been tested, the other platforms are educated guesses.

The shebang situation for Windows seems to be a bit odd. Revert the
change in all the cases where it is likely to cause trouble.

Found and reported by:	damjan

The shebang situation for Windows seems to be a bit odd. Revert the
change here as it's affecting the Windows buildbot.

Found and reported by:	damjan

For consistency with RAND(), replace the old Wichman-Hill with
an even older version of George Marsaglia's KISS algorithm.

For the intended use this is more than enough and
still should pass all statistics tests.

This takes care of a deprecation warning.

Documented:
http://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Writing-Autoconf-Input.html#Writing-Autoconf-Input

Gets ScDBQueryDataIterator::DataAccessInternal::getCurrent() to handle formulas returning strings.
Added a spreadsheet test for this behaviour.

OpenOffice source with clang.

Submitted by:	Don Lewis

get tests to compile, but not all of them pass.

get tests to compile and not to SIGSEGV or die from EPIPE,
and get as many tests to pass as possible.

Some old configuration scripts confuse FreeBSD-10 with FreeBSD-1.0. On
regular FreeBSD port builds these had no visible effect since it's logical
to use up-to-date external dependencies but, in the case of the buildbot, 
these were likely to be the culprits behind breakage.

Huge thanks for the reporter for working these out!

Submitted by:	Don Lewis

during debug builds, since it doesn't want to run with them.

Still the path for python doesn't look right.

Mosrt important perl scripts already use a portable method
but fix these nevertheless.