The present CMake logic wants to pass FIXED:NO to the linker for 64-bit
builds, which on the face of it seems like a Good Thing: it permits code to be
relocated in memory, preventing collisions if two libraries happen to want to
load into overlapping address ranges.

However the way it's being specified is wrong and harmful. Passing /FIXED:NO
to the compiler command line engages /FI (Forced Include!) of a nonexistent
file XED:NO -- producing lots of baffling fatal compile errors.

Thanks Callum for diagnosing this!

Evidently the LL_VIEWER_CHANNEL macro (defined on the compiler command line)
used to contain enclosing double quotes. Something changed (newer CMake
version?) so that the macro now expands as Second Life Release rather than as
"Second Life Release". That leads to syntax errors when it's used.

Add C++ preprocessor trickery to stringize the value of the macro.

Pull in new version of KDU third party package that is build (correctly) as a static library vs. a stub library/DLL

Specifically, DriverCallback() was accepting several DWORD arguments that
should really be DWORD_PTR arguments. In a 32-bit compile, evidently that was
okay, but for 64 bits they're different sizes.

Some day llmake() will be unnecessary because compiler deduction of class
template arguments from constructor arguments has been approved by ISO.

LLPrivateMemoryPool and LLPrivateMemoryPoolManager have assumed that it's
always valid to cast a pointer to U32. With 64-bit pointers, no longer true.

It's never been clear to me why Macs tend to refer to 32-bit Intel processors
as i386 when other platforms tend to refer to them as i686. New CMake logic to
derive ARCH from ADDRESS_SIZE produces i686. Give viewer_manifest.py a
Darwin_i686_Manifest class alias so it continues to work when arch is passed
as i686 as well as i386.

llcorehttp/tests had a clone of llmessage/tests/testrunner.py that was almost
identical save for recognizing an extra optional parameter. Migrate those few
lines into llmessage/tests/testrunner.py; eliminate the copy in llcorehttp;
help test_llcorehttp_peer.py find the testrunner.py in llmessage/tests.

which in turn depends on ADDRESS_SIZE.

There Must Be A Better Way.

with tests on ADDRESS_SIZE, which is now set on the compiler command line.

by adding -DADDRESS_SIZE= to the compile switches.

Remove hack to work around limitations of gcc 4.1 build hosts.

Streamline a bit of logic to specify correct -m32 or -m64 switch.

Use ADDRESS_SIZE instead of ARCH to control -march=pentiumpro.

Migrate the logic formerly used only for LINUX to detect whether ADDRESS_SIZE
is set to 32 or 64, and if not, detect a default for the platform. But instead
of using uname -m, use python's platform.machine().

On Windows, stop forcing ARCH to i686 and ADDRESS_SIZE to 32.

On Mac, reset default to x86_64 instead of i386; stop forcing ADDRESS_SIZE to
32.

autobuild 1.1 now supports expanding $variables within a config file --
support that was explicitly added to address this very problem. So now the
windows platform in autobuild.xml uses $AUTOBUILD_ADDRSIZE,
$AUTOBUILD_WIN_VSPLATFORM and $AUTOBUILD_WIN_CMAKE_GEN, which should handle
most of the deltas between the windows platform and windows64.

This permits removing the windows64 platform definition from autobuild.xml.

The one remaining delta between the windows64 and windows platform definitions
was -DLL_64BIT_BUILD=TRUE. But we can handle that instead by checking
ADDRESS_SIZE. Change all existing references to WORD_SIZE to ADDRESS_SIZE
instead, and set ADDRESS_SIZE to $AUTOBUILD_ADDRSIZE. Change the one existing
LL_64BIT_BUILD reference to test (ADDRESS_SIZE EQUAL 64) instead.

Reverting MAINT-6125

Reverting MAINT-6259

Reverting MAINT-6793

Prervious MAINT-2141 fix causes MAINT-6791 so reverted it