[Issue 15346] New: Calling interface methods on out contracts causes segfaults

via Digitalmars-d-bugs digitalmars-d-bugs at puremagic.com
Mon Nov 16 10:18:18 PST 2015


https://issues.dlang.org/show_bug.cgi?id=15346

          Issue ID: 15346
           Summary: Calling interface methods on out contracts causes
                    segfaults
           Product: D
           Version: D2
          Hardware: x86_64
                OS: Linux
            Status: NEW
          Severity: major
          Priority: P1
         Component: dmd
          Assignee: nobody at puremagic.com
          Reporter: initrd.gz at gmail.com

One of my projects started mysteriously throwing segfaults. I managed to narrow
it down to a line in one of my interface's out contracts that compared the
result with a value returned by a property method defined on the interface.
Even doing `debug stderr.writeln(this);` threw a segfault; GDB says its from
the `_d_interface_cast` function.

The full code is too big to post here, but I think I narrowed it down:


    import std.stdio;
    interface A {
        int* aField() @property pure;

        int foo()
        out(v) {
            debug stderr.writeln(this.aField);
        }
    }

    class F : A{
        this() {}

        int myField;

        int* aField() @property pure { return &myField; }
        int foo() {
            debug stderr.writeln(this.aField);
            return 0;
        }
    }

    void main() {
        auto f = new F();
        f.foo();
    }

I expect that this function would print the same address twice; it prints out
`aField`, a simple function that returns a pointer to a class member, once in a
function body and once in the interface's out contract. Running it with `rdmd
-debug test.d` results in the following output:

    7F21F379F010
    6

The first address differs with each run (as expected with memory addresses) but
the second line is consistently a 6. Moving the out contract to the
implementation of foo method in class F results in the correct output (the same
address printed twice).

I suspect that the pointer adjustment that happens with `this` in an out
contract is being done incorrectly, but I don't have the skills to diagnose it
further.

Using dmd v2.069.1 on x64 XUbuntu

--


More information about the Digitalmars-d-bugs mailing list