14.06.2013 18:31, David пишет:
This code currently fails with a RangeError (used to work in 2.062)
// http://dpaste.dzfl.pl/332a71ec
import std.stdio;
import std.signals;
struct X {
mixin Signal!();
}
class O {
void m() {}
}
void main() {
O o = new O();
X[string] aa;
aa["o"] = X.init;
aa["o"].connect(&o.m);
/*{ // 20
X x = aa["o"];
}*/
}
If you take the uncomment the "block" at line 20 you end up with a
segmentation fault, also worked in 2.062. Both times the problem is in
the __dtor (segfault happens in the call to "_d_toObject(stor.ptr)".
Changing "struct" to "class" and "X.init" to "new X()" it seems to work
as it should.
Is this worth a bugreport or was the old behaviour never intended?
This bug (I consider it one) broke quite a few lines of code... If the
old behaviour was never intended, why wasn't it documentated then... oh
well I am drifting into another D rant here...
http://dlang.org/phobos/std_signals.html#Signal
"Mixin to create a signal within a class object."
So your `X` must be a class.
Also don't use std.signals - it's an incorrect and dangerous mess (see
bug tracker).
--
Денис В. Шеломовский
Denis V. Shelomovskij