Re: [PATCH 1/5] widl: Introduce an "aggregate_return" function attribute.

[Alexandre Julliard]

Henri Verbeet  writes:
> For MSVC style handling of aggregate return values. This probably doesn't make
> anyone particularly happy; I'm open to suggestions. It may be possible to do
> this without an attribute, based on the function declaration, but then we'd
> have to figure out how MSVC decides when to use registers and when to use the
> stack for the return value.

The attribute is certainly ugly. Also it seems to me you are doing this
the wrong way around; you don't want to add a return parameter, you want
to avoid one. Check how div() is handled.

[PATCH 1/5] widl: Introduce an "aggregate_return" function attribute.

[Henri Verbeet]

On 4 September 2014 15:27, Alexandre Julliard  wrote:
> The attribute is certainly ugly.

Well, no argument there. Do we have any alternatives? We could of
course also just change the function prototype, but then that would
break source compatibility.
> Also it seems to me you are doing this
> the wrong way around; you don't want to add a return parameter, you want
> to avoid one. Check how div() is handled.

No, the issue is that gcc returns the result in eax:edx, but native
expects a return parameter, so the reverse of MSVCRT_div(). It's most
obvious when running d2d1 tests (not submitted yet) compiled with
mingw against Windows.

[PATCH 1/5] widl: Introduce an "aggregate_return" function attribute.

[Alexandre Julliard]

Henri Verbeet  writes:
> No, the issue is that gcc returns the result in eax:edx, but native
> expects a return parameter, so the reverse of MSVCRT_div(). It's most
> obvious when running d2d1 tests (not submitted yet) compiled with
> mingw against Windows.

Mingw implements the Windows convention, so it's different from plain
gcc. I'm not sure your test means what you think it means.