Re: Template Parameter Deduction

[Paul D Anderson via Digitalmars-d-learn]

On Wednesday, 11 March 2015 at 23:04:15 UTC, Ali Çehreli wrote:
On 03/11/2015 03:44 PM, Paul D Anderson wrote:
This used to work in D2.065:
given

1) public T mul(T)(in T x, in T y,
    Context context = T.context) if (isDecimal!T)
    // one template parameter for the two input values

and

2) public T mul(T, U)(in T x, U n, Context context = T.context)
    if (isDecimal!T && isIntegral!U)
    // two different template parameters for the two input 
values

then

3) dec9 arg1 = dec9("1.20");
   long arg2 = 3;
   result = mul(arg1, arg2);
   // correctly deduced function

But now (D2.066.1) either 1) has to be changed to

1) public T mul(T, U)(in T x, U y, Context context = T.context)
    if (isDecimal!T && isDecimal!U)
    // two identical template parameters for the two input 
values

or 3) has to be changed to

3) dec9 arg1 = dec9("1.20");
   long arg2 = 3;
   result = mul!(dec9,long)(arg1, arg2);
   // template parameters have to be made explicit

Is this expecded behavior?

Paul

Hint: It makes it much simpler to work with complete code. The 
following code took a while for me to put together:

template isDecimal(T)
{
    enum isDecimal = true;
}

template isIntegral(T)
{
    enum isIntegral = true;
}

public T mul(T)(in T x, in T y,
    Context context = T.context) if (isDecimal!T)
    // one template parameter for the two input values
{
    return x;
}

alias Context = int;

public T mul(T, U)(in T x, U n, Context context = T.context)
    if (isDecimal!T && isIntegral!U)
    // two different template parameters for the two input 
values
{
    return x;
}

struct dec9
{
    string s;
    enum context = 42;
}

void main()
{
    dec9 arg1 = dec9("1.20");
    long arg2 = 3;
    dec9 result = mul(arg1, arg2);
   // correctly deduced function
}

Yes, it fails with 2.066.1 but compiles fine with git head.

Ali

Thanks to you and John for taking the time to work this out. I 
didn't mean for anyone to rebuild the example -- I was just 
hoping there was a quick answer based on known language changes. 
I should have included complete example code.

At any rate, it looks like there is a hiccup in the template 
parameter deduction code and that it is being fixed. And it's 
easy to work around. I'll dig a little deeper to see if I can 
find a related bug report.

Paul