http://d.puremagic.com/issues/show_bug.cgi?id=10785

           Summary: Interface diamond covariance causes silent breakage
           Product: D
           Version: D1 & D2
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: major
          Priority: P2
         Component: DMD
        AssignedTo: nobody@puremagic.com
        ReportedBy: default_357-line@yahoo.de


--- Comment #0 from FeepingCreature <default_357-line@yahoo.de> 2013-08-08 17:01:53 PDT ---
Consider this code:

interface A { A foo(); }

interface A1 : A { A1 foo(); }
interface A2 : A { A2 foo(); }

class C : A1, A2 {
  override C foo() { return new C; }
}

void main() {
  A2 x = new C;
  A2 y = x.foo();
  // y.classinfo.name should be "A2" here; instead, it is "A1". The vtables
also don't match up (not that there's significant vtables in this example, but,
you know)
  assert(y.classinfo.name.find("A2") != -1); // fails
  // Further issues: this should be a no-op, but it isn't.
  A2 z = cast(A2) cast(A) y;
  assert(z.classinfo.name == y.classinfo.name); // fails too
}

The problem seems to be that whatever magic the compiler does to make interface covariance work fails in the case where it has to satisfy the same interface via two covariant child interfaces at once. I suspect this would either require A2.foo to be filled with a stub that does the conversion of C to A2, or a compiletime error.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------

http://d.puremagic.com/issues/show_bug.cgi?id=10785


Infiltrator <lt.infiltrator@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |lt.infiltrator@gmail.com


--- Comment #1 from Infiltrator <lt.infiltrator@gmail.com> 2013-08-11 22:22:22 PDT ---
This also affects non-diamond configurations:

http://dpaste.dzfl.pl/e9df03f9

----------------------
import std.stdio;

interface A1 { A1 foo(); }
interface A2 { A2 foo(); }

class C1 : A1, A2 {
  override C1 foo() { return new C1; }
}

interface B1 { B1 foo(); }
interface B2 { B2 foo(); }

class C2 : B2, B1 {
  override C2 foo() { return new C2; }
}

void main() {
  A2 x = new C1;
  A2 y = x.foo();
  writefln("X: %s", x.classinfo);
  writefln("Y: %s", y.classinfo);


  B2 a = new C2;
  B2 b = a.foo();
  writefln("A: %s", a.classinfo);
  writefln("B: %s", b.classinfo);
}
-----------------------
Application output:

X: f230.A2
Y: f230.A1
A: f230.B2
B: f230.B2

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------

http://d.puremagic.com/issues/show_bug.cgi?id=10785



--- Comment #2 from FeepingCreature <default_357-line@yahoo.de> 2013-08-11 22:40:21 PDT ---
(In reply to comment #1)
> This also affects non-diamond configurations:

Thanks, filed as 10806. Not sure whether to mark this as a duplicate. (10806 is
the more general bug)

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------

http://d.puremagic.com/issues/show_bug.cgi?id=10785



--- Comment #3 from Infiltrator <lt.infiltrator@gmail.com> 2013-08-11 23:02:31 PDT ---
I've set #10806 as a blocker for this, as diamond may still break even after non-diamond is fixed.  Somebody more knowledgeable in dmd can make the duplicate decision.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------