January 14, 2018
vartmpl.d
```
import std.stdio : writeln;
import decimal : decimal32;

template F(T) {
   immutable T c = 3;
}

void foo (T) ()
{
   immutable T t = 1;
}

void main ()
{
//   immutable decimal32 i = 1; // Error: none of the overloads of '__ctor' are
callable using a immutable object
//   foo!decimal32; //  Error: none of the overloads of '__ctor' are callable us
ing a immutable object, candidates are:
   alias c = F!decimal32.c;
   c.writeln;
   writeln (typeof (c).stringof);
}
```

$ dmd -g vartmpl.d decimal.git/libdecimal.a
$ ./vartmpl

3
immutable(Decimal!32)

Why does this compile while both of the commented lines give a compile error. decimal ist http://rumbu13.github.io/decimal/
January 14, 2018
On Sunday, 14 January 2018 at 16:23:18 UTC, kdevel wrote:
> Why does this compile while both of the commented lines give a compile error.

The code boils down to this:

----
struct decimal32
{
    this(int x) {}
}

immutable decimal32 c = 3; /* works */

void main ()
{
   immutable decimal32 i = 1; /* error */
}
----

I think this is CTFE being unexpectedly smart.

If you add the `pure` attribute to the constructor, then the `i` line works as well. That's because a strongly pure constructor is guaranteed to return a unique object, and a unique object can be converted implicitly to other mutability levels.

The `pure` attribute is needed for `i`, because here the compiler only looks at the function attributes to determine purity. No `pure` attribute -> function is regarded as impure.

But for `c`, the constructor goes through CTFE, and CTFE doesn't care all that much about the `pure` attribute. Instead, CTFE just tries to evaluate the function and aborts when it encounters an action that would be impure.