2 trivial patches for libc++

Clang Frontend
1

Hi

Does anyone object to these trivial patches? (If not someone please push)

#1

FreeBSD has a local patch that for some stupid reason they have’t pushed upstream. We recently added a commit that without the libc++ fix causes multiple symbols for us and a build regression.

Here’s the patch
https://svnweb.freebsd.org/base/head/contrib/libc%2B%2B/src/new.cpp?r1=261283&r2=273434

commit message - “libcxxrt now implements bad_array_new_length and need to guard against multiple defines”

#2 The attached patch allows libc++ to use libcxxrt on OSX.

Thanks

libcxx.patch (1.09 KB)

2

FWIW, your description doesn't seem bad, but we can't commit a patch to
libc++ without the author of the patch contributing it by posting it to the
list... Hopefully whomever wrote this is willing to upstream it?

3

The CMake part has a bug:

diff --git a/lib/CMakeLists.txt b/lib/CMakeLists.txt
index cd8553c..d489c90 100644
--- a/lib/CMakeLists.txt
+++ b/lib/CMakeLists.txt
@@ -53,7 +53,7 @@ target_link_libraries(cxx ${libraries})
append_if(compile_flags LIBCXX_HAS_FPIC_FLAG -fPIC)
append_if(link_flags LIBCXX_HAS_NODEFAULTLIBS_FLAG -nodefaultlibs)

-if ( APPLE )
+if ( APPLE AND "${LIBCXX_CXX_ABI}" STREQUAL "libcxxaib" )

This should be "libcxxabi"

Other than that (and needing to test this on OS X with libc++abi to make
sure the typo fix works) the CMake side looks fine. Sadly, the
exception.cpp code is stuff I have never rightly understood well enough to
help with. Hopefully Marshall or someone can help there.

-Chandler

4

The patch is too trivial to be covered by copyright (it moves an #endif down a few lines). Baptiste committed it to the FreeBSD repo, but on my advice so I can count as the person who wrote it if that's really a problem.

David

5

I'd much rather have someone just post a patch they authored than think
about the details of such things.... Much simpler....

6

I agree. Doesn't cost anything, and makes things a lot clearer.

--renato

7

We're talking about moving a #if a couple lines down. I'd agree there's a
line nobody wants to cross, but it's unrealistic to think this could be
done another way. The time taken for 3 people to write emails was longer
than the commit message.

(Now I guess I understand why FreeBSD has local changes that aren't
upstream..)

UGH

8

I’ve reached out to the author of the patch to get his permission. I’ll review and test the CMake patch tonight and then commit it.

/Eric