Defining an own memmove() here is silly. It breaks compilation against
MacOSX 10.6 SDK where memmove is a macro. If we really wanted to avoid
the system memmove() here, surely we should then use
rtl_moveMemory(). But since when is idlcpp performance critical?

Factor out the part of the iOS patch that comments out ATSUI-using API
into a new patch file, and use that when building against the MacOSX
SDK 10.5.

Whether the callers of cairo will build then, no idea... Will see.

Happened on non-standard locales (like hu_HU on Linux) when osl_loadModuleRelative
wanted to access non-standard text encodings.

For Windows, the fix is still only a "TODO: FIXME" dummy.

In the spirit of 12759f67, change
external lib's config.sub to eat the arm-unknown-linux-androideabi
host os string. Also, permit shared libs again - seems Android can
handle those.

Added dictionaries to cross-build-toolset - idxdict is needed.

Should build up to sfx2 - some residual static lib issues there,
and in raptor.

No semantic changes, the iOS code still is mostly dummy.

and deletion of a varialbe used as constant since 2002

but left the call to the function behind, since not sure there is
no effect on data

partially revert 6985b28d as it breaks the whole idea of testing for bug 20033