1`.to!int;";
>     enum c = __traits(compiles, mixin("{auto x = "~s~";}")); // true
> }

Hmm.. And this also works:
enum c = __traits(compiles, mixin("{auto a = new 1;}"));

Something to do with CTFE combined with mixins?

On 3/5/13, cal <callumenator@gmail.com> wrote:
> So why does this work:
>
> import std.conv;
>
> void main()
> {
>      enum s = "`1`.to!int;";
>      enum c = __traits(compiles, mixin("{auto x = "~s~";}")); //
> true
> }

That's a function literal, i.e. an expression.

On Tuesday, 5 March 2013 at 08:14:58 UTC, simendsjo wrote:
> Hmm.. And this also works:
> enum c = __traits(compiles, mixin("{auto a = new 1;}"));
>
> Something to do with CTFE combined with mixins?

But it gets this right, in that c is false. I had thought that by wrapping the declaration in braces, making it a function, that got around the 'no declarations' limitation.

On 03/05/2013 08:53 AM, cal wrote:
> I'm confused about this:
>
> import std.conv;
>
> void main()
> {
>      enum s = "`1`.to!int;";
>      enum c = __traits(compiles, mixin("{auto a = new "~s~";}")); // line 1
>      mixin("auto a = "~s~";"); // line 2
> }
>
> This does not compile, giving errors about instantiating
> std.conv.to!(int).to!(string). If I switch the order of lines 1 and 2 it
> compiles. Is this a bug or something I don't understand about
> __traits(compiles)?

Compiles as expected with DMD 2.060. It is probably a regression.

On Tuesday, 5 March 2013 at 12:58:57 UTC, Timon Gehr wrote:
> Compiles as expected with DMD 2.060. It is probably a regression.

Ah you're right, also with 2.061. I'll file.