Slience bogus -Werror=maybe-uninitialized

...as emitted by at least GCC 8.2 with --enable-optimized, e.g. at:

> In file included from include/rtl/ustring.hxx:37,
>                  from include/cppuhelper/factory.hxx:26,
>                  from unoxml/source/rdf/librdf_repository.hxx:24,
>                  from unoxml/source/rdf/librdf_repository.cxx:20:
> include/rtl/string.hxx: In static member function ‘static std::shared_ptr<{anonymous}::librdf_TypeConverter::Node> {anonymous}::librdf_TypeConverter::extractNode_NoLock(const com::sun:⭐:uno::Reference<com::sun:⭐:rdf::XNode>&)’:
> include/rtl/string.hxx:294:27: error: ‘*((void*)(& type)+8).rtl::OString::pData’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
>          rtl_string_release( pData );
>          ~~~~~~~~~~~~~~~~~~^~~~~~~~~
> unoxml/source/rdf/librdf_repository.cxx:2094:30: note: ‘*((void*)(& type)+8).rtl::OString::pData’ was declared here
>      boost::optional<OString> type;
>                               ^~~~

This appears to be a common pattern of false positives with uses of
boost::optional, common enough to disable the warning globally for affected
compilers, even if there would also be useful findings by that warning (e.g.,
<https://gerrit.libreoffice.org/#/c/66619/> "Fix -Werror=maybe-uninitialized").

I didn't bother to file a GCC bug for the reproducer used in configure.ac,
<https://gcc.gnu.org/bugzilla/buglist.cgi?quicksearch=maybe-uninitialized>
already shows lots of open bugs in that area, and the documentation at
<https://gcc.gnu.org/onlinedocs/gcc-8.2.0/gcc/Warning-Options.html> states that
"GCC may not be able to determine when the code is correct in spite of appearing
to have an error."

(Clang appears to not support -Wmaybe-uninitialized at all, so exclude it from
the configure.ac check, to not have the check's failure result in an unsupported
-Wno-maybe-uninitialized end up on the compiler command line.)

Change-Id: Ifb9ca4c342750eae54f7e1a01506101310484c7e
Reviewed-on: https://gerrit.libreoffice.org/66621
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>

config_host.mk.in

@@ -244,6 +244,7 @@ export GTK_PRINT_CFLAGS=$(gb_SPACE)@GTK_PRINT_CFLAGS@
 export GTK_PRINT_LIBS=$(gb_SPACE)@GTK_PRINT_LIBS@
 export USING_X11=@USING_X11@
 export HAMCREST_JAR=@HAMCREST_JAR@
+export HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED=@HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED@
 export HAVE_GCC_AVX=@HAVE_GCC_AVX@
 export HAVE_GCC_STACK_PROTECTOR_STRONG=@HAVE_GCC_STACK_PROTECTOR_STRONG@
 export HAVE_GCC_BUILTIN_ATOMIC=@HAVE_GCC_BUILTIN_ATOMIC@

configure.ac

@@ -6540,6 +6540,63 @@ AC_COMPILE_IFELSE([AC_LANG_SOURCE([
 CXXFLAGS=$save_CXXFLAGS
 AC_LANG_POP([C++])
 
+dnl At least GCC 8.2 with -O2 (i.e., --enable-optimized) causes a false-positive -Wmaybe-
+dnl uninitialized warning for code like
+dnl
+dnl   OString f();
+dnl   boost::optional<OString> * g(bool b) {
+dnl       boost::optional<OString> o;
+dnl       if (b) o = f();
+dnl       return new boost::optional<OString>(o);
+dnl   }
+dnl
+dnl (as is e.g. present, in a slightly more elaborate form, in
+dnl librdf_TypeConverter::extractNode_NoLock in unoxml/source/rdf/librdf_repository.cxx); the below
+dnl code is meant to be a faithfully stripped-down and self-contained version of the above code:
+HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED=
+if test "$GCC" = yes && test "$COM_IS_CLANG" != TRUE; then
+    AC_MSG_CHECKING([whether $CXX might report false -Werror=maybe-uninitialized])
+    AC_LANG_PUSH([C++])
+    save_CXXFLAGS=$CXXFLAGS
+    CXXFLAGS="$CXXFLAGS $CXXFLAGS_CXX11 -Werror -Wmaybe-uninitialized"
+    if test "$ENABLE_OPTIMIZED" = TRUE; then
+        CXXFLAGS="$CXXFLAGS -O2"
+    else
+        CXXFLAGS="$CXXFLAGS -O0"
+    fi
+    AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
+            #include <new>
+            void f1(int);
+            struct S1 {
+                ~S1() { f1(n); }
+                int n = 0;
+            };
+            struct S2 {
+                S2() {}
+                S2(S2 const & s) { if (s.init) set(*reinterpret_cast<S1 const *>(s.stg)); }
+                ~S2() { if (init) reinterpret_cast<S1 *>(stg)->S1::~S1(); }
+                void set(S1 s) {
+                    new (stg) S1(s);
+                    init = true;
+                }
+                bool init = false;
+                char stg[sizeof (S1)];
+            } ;
+            S1 f2();
+            S2 * f3(bool b) {
+                S2 o;
+                if (b) o.set(f2());
+                return new S2(o);
+            }
+        ]])], [AC_MSG_RESULT([no])], [
+            HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED=TRUE
+            AC_MSG_RESULT([yes])
+        ])
+    CXXFLAGS=$save_CXXFLAGS
+    AC_LANG_POP([C++])
+fi
+AC_SUBST([HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED])
+
 dnl ===================================================================
 dnl system stl sanity tests
 dnl ===================================================================

solenv/gbuild/platform/com_GCC_defs.mk

@@ -79,6 +79,10 @@ gb_CXXFLAGS_COMMON := \
 	-fno-common \
 	-pipe \
 
+ifeq ($(HAVE_BROKEN_GCC_WMAYBE_UNINITIALIZED),TRUE)
+gb_CXXFLAGS_COMMON += -Wno-maybe-uninitialized
+endif
+
 gb_CXXFLAGS_Wundef = -Wno-undef
 
 ifeq ($(ENABLE_GDB_INDEX),TRUE)