I'm trying to initialize an AA, but initializing at definition fails, as do in a shared module ctor.

// how is this now constant?
//shared string[int] aa = [1: "a"]; // Error: non-constant expression [1:"a"]

// ok
string[int] aa2;
static this()
{
    aa2 = [1: "a"];
}

// The same with shared fails
// segfault
/+
shared string[int] aa3;
shared static this()
{
    aa3 = [1: "a"];
}
+/

// Using __gshared works though
__gshared string[int] aa4;
shared static this()
{
    aa4 = [1: "a"];
}

void main() {
    assert(aa4[1] == "a");
}

On Sun, Mar 18, 2012 at 02:50:00PM +0100, simendsjo wrote:
> I'm trying to initialize an AA, but initializing at definition fails, as do in a shared module ctor.
> 
> // how is this now constant?
> //shared string[int] aa = [1: "a"]; // Error: non-constant
> expression [1:"a"]

This is a shortcoming in the current AA implementation: all AA's have to be initialized at runtime, even AA literals.

In my prospective AA replacement, I've worked out a way to create compile-time AA literals via CTFE and mixins, though I haven't actually implemented it yet. Preliminary testing code seems to show that this should be quite possible.


> // ok
> string[int] aa2;
> static this()
> {
>     aa2 = [1: "a"];
> }

Yes, this is the workaround currently necessary.


> // The same with shared fails
> // segfault
> /+
> shared string[int] aa3;
> shared static this()
> {
>     aa3 = [1: "a"];
> }
> +/
[...]

Now this I've no idea about. But I know there are some bugs with shared currently; you might want to search the bugtracker for issues that might be causing this.


T

-- 
He who sacrifices functionality for ease of use, loses both and deserves neither. -- Slashdotter