On 8/26/12, Andrej Mitrovic <andrej.mitrovich@gmail.com> wrote:
> If I replace it
> with my own hardcoded function below it takes only 5 seconds and uses
> 150 MB RAM. This is what the function looks like:

An even faster version, twice as fast:

template myStaticIndexOf(T, TList...)
{
    static if (!__traits(compiles, TList[0]))
        enum myStaticIndexOf = -1;
    else
    static if (is(T == TList[0]))
        enum myStaticIndexOf = 0;
    else
    // repeat..
}

On 08/26/12 23:03, Andrej Mitrovic wrote:
> Maybe this is CTFE to blame more than the function itself.
> 
> I have a project where I have a TypeTuple that holds a class tree of a wrapped C++ library. The tuple is iterated from in several places where an index has to be retrieved. Compiling this project takes 46 seconds when using staticIndexOf and uses 700 MB RAM. If I replace it with my own hardcoded function below it takes only 5 seconds and uses 150 MB RAM. This is what the function looks like:
> 
> template myStaticIndexOf(T, TList...)
> {
>     static if (is(typeof(T == TList[0])) && is(T == TList[0]))
>         enum myStaticIndexOf = 0;
>     else
>     static if (is(typeof(T == TList[1])) && is(T == TList[1]))
>         enum myStaticIndexOf = 1;
>     else
>     // ... and so on ...
> }
> 
> The body is pregenerated of course, using mixin() would slow down
> compilation here as well.
> 
> When wrapping larger libraries (and hence having a larger TypeTuple) and using staticIndexOf the memory usage becomes so high that the compiler runs out of memory and crashes.
> 
> I really think it sucks that I have to resort to manually pre-generating a function body externally as if I were using a lame (CTFE-wise) language like C++03.
> 
> D *has* to be better than this..

   template staticIndexOf(T, TS...) {
      enum staticIndexOf = {
         foreach(size_t i, I; TS)
            static if (is(typeof(T == I)) && is(T == I))
               return i;
         return -1;
      }();
   }

But i have no idea how it will do - it appears to be faster than the std one, but i couldn't really test it, because the std one keeps failing when i try it with a large enough tuple...

artur

On Sun, Aug 26, 2012 at 11:58 PM, Artur Skawina <art.08.09@gmail.com> wrote:

>    template staticIndexOf(T, TS...) {
>       enum staticIndexOf = {
>          foreach(size_t i, I; TS)
>             static if (is(typeof(T == I)) && is(T == I))
>                return i;
>          return -1;
>       }();
>    }

Is the is(typeof(T == I)) really necessary? If T == I is an error,
then is() returns false, or am I mistaken?

On 08/27/12 00:31, Philippe Sigaud wrote:
> On Sun, Aug 26, 2012 at 11:58 PM, Artur Skawina <art.08.09@gmail.com> wrote:
> 
>>    template staticIndexOf(T, TS...) {
>>       enum staticIndexOf = {
>>          foreach(size_t i, I; TS)
>>             static if (is(typeof(T == I)) && is(T == I))
>>                return i;
>>          return -1;
>>       }();
>>    }
> 
> Is the is(typeof(T == I)) really necessary? If T == I is an error,
> then is() returns false, or am I mistaken?
> 

It didn't look necessary to me either; i took the condition verbatim from Andrej's example. Removing the check made no difference to performance, so i kept it to minimize the diff between both versions.

artur

On 8/27/12, Philippe Sigaud <philippe.sigaud@gmail.com> wrote:
> Is the is(typeof(T == I)) really necessary? If T == I is an error,
> then is() returns false, or am I mistaken?
>

In my OP example I've had to use a typeof() test because otherwise I
could get an out of bounds compile-time error. The is() expression
doesn't seem to gag these errors, e.g.:

void main()
{
    alias TypeTuple!(int) x;
    static if (is(int == x[4]))
    {
    }
}

test.d(17): Error: tuple index 4 exceeds 1

I just noticed something though, if the order is swapped there's no error:
void main()
{
    alias TypeTuple!(int) x;
    static if (is(x[4] == int))
    {
    }
}

Bug maybe?

On Mon, Aug 27, 2012 at 1:00 AM, Andrej Mitrovic <andrej.mitrovich@gmail.com> wrote:

>
> test.d(17): Error: tuple index 4 exceeds 1
>
> I just noticed something though, if the order is swapped there's no error:
> void main()
> {
>     alias TypeTuple!(int) x;
>     static if (is(x[4] == int))
>     {
>     }
> }
>
> Bug maybe?

Certainly.

On 08/27/2012 07:12 AM, Philippe Sigaud wrote:
> On Mon, Aug 27, 2012 at 1:00 AM, Andrej Mitrovic
> <andrej.mitrovich@gmail.com> wrote:
>
>>
>> test.d(17): Error: tuple index 4 exceeds 1
>>
>> I just noticed something though, if the order is swapped there's no error:
>> void main()
>> {
>>      alias TypeTuple!(int) x;
>>      static if (is(x[4] == int))
>>      {
>>      }
>> }
>>
>> Bug maybe?
>
> Certainly.
>

Actually it is according to the specification:

"3. is ( Type == TypeSpecialization )
The condition is satisfied if Type is semantically correct and is the same type as TypeSpecialization."

=> TypeSpecialization is compiled without suppressing errors.

On 8/27/12, Philippe Sigaud <philippe.sigaud@gmail.com> wrote:
> Certainly.

http://d.puremagic.com/issues/show_bug.cgi?id=8593