C++ codegen regression: memset called on a result variable of importcpp type

[nigredo-tori]

Reduced example:

{.emit:"""
#import <cassert>

template <typename X> class C {
  public:
    int d;

    C(): d(1) { }

    C<X>& operator=(const C<X> other) {
      assert(d == 1);
    }
};
""".}

type C{.importcpp, header: "<stdio.h>", nodecl.} [X] = object
proc mkC[X]: C[X] {.importcpp: "C<'*0>()", constructor, nodecl.}

proc foo(): C[int] =
  result = mkC[int]()

discard foo()

The assertion fails, because the generated code for foo looks like this:

C<NI>  result0;
nimfr("foo", "test_default_constructor.nim")
memset((void*)(&result0), 0, sizeof(result0));
nimln(20, "test_default_constructor.nim");
result0 = C<NI>();
popFrame();
return result0;

This memset breaks use cases relying on default constructor (in my case this was QList).

Tested on: 80f9dcc
Bisect points to: 860cbd3

[philip-wernersbach]

This should be reopened, this regressed in 0.17.0.

[Yardanico]

@philip-wernersbach I don't have any error messages after running this example on Arch Linux with latest gcc and latest nim devel

[Yardanico]

[root@tiberium ~]# nim cpp test.nim
Hint: used config file '/root/.choosenim/toolchains/nim-#devel/config/nim.cfg' [Conf]
Hint: system [Processing]
Hint: test [Processing]
CC: test
CC: stdlib_system
Hint:  [Link]
Hint: operation successful (10912 lines compiled; 5.415 sec total; 17.938MiB peakmem; Debug Build) [SuccessX]
[root@tiberium ~]# ./test
[root@tiberium ~]#

[Yardanico]

Can this be closed?

[Araq]

The example program continues to compile. Please open a new bug with a new snippet that fails.