Thread overview
Allow illegal code in enum initialization?
Oct 23, 2013
simendsjo
Oct 23, 2013
Ali Çehreli
Oct 23, 2013
simendsjo
October 23, 2013
Illegal code is accepted in static if, but not in enum declarations. This leads to having three lines when one is enough. Is this just an oversight, or by design?

template T(alias A) {
    enum T = true;
}

void main() {
    struct S { }
    static if(__traits(compiles, S.a) && T!(S.a)) // ok
        enum e1 = true;
    else
        enum e1 = false;
    enum e2 = __traits(compiles, S.a) && T!(S.a); // No property S.a
}
October 23, 2013
On 10/23/2013 02:55 AM, simendsjo wrote:
> Illegal code is accepted in static if, but not in enum declarations.
> This leads to having three lines when one is enough. Is this just an
> oversight, or by design?
>
> template T(alias A) {
>      enum T = true;
> }
>
> void main() {
>      struct S { }
>      static if(__traits(compiles, S.a) && T!(S.a)) // ok
>          enum e1 = true;
>      else
>          enum e1 = false;
>      enum e2 = __traits(compiles, S.a) && T!(S.a); // No property S.a
> }

I don't know whether it is even specified but it feels like a feature to me.

Just like the shortcut behavior of runtime if helps with avoiding illegal memory accesses, this helps with avoiding illegal code altogether:

// null access avoided
if ((p !is null) && (p.member == 42))

// illegal code avoided
static if (__traits(compiles, S.a) && (S.a == 42))

Ali

October 23, 2013
On Wednesday, 23 October 2013 at 16:27:47 UTC, Ali Çehreli wrote:
> On 10/23/2013 02:55 AM, simendsjo wrote:
>> Illegal code is accepted in static if, but not in enum declarations.
>> This leads to having three lines when one is enough. Is this just an
>> oversight, or by design?
>>
>> template T(alias A) {
>>     enum T = true;
>> }
>>
>> void main() {
>>     struct S { }
>>     static if(__traits(compiles, S.a) && T!(S.a)) // ok
>>         enum e1 = true;
>>     else
>>         enum e1 = false;
>>     enum e2 = __traits(compiles, S.a) && T!(S.a); // No property S.a
>> }
>
> I don't know whether it is even specified but it feels like a feature to me.
>
> Just like the shortcut behavior of runtime if helps with avoiding illegal memory accesses, this helps with avoiding illegal code altogether:
>
> // null access avoided
> if ((p !is null) && (p.member == 42))
>
> // illegal code avoided
> static if (__traits(compiles, S.a) && (S.a == 42))
>
> Ali

The question is if it would make sense to allow it for enum as well as static if. As enum is a compile-time constant I think it would be consistent.