RenderFile should add a source filter and call Render on each output
pin, while Render() should try to connect, first directly to any
existing renderers, and then using intermediates. It uses recursion
since this is the only possible way to implement Render sanely.

Does some security checks that it won't connect a filter to itself,
and also cleans up after a failed connection attempt.

These internal flags conflict with the Win32 API flags TPM_RECURSE,
TPM_RIGHTBUTTON, and TPM_CENTERALIGN respectively, which are all stored
in the same flag UINT.

As far as I can tell we support post ps blending in combination with
MRTs fine. Tabula Rasa needs this cap in order to enable some of the
higher graphics settings.

When the caret is at the start of a run, it uses the font of the
previous run for inserting characters.  The caret size previously was
the wrong height for the characters being inserted when the caret was at
the start of a line, but not the start of a paragraph so this patch
fixes this bug.

The check to see if the dwOutputBuffer is full was performed incorrectly
in RTFPutUnicodeString, however, this mistake was actually harmless
since it would just cause an extra loop that does nothing, then the
check would work properly with the variable fit equal to 0.

Checks were added for hexadecimal values that did not have valid
characters, and for EOF received before the final closing brace of the
rich text stream.  The error values were tested on richedit versions 1,
2, 3 & 4.1, and they were all the same for these cases.

The function ME_GetInsertStyle already checks for the case where there
is a selection, so there is no need to duplicate this code for
ME_GetSelectionInsertStyle.

Consecutively typed characters are grouped together to be undone
together.  The grouping of typed characters can be stopped by certain
events that are mentioned in MSDN's remarks on the EM_STOPGROUPTYPING
message, which is also implemented by this patch.

Previously bold needed to be set by setting CFM_WEIGHT in the
CHARFORMAT2 structure, and then setting the appropriate wWeight value.
This approach isn't even supported in version 3.0 of the richedit
control. Now bold can be set/unset properly for Windows or Wine using
CFE_BOLD in dwEffects and with CFM_BOLD set in the dwMask flag.

The LOGFONT's lfHeight member is in logical units, and is being used to
set the yHeight member of CHARFORMAT2 which is supposed to be in twips.

Opening a text file with a NULL terminated character in it was causing
an assertion error after a run was being split due to word wrap.
Windows allows NULL terminated characters to be in the text.

Using Ctrl-RightArrow to move to the start of the next word did not
previously work when at the start of a word.  This means that
Ctrl-RightArrow would not work twice in a row since it should move to
the start of the next word.

The call to GetClientRect returns 0 values for the returned RECT when
called in WM_NCCREATE in on Windows, which ended up causing an assertion
error when Wine's riched20.dll replaces the native version. Moving the
call to WM_CREATE fixes this problem (probably because NCCALCSIZE is
called in between).

There was a missing call to ShowCaret after the caret was created.

The function ME_FindRunInRow uses two parameters to return values by
reference, and treated these parameters as if they were optional except
for the start of the function which set *pbCaretAtEnd without checking
to see if was a NULL pointer.