Update libcxx

[waywardmonkeys]

This updates things to the current version of libcxx.

I'm still running the full test suite ...

There's a change that I made:

+#elif defined(__EMSCRIPTEN__)
+#define _LIBCPP_ELAST 256

I'll be upstreaming that as soon as I remember how to use the phabricator stuff. I've mentioned to the Musl people that they might add a ELAST, but I've no horse in the race and little intention at this point to argue strongly with them either way.

cc: @chadaustin

[waywardmonkeys]

Well ...had a scare for a while. test_time was failing ... but then I found that it is failing on the incoming branch for me, so it isn't related to my work here.

All other tests pass that I ran (the normal battery of tests).

[kripken]

By normal do you mean default? how about `other?

[waywardmonkeys]

I mean:

python ./tests/runner.py

I'll try other now ... although I don't recall libcxx having issues in that area in the past.

[waywardmonkeys]

Of course, it would've been nice if other worked on the incoming branch ... (it doesn't)

[kripken]

./tests/runner.py other fails?

[waywardmonkeys]

On incoming? It does for me:

test_EMCC_BUILD_DIR (test_other.other) ... ERROR
test_abspaths (test_other.other) ... ok
test_aliased_func_pointers (test_other.other) ... aliasing
['/usr/bin/python2', '/Users/bruce/Development/emscripten/emcc', 'src.cpp', '-O1']
['/usr/bin/python2', '/Users/bruce/Development/emscripten/emcc', 'src.cpp', '-O1', '-s', 'ALIASING_FUNCTION_POINTERS=0']
['/usr/bin/python2', '/Users/bruce/Development/emscripten/emcc', 'src.cpp', '-O1', '-s', 'ALIASING_FUNCTION_POINTERS=1']
ok
test_argv0_node (test_other.other) ... ERROR

And then I realized I was on incoming and swapped to my update-libcxx branch.

[kripken]

Looks good to me. Just one question - I see some emscripten ifdefs were replaced with LIBCPP_HAS_NO_THREADS. Don't we need to set that to 1?

[kripken]

Those pass on the bots and for me locally, not sure why they would fail for you. But probably not relevant to this pull.

[waywardmonkeys]

Good eye!

You're referring to this and a companion change:

-#if __has_feature(cxx_atomic) && !defined(__EMSCRIPTEN__)
+#if __has_feature(cxx_atomic) && !defined(_LIBCPP_HAS_NO_THREADS)

The reason for that _EMSCRIPTEN__ check there was that there was previously a bug in how emscripten + clang handled atomics ... but that's long been fixed.

Turning on _LIBCPP_HAS_NO_THREADS would be interesting, but:

1. A separate change from this pull
2. It removes a lot of stuff from the headers and might break some people.

[waywardmonkeys]

The usual suspects in other failed ... nothing from libcxx as far as I can see.

[kripken]

Ok, sounds good.