While testing I have found a problem in my code, I have reduced the code to the following lines:

template IsArray(T) {
    const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
                         is(typeof(T.reverse)) && is(typeof(T.dup));
}

TA[] foo(TA, TB)(TA[] a, TB b) {
    TA[] result;

    static if (IsArray!(TB)) {
        if (b.length == 0) {
            result = a;
        } else if (b.length == 1) {
            result = foo(a, b[0]);
        }
    }

    return result;
}

void main() {
    foo("test", "");
}


Its output:

bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
bug.d(21): template instance bug.foo!(char,char[0u]) error instantiating

I have found ways to solve this problem, but can someone please explain me what's the problem?

Bye and thank you,
bearophile

On Sat, 16 Feb 2008 19:43:54 -0500, bearophile wrote:

> While testing I have found a problem in my code, I have reduced the code to the following lines:
> 
> template IsArray(T) {
>     const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
>                          is(typeof(T.reverse)) && is(typeof(T.dup));
> }
> 
> TA[] foo(TA, TB)(TA[] a, TB b) {
>     TA[] result;
> 
>     static if (IsArray!(TB)) {
>         if (b.length == 0) {
>             result = a;
>         } else if (b.length == 1) {
>             result = foo(a, b[0]);
>         }
>     }
> 
>     return result;
> }
> 
> void main() {
>     foo("test", "");
> }
> 
> 
> Its output:
> 
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0] bug.d(13):
> Error: array index 0 is out of bounds b[0 .. 0] bug.d(13): Error: array
> index 0 is out of bounds b[0 .. 0] bug.d(21): template instance
> bug.foo!(char,char[0u]) error instantiating
> 
> I have found ways to solve this problem, but can someone please explain me what's the problem?
> 
> Bye and thank you,
> bearophile

I tried looking at this.  It's weird (and tripping me out).
I would love to see the explanation of what's going wrong here.

"bearophile" <bearophileHUGS@lycos.com> wrote in message news:fp800a$2781$1@digitalmars.com...
> While testing I have found a problem in my code, I have reduced the code to the following lines:
>
> template IsArray(T) {
>    const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
>                         is(typeof(T.reverse)) && is(typeof(T.dup));
> }
>
> TA[] foo(TA, TB)(TA[] a, TB b) {
>    TA[] result;
>
>    static if (IsArray!(TB)) {
>        if (b.length == 0) {
>            result = a;
>        } else if (b.length == 1) {
>            result = foo(a, b[0]);
>        }
>    }
>
>    return result;
> }
>
> void main() {
>    foo("test", "");
> }
>
>
> Its output:
>
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(21): template instance bug.foo!(char,char[0u]) error instantiating
>
> I have found ways to solve this problem, but can someone please explain me what's the problem?
>
> Bye and thank you,
> bearophile

As far as I can tell the cause is that the expression b[0] is being evaluated even if the enclosing if statement hasn't been entered. For example if the line containing b[0] is replaced with a writef() it is never output to the console.

I can only assume that this is a compiler bug.

On Sun, 17 Feb 2008 11:37:53 +0900, Neil Vice wrote:

> "bearophile" <bearophileHUGS@lycos.com> wrote in message news:fp800a$2781$1@digitalmars.com...
>> While testing I have found a problem in my code, I have reduced the code to the following lines:
>>
>> template IsArray(T) {
>>    const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
>>                         is(typeof(T.reverse)) && is(typeof(T.dup));
>> }
>>
>> TA[] foo(TA, TB)(TA[] a, TB b) {
>>    TA[] result;
>>
>>    static if (IsArray!(TB)) {
>>        if (b.length == 0) {
>>            result = a;
>>        } else if (b.length == 1) {
>>            result = foo(a, b[0]);
>>        }
>>    }
>>
>>    return result;
>> }
>>
>> void main() {
>>    foo("test", "");
>> }
>>
>>
>> Its output:
>>
>> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0] bug.d(13):
>> Error: array index 0 is out of bounds b[0 .. 0] bug.d(13): Error: array
>> index 0 is out of bounds b[0 .. 0] bug.d(21): template instance
>> bug.foo!(char,char[0u]) error instantiating
>>
>> I have found ways to solve this problem, but can someone please explain me what's the problem?
>>
>> Bye and thank you,
>> bearophile
> 
> As far as I can tell the cause is that the expression b[0] is being evaluated even if the enclosing if statement hasn't been entered. For example if the line containing b[0] is replaced with a writef() it is never output to the console.
> 
> I can only assume that this is a compiler bug.

Behaviour exists in DMD 2.010 and GDC 0.25(1.021)

bearophile wrote:
> While testing I have found a problem in my code, I have reduced the code to the following lines:
> 
> template IsArray(T) {
>     const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
>                          is(typeof(T.reverse)) && is(typeof(T.dup));
> }

...
Y'know, there's a template for this in std.traits and tango.core.Traits.

> TA[] foo(TA, TB)(TA[] a, TB b) {
>     TA[] result;
> 
>     static if (IsArray!(TB)) {
>         if (b.length == 0) {
>             result = a;
>         } else if (b.length == 1) {
>             result = foo(a, b[0]);
>         }
>     }
> 
>     return result;
> }
> 
> void main() {
>     foo("test", "");
> }
> 
> 
> Its output:
> 
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(13): Error: array index 0 is out of bounds b[0 .. 0]
> bug.d(21): template instance bug.foo!(char,char[0u]) error instantiating

Ouch! Constant expansion for the lose!

What's going on is, this is being evaluated at compile time, when you've got a runtime check for length. So it's expanded like this:

if ("".length == 0) {
   // blah
} else if ("".length == 1) {
   result = foo(blah, ""[0]);
}

Then the compiler sees: Array literal! Constant index! Must expand!
Obviously, it can't do that. So it gives an error.

So, you can use a static if, but that'll fail for a dynamic array. Or you can pass in ""[] rather than "". Or you can check if it's a static array, do a 'static if (array.length)' in that case, and in the case of a dynamic array, use a non-static if.

But yeah, it's a bug.

> I have found ways to solve this problem, but can someone please explain me what's the problem?
> 
> Bye and thank you,
> bearophile

bearophile <bearophileHUGS@lycos.com> wrote:
> TA[] foo(TA, TB)(TA[] a, TB b) {
>     TA[] result;
> 
>     static if (IsArray!(TB)) {
>         if (b.length == 0) {
>             result = a;
>         } else if (b.length == 1) {
>             result = foo(a, b[0]);
>         }
>     }
> 
>     return result;
> }

It seems to replace the line
result = foo(a, b[0]);
with
result = foo("test", ""[0]);
and then complain that the constant expression ""[0] is wrong.  It's
about distinguishing between static and dynamic arrays which I don't
know how to do.

In your case, adding static if(b.sizeof) should do the trick: only empty static arrays have zero size.

-- 
SnakE

Christopher Wright <dhasenan@gmail.com> wrote:
> if ("".length == 0) {
>     // blah
> } else if ("".length == 1) {
>     result = foo(blah, ""[0]);
> }
> 
> Then the compiler sees: Array literal! Constant index! Must expand! Obviously, it can't do that. So it gives an error.
> 
> [...]
> But yeah, it's a bug.

But how the compiler should handle this ?  In trivial case above, the code with error is unreachable.  But as a generic solution, it probably should expand into something like:

if (condition) {
    // blah
} else if (other condition) {
    blah;  // stays here for possible side effects
    // ""[0] is replaced with this at compile time
    throw new ArrayBoundsException();
    // the rest of the block is unreachable, throw away
}

-- 
SnakE

Thanks to everybody, in my code I did solve the problem like this:
result = foo(a, b.dup[0]);

But I was curious to know what you think about this situation. Now I know it's probably a compiler bug.


Christopher Wright:
> Y'know, there's a template for this in std.traits and tango.core.Traits.

I don't use Tango (yet). I use always the std lib when possible, I know there's no point in duplicating the std lib. I don't remember the bugs or the problems, but I have found std.traits not enough for similar array purposes, and I have found a bug in the traits of Tango to do that array purposes, so I have created IsArray and IsDynamicArray, you can find them in my d libs:

template IsArray(T) {
    const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
                         is(typeof(T.reverse)) && is(typeof(T.dup));
}
template IsDynamicArray(T) {
    const bool IsDynamicArray = is( typeof(T.init[0])[] == T );
}
template IsStaticArray(T) {
    const bool IsStaticArray = IsArray!(T) && (!IsDynamicArray!(T));
}

That IsArray may look a bit like "duck typing at compile time", but that's the only thing that works that I have found. I use similar tools all the time, so I need to have them quite sharp :-)

Bye,
bearophile

bearophile wrote:
> Thanks to everybody, in my code I did solve the problem like this:
> result = foo(a, b.dup[0]);
> 
> But I was curious to know what you think about this situation. Now I know it's probably a compiler bug.
> 
> 
> Christopher Wright:
>> Y'know, there's a template for this in std.traits and tango.core.Traits.
> 
> I don't use Tango (yet). I use always the std lib when possible, I know there's no point in duplicating the std lib. I don't remember the bugs or the problems, but I have found std.traits not enough for similar array purposes, and I have found a bug in the traits of Tango to do that array purposes, so I have created IsArray and IsDynamicArray, you can find them in my d libs:

std.traits.isArray/isStaticArray/isDynamic doesn't work? I don't recall seeing such a bug report.

> template IsArray(T) {
>     const bool IsArray = is(typeof(T.length)) && is(typeof(T.sort)) &&
>                          is(typeof(T.reverse)) && is(typeof(T.dup));
> }
> template IsDynamicArray(T) {
>     const bool IsDynamicArray = is( typeof(T.init[0])[] == T );
> }
> template IsStaticArray(T) {
>     const bool IsStaticArray = IsArray!(T) && (!IsDynamicArray!(T));
> }

const bool IsStaticArray(T) {
   const bool IsStaticArray = is(typeof(T[0])) && is(typeof(T[0])[T.sizeof/T[0].sizeof] == T);
}

Then, since IsStaticArray doesn't depend on IsArray, you can change IsArray to return true iff IsStaticArray or IsDynamicArray.

> That IsArray may look a bit like "duck typing at compile time", but that's the only thing that works that I have found.
> I use similar tools all the time, so I need to have them quite sharp :-)
> 
> Bye,
> bearophile

Christopher Wright:
> Then, since IsStaticArray doesn't depend on IsArray, you can change IsArray to return true iff IsStaticArray or IsDynamicArray.

You are right, thank you. I'm learning D still.

Bye,
bearophile