- Move sending WM_MEASUREITEM message for HBMMENU_CALLBACK bitmaps to
MENU_GetBitmapItemSize();
- Save the received bitmap size in the menu item structure as it will
need to be sent in the WM_DRAWITEM message;
-  Move sending WM_DRAWITEM message for HBMMENU_CALLBACK bitmaps to
MENU_DrawBitmapItem();
- This also means that HBMMENU_CALLBACK bitmaps are now supported in
menu bars.

level of some large if statements in the drawing code. Some updates to
the comments.

the relative location of the wine libraries.

- remove unused argument from functions
- print the pointer in the str_array_push TRACE to be able to distinguish
  the different arrays.
- comment and white space cleanups

Add some tests for unDName() checking:
- data type modifiers A,B,P,Q,R,S
- literal strings
- back references to names

Use a separate array/stack for the back references to names as
sym->stack is used to push temporary strings to it which breaks back
referencing in some occasions (26 symbols in the mfc42 dll).  During
this cleanup get_class() a little bit.

unDName: Move the code to parse a literal string from get_class() to a
separate function. Add some error handling to the new function.

In test_demangle use a custom strcmp that treats multiple spaces as
single space. __unDName() from older msvcrt (including W2K it seems)
emit in some places 2 spaces instead of one. Fixes 20 failing tests on
those old Windows versions.

Change the apartment model field to a boolean that just specifies
whether the apartment is multi-threaded or not and fix up all the
places where this is used.
Fixes a bug where we would return an error if the previous apartment
model passed into CoInitializeEx matches the new apartment model, but
additional optimisation flags are specified.

criteria (it was mainly doing the opposite).

E_NOTIMPL in Windows as well.

Implement CertGetEnhancedKeyUsage, CertSetEnhancedKeyUsage,
CertAddEnhancedKeyUsageIdentifier, and
CertRemoveEnhancedKeyUsageIdentifier.

x86_64 in 32-bit mode.

Setting the EOF flag only if _read() returns zero is not enough as we
could also read something AND hit the end of file.
_read() already sets another internal EOF flag, so just use this too.
Removing todo_wine from the test case.

When calling fread() with a large enough buffer to reach the end of
the file a subsequent feof() should return true; (marked as todo_wine
because it doesn't work atm).

We can't just pop a exception record from the exception handler list
which we haven't put there.