Created attachment 550
an example of the bug
Delegates of an interface that refer to methods in sub interfaces that are not
the first one fail.
Somehow when done through ThreadLocal variables the bug is slightly different
(nothing seems to be called), whereas the other examples that I show another
method get called.
I set it as major because it is a subtle bug that took me very long to track
down, its effects a similar to memory pollution...
This is a bug in both 1.047 and 1.055
Comment #1 by fawzi — 2010-01-14T12:58:44Z
Created attachment 551
a shorter example
Comment #2 by clugdbug — 2010-07-22T12:20:12Z
Here's a slightly reduced test case.
--------------
interface I{
void h();
}
interface K{
void f();
}
interface J:I,K {}
class A:J{
void f(){ }
void h(){ }
}
void main(){
auto a = new A();
J b = a;
K c = a;
assert(&b.f == &c.f); // fails: &b.f returns &a.h.
}
Comment #3 by clugdbug — 2010-07-22T13:55:22Z
The problem lies in delegate expressions formed from interfaces. We need to ensure that the correct interface is used, otherwise the vtblIndex will refer to the wrong one.
Applies to both D1 and D2.
PATCH: expression.c, not yet fully tested in the test suite. Maybe this patch should be made in getRightThis() instead?
Expression *DelegateExp::semantic(Scope *sc)
{
#if LOGSEMANTIC
printf("DelegateExp::semantic('%s')\n", toChars());
#endif
if (!type)
{
e1 = e1->semantic(sc);
type = new TypeDelegate(func->type);
type = type->semantic(loc, sc);
AggregateDeclaration *ad = func->toParent()->isAggregateDeclaration();
if (func->needThis())
e1 = getRightThis(loc, sc, ad, e1, func);
+ if (ad && ad->type != e1->type)
+ { // A downcast is required for interfaces
+ e1 = new CastExp(loc, e1, ad->type);
+ e1 = e1->semantic(sc);
+ }
}
return this;
}