Refcounting in Direct3DDevice8::Release (again) about d3d8to9

[elishacloud]

Yes, I just verified #18 and it seemed to work fine. I did not find any issues with it.

[crosire]

auto &rt = _current_rendertarget;
auto &ds = _current_depthstencil;
if (rt != nullptr && ds != nullptr && myRef <= 3)
{
    rt->Release(); rt = nullptr;
    ds->Release(); ds = nullptr;
}
else if (rt != nullptr && myRef <= 2)
{
    rt->Release(); rt = nullptr;
}
else if (ds != nullptr && myRef <= 2)
{
    ds->Release(); ds = nullptr;
}

[elishacloud]

I don't think those changes are safe. I tried that with one of my games (Rayman 3: Hoodlum Havoc) and it crashed on exit, whereas it never crashed before. Event Viewer showed it as crashing in d3d8.dll (this wrapper).

[crosire]

Mmh. Well, then it's probably better to follow the "Never change a running system" matra.

[elishacloud]

Ok, if I change the declaration then it works without crashing:

auto* rt = _current_rendertarget;
auto* ds = _current_depthstencil;
if (rt != nullptr && ds != nullptr && myRef <= 3)
{
	rt->Release(); rt = nullptr;
	ds->Release(); ds = nullptr;
}
else if (rt != nullptr && myRef <= 2)
{
	rt->Release(); rt = nullptr;
}
else if (ds != nullptr && myRef <= 2)
{
	ds->Release(); ds = nullptr;
}

[crosire]

That doesn't reset the fields to nullptr though, which is bad too.

[CookiePLMonster]

Indeed, auto& shouldn't be used in here. But other than this, it looks good and it does reset the fields to nullptr (missed the assignments after Release(), crosire?).

The only thing I'd change is <= to == - defensive approach is really not needed in here, I guess.

[elishacloud]

Yes, that code seems to work for me. No crashes or other issues.

[CookiePLMonster]

The one I posted in here was wrong again though (too sleepy, it seems). I'll fix and issue a PR.

[crosire]

No, it doesn't reset the fields to nullptr. It only sets the local auto variables to nullptr. But I see you fixed that in the PR.

[crosire]

I don't have any way to test #18 currently, so would greatly appreciate if @elishacloud could verify things still work with it.