I'm just getting started with D, and was playing around with string mixins.  I've hit a snag, and have reduced it to a minimal test case:

    import std.conv;

    string test()
    {
        return to!string(0.0);
    }

    immutable auto testvar = mixin(test());

This gives this result when compiling:

/usr/include/phobos2/std/format.d(1479): Error: snprintf cannot be interpreted at compile time, because it has no available source code
/usr/include/phobos2/std/conv.d(99):        called from here: formatValue(w,src,f)
/usr/include/phobos2/std/conv.d(824):        called from here: toStr(value)
/usr/include/phobos2/std/conv.d(268):        called from here: toImpl(_param_0)
test.d(6):        called from here: to(0)
test.d(9):        called from here: test()
test.d(9): Error: argument to mixin must be a string, not (test())

I guess converting a double to a string can't be done at compile time because it requires calling the C snprintf function?  It compiles fine if I replace the 0.0 with an int literal.  Is there any way around this limitation?

On Tuesday, 21 August 2012 at 04:43:09 UTC, Bobby Bingham wrote:
> I'm just getting started with D, and was playing around with string
> mixins.  I've hit a snag, and have reduced it to a minimal test case:
>
>     import std.conv;
>
>     string test()
>     {
>         return to!string(0.0);
>     }
>
>     immutable auto testvar = mixin(test());
>
> This gives this result when compiling:
>
> /usr/include/phobos2/std/format.d(1479): Error: snprintf cannot be interpreted at compile time, because it has no available source code
> /usr/include/phobos2/std/conv.d(99):        called from here: formatValue(w,src,f)
> /usr/include/phobos2/std/conv.d(824):        called from here: toStr(value)
> /usr/include/phobos2/std/conv.d(268):        called from here: toImpl(_param_0)
> test.d(6):        called from here: to(0)
> test.d(9):        called from here: test()
> test.d(9): Error: argument to mixin must be a string, not (test())
>
> I guess converting a double to a string can't be done at compile time
> because it requires calling the C snprintf function?  It compiles fine
> if I replace the 0.0 with an int literal.  Is there any way around
> this limitation?

I have had to work around this also. One way is to first multiply your float by a large factor (say 10000 depending on what precision you want) and then adding a decimal point back in to the string. Kinda hacky, but... Here is an example:

int fx = cast(int) (cos(angle)*1000000.);
string fsx = fx.to!string;

string xprefix;
if (fsx[0] == '-')
{
  xprefix = "-";
  fsx = fsx[1..$];
}

if (fsx.length == 7)
  sx = xprefix ~ fsx[0] ~ "." ~ fsx[1..$];
else
  sx = xprefix ~ "0." ~ fsx;

sx is then the string you want to mixin.

On Tue, Aug 21, 2012 at 6:43 AM, Bobby Bingham <uhmmmm@gmail.com> wrote:
> I'm just getting started with D, and was playing around with string mixins.  I've hit a snag, and have reduced it to a minimal test case:
>
>     import std.conv;
>
>     string test()
>     {
>         return to!string(0.0);
>     }
>
>     immutable auto testvar = mixin(test());
>

> I guess converting a double to a string can't be done at compile time because it requires calling the C snprintf function?  It compiles fine if I replace the 0.0 with an int literal.  Is there any way around this limitation?

A possibility is to use a function template, passing the double as a template argument:

string test(double d)() // d is a template argument
{
    return d.stringof;
}

enum testvar = mixin(test!(3.14));

void main()
{
    pragma(msg, testvar);
}

On Thursday, 23 August 2012 at 13:56:05 UTC, Philippe Sigaud wrote:
> On Tue, Aug 21, 2012 at 6:43 AM, Bobby Bingham <uhmmmm@gmail.com> wrote:
>>[...]
>
>> [...]
>
> A possibility is to use a function template, passing the double as a template argument:
>
> string test(double d)() // d is a template argument
> {
>     return d.stringof;
> }
>
> enum testvar = mixin(test!(3.14));
>
> void main()
> {
>     pragma(msg, testvar);
> }

https://issues.dlang.org/show_bug.cgi?id=15497