That's what I figured. I thought I would try anyway.
On Tue, Mar 27, 2018, 12:45 AM Austin English austinenglish@gmail.com wrote:
On Mon, Mar 26, 2018 at 11:35 PM, Kieran Duggan kieranduggan15@gmail.com wrote:
So I took another look at the ideas list and I thought that writing micro benchmarks for the D3D components would be about my speed and also fit
into
my interests. My only issue is that I am not sure which operations to test.
While I was looking for inspiration I came across this project[1] and thought that it could be a good focus for a GSoC project. That is, specifically writing micro benchmarks to measure the
improvements
of components effected by changes in wine-pba. I'm very uncertain about
this
however because it isn't officially in the master branch or even submitted at all. As far as I can tell the developer isn't directly associated with Wine. On the other hand having conformance tests and benchmarks made would save the developer time and allow his patch to be moved through quicker. But really it just looks interesting so I thought I would bring it up.
If this isn't a possibility, then I could use some help finding
operations
that I can work on.
I know this is very close to the deadline, so I apologize for my poor timing. I underestimated how long it would take me to submit a patch and ended up investing not enough time on the proposal. I hope that I can make up for this in the final
hours.
Again, thank you for your assistance.
[1] https://comminos.com/posts/2018-02-21-wined3d-profiling.html
On Mon, Mar 26, 2018 at 10:02 AM, Stefan Dösinger stefandoesinger@gmail.com wrote:
Am 26.03.2018 um 06:12 schrieb Kieran Duggan <kieranduggan15@gmail.com
:
The hard part of this for me was figuring out who was responsible for freeing container. After some time I came to the conclusion that the
caller
of the AtlAxAttachControl function was intended to free container. I namely came to this conclusion because I couldn't really see how the
DestroyWindow
function would be able to free the container. If DestoyWindow is
supposed
to be responsible for freeing container, I just wasn't smart enough to figure it out and will have too look again.
As far as I can tell (and I haven't touched the atl code myself before) your conclusion is correct. AtlAxAttachControl creates the container
object
and returns the interface, so the caller is responsible for eventually destroying it. I would say submit your patch :-) . And don't forget to submit your gsoc proposal in time on the google website!
Ignore the following if it confuses you. It's some semi-educated
guesswork
on my part:
DestroyWindow doesn't know anything about COM or atl, so the
DestroyWindow
implementation is certainly not the right place. However, one thing is theoretically possible: DestroyWindow will send WM_DESTROY to the window callback procedure, which could in theory be responsible for releasing
the
container. AtlAxAttachControl appears to overwrite the wndproc (in IOCS_Attach). IOCS_Detach might be a candidate for releasing the
container,
it is called on WM_DESTROY.
However, I think this is unlikely because AtlAxAttachControl returns the interface to the caller. And convention says that the caller that
receives
an interface must release it once it is done. Of course Microsoft
screws up
its own COM rules a lot.
You can try to do to find out by looking at how Windows behaves when it
is
running this test: You can read the reference count by calling AddRef() followed by Release(). (there is a function get_refcount in numerous
places,
e.g. dlls/ddraw/tests). After AtlAxAttachControl I'd expect it to be 1.
If
it is still 1 after DestroyWindow(), the window callbacks should leave
the
refcount alone. If trying to call AddRef() after DestroyWindow crashes,
or
the refcount is zero, something in the window procedure released the container.
I don't think making a project about a wine fork (that's not upstreaming it) is a useful use of GSOC resources.
-- -Austin GPG: 267B CC1F 053F 0749 (expires 2021/02/18)