...in LIBO_INTERNAL_ONLY, __cplusplus, non-MSVC case.
It turns out that sal_Unicode happens to not be mangled into any symbols that
make up the stable URE interface, so (for LIBO_INTERNAL_ONLY, at least) we are
free to replace the typedef to sal_uInt16 with a typedef to any integral type
layout-compatible with that. (sal_Unicode does appear in some symbols in sal's
PRIVATE_textenc.1 section, but that is private between the sal and sal_textenc
libraries, so changing those symbols does not require a change of SONAME.)
C++11 chart16_t is the obvious choice (and will ultimately allow using u"..."
to write literals of type array-of-sal_Unicode). Reportedly, char16_t is
supported since GCC 4.4 and Clang 2.9 but will only be available in MSVC 2015.
For plain C, we continue to use sal_uInt16. We could theoretically use C11
char16_t from <uchar.h>, but at least the Mac OS X 10.11 SDK still does not
offer that C11 header.
For MSVC, we continue to use wchar_t (which is actually unsigned short, due to
/Zc:wchar_t-) for now. Potential options there include dropping /Zc:wchar_t-
and using true wchar_t, or using C++11 char16_t once support for MSVC 2013 is
dropped.
Some code needed to be adapted that was written in a way assuming that
sal_Unicode is unsigned short (which indicates that changing sal_Unicode for
non-LIBO_INTERNAL_ONLY would be an ABI change). OUStringBuffer::append can now
differentiate between being called with sal_Unicode (to append a single
character) and erroneously being called with sal_uInt16 (intending to append a
number's textual representation, for which the sal_Int32 overload must be used
instead). Bugs found are 379fe0409e7973b36210cffa3dd1dfd4032f0ecc "Assume that
this code wants to append a number, not a character" and
dc148335a6a438848325f24c49198fba81043279 "Assume this wants to append the
numerical representation."
The GDB support for pretty-printing of sal_Unicode-related data in
solenv/gdb/libreoffice/sal.py can presumably be simplified now.
Change-Id: I445b3a80e65b7cb004d9e08b38bdc9ee93bc9401
Reviewed-on: https://gerrit.libreoffice.org/20036
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Tested-by: Stephan Bergmann <sbergman@redhat.com>
ie.
void f(void);
becomes
void f();
I used the following command to make the changes:
git grep -lP '\(\s*void\s*\)' -- *.cxx \
| xargs perl -pi -w -e 's/(\w+)\s*\(\s*void\s*\)/$1\(\)/g;'
and ran it for both .cxx and .hxx files.
Change-Id: I314a1b56e9c14d10726e32841736b0ad5eef8ddd
...using C++11-only std::remove_reference, but decltype is also only C++11, so
3rd party C++03 code should have no need for it anyway. (Commits acutally
making use of it to follow shortly.)
Change-Id: I8dd7a64ca73d2082a3e7b74d3599848e65e81da5
Unfortunately iwyu gets quite confused by the weird cyclic dependencies
between various foo.h/foo.hxx and cppumaker generated headers, so it's
not obvious if any improvement here is realistic...
Change-Id: I0bc66f98b146712e28cabc18d56c11c08418c721
...to improve diagnosing misuses of boolean expressions in client code (cf.
compilerplugins/clang/implicitboolconversion.cxx). This change should be
transparent to client code.
Change-Id: I4c9a55f8c7d0f05fda617f9b33d93ce78bb8fbeb
While strictly speaking an incompatible change, it is extremely unlikely that
external code uses cppu::Enterable at all; this should always have been a
private URE implementation detail.
Change-Id: I2c3fe754fe6268b18ca03532229f3403736f6f6e
