Fixes Bugzilla #429.


Selected notes from the patch's README:

= FIXES =

This patch fixes the above issues as follows.

== X11 events ==

Moved XFilterEvent just after XNextEvent so that all events are passed
to it.  Also, XFilterEvent will receive masks indicated by IM through
XNFilterEvents IC value as well as masks surpplied by SDL.

X11_KeyRepeat is called between XNextEvent and XFilterEvent, after
testing an event is a KeyRelease.  I'm not 100% comfortable to do so,
but I couldn't find a better timing to call it, and use of the
function is inevitable.

== Xutf8LookupString ==

Used a longer buffer to receive UTF-8 string.  If it is insufficient,
a dynamic storage of the requested size will be allocated.  The
initial size of the buffer is set to 32, because the Japanese text
converted from the most widely used benchmark key sequence for
Japanese IM, "WATASHINONAMAEHANAKANODESU." has ten Japanese characters
in it, that occupies 30 bytes when encoded in UTF-8.

== SDL_keysym.unicode ==

On Windows version of SDL implementation, SDL_keysym.unicode stores
UTF-16 encoded unicode characters, one UTF-16 encoding unit per an SDL
event.  A Unicode supplementary characters are sent to an application
as two events.  (One with a high surrogate and another with a low
surrogate.)  The behavior seems reasonable since it is upward
compatible with existing handling of BMP characters.

I wrote a UTF-8 to UTF-16 conversion function for the purpose.  It is
designed with the execution speed in mind, having a minimum set of
features that my patch requires.

--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%402378