BUG: struct timeval uses unsigned types on at least one platform (QNX).  Alot of the time logic assumes a signed type.  Switch to using a signed representation that is converted to the native representation only for system calls.

ENH: Disabling test 6 on QNX until process killing can be resolved.  It will just fail always for now.

BUG: Extending timeout of test 6 from 0.1 seconds to 3 seconds.  This should avoid missed signals and intermittent failures.

BUG: Drastically extending test5's timeouts to get it to pass when running on a heavily-loaded machine.

COMP: Added KWSys try-compiles KWSYS_STL_HAS_ALLOCATOR_TEMPLATE and KWSYS_STL_HAS_ALLOCATOR_OBJECTS.  Needed for more old-stl support in the hashtable.

BUG: When killing a child all the pipe read ends should be closed.  This will allow a child that is blocking while waiting to write to the pipe to wake up and receive the kill signal properly on cygwin.

BUG: Expanded difference in timeouts between tests 4 and 5 so that 5 does not timeout while waiting for 4 to timeout.  This should fix the intermittent failure of test 5 on cygwin.  ENH: When a mismatch is reported it now reports what it was as well as what it should have been.

BUG: For some reason the non-template allocator test compiles on VS6 even though its allocator is a template.  Adding ::size_type to be sure it accesses a member of the allocator.

BUG: Cannot use Win32 implementation for cygwin processes because then cygwin paths to executables like /usr/bin/ls are not found.

COMP: Added KWSYS_CXX_HAS_CSTDDEF try-compile to KWSys to provide kwsys/cstddef header (to get size_t for hash_fun.hxx).

ENH: Changing kwsysProcess implementation on Cygwin to use the Win32 implementation instead of the UNIX implementation.  This makes dealing with misbehaving children work better.  The KWSys Win32 process implementaion is more robust than the Cygwin implementation (partly because it doesn't have to exactly reproduce the POSIX api).