Re: [Vala] Incorrect C generated with union containing anonymous struct

[Evan Nemerson <evan coeus-group com>]

On Thu, 2011-10-13 at 00:10 +0200, Jan-Jaap van der Geer wrote:
> (Excuses if this appears twice, but after waiting half an hour my
> first attempt still hasn't appeared...)
>
> I have the following struct:
>
> typedef union foobar foobar;
>
> union foobar
>    {  struct
>       {  char *bar;
>       }
>       foo;
>    };
>
> This is represented in the vapi like this:
>
> [CCode (cheader_filename = "struct.h")]
> namespace StructTest
> {
>   [CCode (cname = "foo")]
>   public struct Foo
>   {
>     [CCode (cname = "bar")]
>     public unowned string bar;
>   }
>
>   [CCode (cname = "foobar")]
>   public struct Foobar
>   {
>     [CCode (cname = "foo")]
>     public unowned Foo foo;
>   }
> }

There are basically two ways to do this... what you usually see in vapis
is something like this:

public struct Foobar {
  [CCode (cname = "foo.bar")]
  public unowned string foo_bar;
}

The other option, which I haven't tested in a while but used to work
well enough, is something like:

[CCode (cname = "struct { char *bar; }")]
public struct Foo {
  public unowned string bar;
}

public struct Foobar {
  public Foo foo;
}

> I use it like this:
>
>     var fbar = Foobar();
>     Memory.copy (fbar.foo.bar, "tralala", 10);
>
> In Vala 0.13.1 this worked fine and the following code was
> generated:
>
>     foobar fbar = {0};
>     memset (&fbar, 0, sizeof (foobar));
>     memcpy (fbar.foo.bar, "tralala", (gsize) 10);
>
> However, in version 0.13.2 and beyond the following code is
> generated:
>
>     foobar fbar = {0};
>     foobar _tmp0_;
>     foo _tmp1_;
>     const gchar* _tmp2_;
>     memset (&fbar, 0, sizeof (foobar));
>     _tmp0_ = fbar;
>     _tmp1_ = _tmp0_.foo;
>     _tmp2_ = _tmp1_.bar;
>     memcpy (_tmp2_, "tralala", (gsize) 10);
>
> This fails since foo, the variable of _tmp1_ is undefined in C. I
> could of course remove the [CCode (cname = "foo")] in the vapi but
> that only changes the name to something else that is not known.
>
> Is this a bug? Is there a workaround? Note that the C-code is
> beyond my control, so workarounds involving changing the definition
> of the struct in C will most probably not work...

No, it's not a bug. The fact that it used to work was a bit of a
fluke... there are situations where it never did (such as assigning
Foobar.foo to a variable).


-Evan