Re: DDraw: Split up the ddraw refcount

2006-07-10 Thread Stefan Dösinger 
Am Montag 10 Juli 2006 10:47 schrieb Stefan Dösinger:
 Here is a patch with the improvement suggestions from AJ. The changes to
 the older versions are:

 * Use the counter from InterlockedDecrement to check if the object has to
 be destroyed
 * Get rid of IDirectDrawImpl::DoNotDestroy
 * Minor spelling fixes in the added comments :-)

 The aim of this is to fix the startup Star Wars Episode 1 racer and most
 likely Star Wars Jedi Knight 1(Not tested, I do not have this game), and
 fix the shutdown of CC Red Alert. The 2 star wars games depend on the
 individual refcounts of the interface versions, while Red Alert depends on
 the different palette refcounting behavior in older interface versions.

 This patch also requires the palette refcounting fix for older interfaces
 which I will resend too.

 ChangeLog:
 Stefan Dösinger: Split up the ddraw refcount
Grrr. Got to send another updated patch :-(

I accidentally renamed Thunk_IDirect3D_3_Release to Thunk_IDirect3D_1_Release, 
so the older patch didn't even compile. I wonder how I could run the 
regression tests and my games with it.

The attached patch corrects this wrong numer.


0005-DDraw-Split-the-ddraw-refcount.txt.bz2
Description: BZip2 compressed data


pgp5NjjQKmKSW.pgp
Description: PGP signature

Re: DDraw: Split up the ddraw refcount

2006-07-07 Thread Alexandre Julliard 
Stefan Dösinger [EMAIL PROTECTED] writes:

 +if(ref == 0) InterlockedDecrement(This-numIfaces);
  
 -/* This is for the dll cleanup code in DllMain() */
 -if(!This-DoNotDestroy)
 -IDirectDrawImpl_Destroy(This);
 -}
 +/* Checks the refcount before actually destroying the object */
 +IDirectDrawImpl_Destroy(This);

This isn't thread safe, you should only call Destroy when the
InterlockedDecrement returns 0, otherwise there's no point in using
interlocked functions at all. Also you should get rid of the
DoNotDestroy flag and use the refcount for that.

-- 
Alexandre Julliard
[EMAIL PROTECTED]

Re: DDraw: Split up the ddraw refcount

2006-07-06 Thread Stefan Dösinger 
Am Mittwoch 05 Juli 2006 18:27 schrieb Stefan Dösinger:
 Am Mittwoch 05 Juli 2006 02:27 schrieb Stefan Dösinger:
  The DirectDraw reference counting is a bit more complex than it seemed at
  first. The unit test in that patch shows that the IDirectDraw,
  IDirectDraw2, IDirectDraw4 and IDirectDraw7 interfaces of one object have
  their own reference count. The object is the same, the test with
  SetCooperativeLevel shows this(Setting a focus window returns an error in
  exclusive mode). Interestingly the IDirect3DX interfaces do not have
  their own refcount, instead they are linked to a DirectDraw interface
  refcount.
 
  This patch splits up the IDirectDrawImpl::ref member in ref7, ref4, ref2
  and ref1 and changes the AddRef and Release methods accordingly. Of
  course a number of tests to show this behavior are included :-)
 
  I will also send a patch which fixes some reference counting issues with
  Palettes, and both patches depend on each other. So please apply either
  both patches or none of them if one is wrong.
 
  ChangeLog:
  Stefan Dösinger: Split the ddraw refcount

 Here is an updated patch, it implements the destroy condition as suggested
 by Rob and adds the accidentally dropped SetCooperativeLevel(NULL,
 DDSCL_NORMAL) cooplevel and screen mode restoring call.
Oh wait, this is not necessarilly correct. Considered that Direct3D came with 
DirectX 3.0 (I think) IDirect3D(1) isn't necessarily linked to 
IDirectDraw(1).

I will add some more tests to check for that, do not apply this patch for now


pgpiFO78bK8Lm.pgp
Description: PGP signature