Generalized the OSX ppc workaround for ancient bison versions,
factored out for all gcc platforms. Put bison version detection
into configure accordingly, to switch on that, and not on platform.

This patch fixzex bug fdo43586, more room needed for localized strings in named range dialog.

Otherwise we would fail to import the cell contents of those documents
that don't include table styles at all.  Some hand-crafted ods documents
don't provide table styles, which 3.4 imported just fine.

Some parameters don't work with external references, however, such
as PROTECT, PREFIX and WIDTH.

test XDataPilotFieldGrouping::createDateGrouping is broken

Otherwise that method will set #REF! to the cell when it sees the tab
index being -1.  We don't use that index in that code, so it's safe to
do this.

Older releases stored a few function names not defined by ODFF, namely
EASTERSUNDAY instead of ORG.OPENOFFICE.EASTERSUNDAY, TDIST instead of
LEGACY.TDIST and B instead of BINOM.DIST.RANGE.

Since OOo/LibO 3.3 the proper function names can be read, additionally to the
"wrong" names. Now it's time to write the proper names and still accept the
incorrect ones.

Test cases are attached to AOOo issues:
ORG.OPENOFFICE.EASTERSUNDAY
https://issues.apache.org/ooo/show_bug.cgi?id=112882
LEGACY.TDIST and BINOM.DIST.RANGE
https://issues.apache.org/ooo/show_bug.cgi?id=110229
Note that the FALSE in A2 and A3 is a result of the string comparison of the
actual formula, that differs in separators (, vs ;) only.
(cherry picked from commit a9b03bd1)

This is also another unfortunate bug due to the logic change caused
by the DECLARE_LIST removal.  There was one crasher due to out-of-bound
array access, plus one incorrect behavior concerning cell content
placement.  Both are now fixed.

still missing are all properties that are tested automatically by the
PropertyTester

it seems that the java based test fails by "design" and not due to an
error

For now, only COL and ROW are supported.  More on the way.  Note that
we can't support all the information types that we do for internal
references.

 * on OSX 10.5 PPC, ld crashes with a bus error, presumably because
   of the empty ar[] string array.
 * the "size = 0" default case looks wrong, replaced by *size=0

Some places in the code assumed that if the stack is empty a null pointer is
returned by top() (or old Pop()), this doesn't work anymore with ::std::stack
that instead has undefined behavior in that case, so check !stack.empty()
first before accessing top.
(cherry picked from commit ac40f7d6)

gsoc input bar now (should) snuggly wrap the text when collapsed, if multilines exist then scrolling with the keys should be clean and no spill from upper or lower lines visible. Removed some strange ( pseudo padding ) apparently for taking into account the window border and reorganized the resize logic so the padding is added ( and commented why ) in just one place.

Another unfortunate logic change introduced via DECLARE_LIST removal.
Reverting the logic to the original one.