Nikolay Sivov schrieb:
On 12/13/2009 15:15, Roderick Colenbrander wrote:
The main test which AJ suggested would be to 'force' native user32 to call RegisterClassNameW. There would be a dummy dll containing a RegisterClassNameW to which lets say the Button control would be redirected using a manifest.
If I got it right you're talking about a dummy dll with compiled in (or separate doesn't matter) manifest with 'windowclass' entry (and without to check it's actually used) to Button, after that we trigger test application reload and dump this RegisterClassNameW call someway?
Yeah that's the idea. I'm quite certain that this mechanism is used to register the class. I just searched some more this morning and found stuff I didn't fnd before about subclassing. Read this http://msdn.microsoft.com/en-us/library/bb773183%28VS.85%29.aspx It talks about four new subclassing related functions which were introduced in XP, so it might mean that the controls are subclassed after all?
Roderick
Looks like it's definitely the way it works. Finally I'm able to reproduce that. That's what I've done:
- a DLL with a call:
BOOL RegisterClassNameW(WCHAR *name) { wprintf(L"%s\n", name); return FALSE; }
- an executable with main like that:
int _tmain(int argc, _TCHAR* argv[]) { HWND button;
button = CreateWindowW(L"Button", NULL, WS_VISIBLE, 0, 0, 10, 10, NULL, NULL, NULL, NULL); printf("%s, %p\n", "blah", button); DestroyWindow(button); return 0;}
- A generated a hash for DLL with mt.exe and updated a DLL manifest
with it 4) After that embedded a dll manifest to dll module 5) Looks like publicKeyToken attribute is necessary for this to work - I defaulted it to publicKeyToken="1111111111111111".
The final output for .exe is:
Button blah, 00000000
So as an additional "feature" - user32 doesn't attempt for register default classes if a custom RegisterClassNameW() fails.
Now I'm not sure what's next step. It seems that another test is about more than 1 dll with exported RegisterClassNameW loaded with same application (what dll will be called to register class isn't obvious at all).
of course that would only be a problem if they both export the same class with their manifest, maybe that will lead to a loader error.
Another question here is current Wine's actctx support ready for such tricks?
P.S. If somebody wants a bunch of dull sources to build this testcase let me know.
can you zip me a bundle of your sources and binaries please? You made a great step with this. congratulations!