Summary: All is good now.

Test Plan: Run all the tests.

Reviewed By: hans@fb.com

Subscribers: jsedgwick, trunkagent, fugalh, folly-diffs@

FB internal diff: D1758272

Signature: t1:1758272:1421889405:a015d30783715e106a1e6667e971f7cf47c8392d

Summary: Not sure why we'd need that? This avoids a move, I think.

Test Plan:
Run all the tests. Also unbreaks this code, which used to work at some
point.
https://phabricator.fb.com/diffusion/FBCODE/browse/master/taoThriftService/TaoServiceHandler.cpp;a27bc2ef36cd671d65ae0fd2cc1fb1f823e68ae4$246

Reviewed By: hans@fb.com

Subscribers: trunkagent, folly-diffs@, jsedgwick

FB internal diff: D1800619

Signature: t1:1800619:1422053903:f9e65a0be3d820a9a9989b3cf5dfaf2a61e2e900

Test Plan: g-unittest

Reviewed By: philipp@fb.com

Subscribers: folly-diffs@

FB internal diff: D1801665

Signature: t1:1801665:1422093104:eecf86ea8c1ab39dcb3d01442ed66ae45ef3bede

Summary: I'm in unfamiliar territory, so shout if I'm doing something dumb.  Perhaps it's a bad assumption that if the ipv4 port is free that the ipv6 port is also free?

Test Plan: g-unittest

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, ps, bmatheny, folly-diffs@

FB internal diff: D1795120

Signature: t1:1795120:1422034693:bd315023ab6cd9e9bda12161d05dd781dc401546

Summary: Now all connections are always scheduled with connectionManager's default timeout.  This diff enables us to specify the different timeouts for each managed connections.

Test Plan: prospect testing

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, folly-diffs@, jsedgwick

FB internal diff: D1797193

Tasks: 5343753

Signature: t1:1797193:1422034092:ed67b96efe8af8f8d1355d3f86fb1289daafb178

Summary: Future service interfaces similar to finagle.  Service creators for client, filters

Test Plan: Unittests included - also sets up a simple pipeline to test a full stack client/server.

Reviewed By: hans@fb.com

Subscribers: jsedgwick, trunkagent, njormrod, folly-diffs@, doug, fugalh

FB internal diff: D1573086

Tasks: 5002456

Signature: t1:1573086:1421970698:328453c4a980bb6950fc9aeed6a2b6d9819c20db

Summary:
Subject:

This reverts commit d8eeb9a12f932b74d268c74ce7ce2610dc14b152.

Test Plan: Watch contbuild?

Reviewed By: njormrod@fb.com

Subscribers: trunkagent, doug, fbcode-common-diffs@, hero-diffs@, cold-storage-diffs@, adamsyta, zhuohuang, darshan, micha, folly-diffs@, lins, tingy, hannesr, jsedgwick

FB internal diff: D1797163

Tasks: 5940008

Signature: t1:1797163:1421959040:c9d528367fed2de2f7dafdac9416aa01bdb47df9

Summary:
This also adds a dependency from all folly::Singletons on RequestContext. Better fix would probaly be to switch RequestContext to be managed by folly::Singleton, however that would require us to switch all singletons which use it to be folly::Singletons too.

Facebook:
Making sure we initialize SR singletons when OBCClient singleton is created, so that SR is destoyed after OBCClient.

Test Plan:
fbconfig -r --sanitize=address --platform-all=gcc-4.8.1-glibc-2.17 dragon/
fbmake runtests

Reviewed By: meyering@fb.com, davejwatson@fb.com

Subscribers: trunkagent, folly-diffs@, meyering

FB internal diff: D1796847

Tasks: 5986816

Signature: t1:1796847:1421954531:e9ec177d5289a33c12fe0947426d89963c8e262c

Summary:
Histogram::getPercentileBucketIdx() and getPercentileEstimate() are
implemented using template methods of HistogramBuckets.

This updates Instantiations.cpp to also instantiate the versions of
these template methods that are required by Histogram<int64_t>.  Without
this, anyone using these methods was required to also include
Histogram-defs.h, or otherwise they may get link errors.  This makes it
so they no longer need to include Histogram-defs.h

The C++ syntax required is pretty horrendous.  We could potentially
avoid this by refactoring the code so that these methods are no longer
templates (and require that the Bucket class provide methods for this
purpose).  At the moment adding instantiations seems simpler, and
doesn't change any public APIs.

Test Plan:
Successfully linked code using
Histogram<int64_t>::getPercentileEstimate() without including
Histogram-defs.h

Reviewed By: ldbrandy@fb.com, meyering@fb.com

Subscribers: trunkagent, doug, net-systems@, exa, folly-diffs@

FB internal diff: D1786346

Signature: t1:1786346:1421429734:da17b7ea326c64a1e158fec3b87dc49c3db3d848

Summary:
This splits out some code from the existing getIov() method, to support
appending to an existing iovec array, rather than always returning a new
one.

I have a use case where I have multiple IOBuf chains to write at once,
preceded by some header data.  appendToIov() makes it easier to populate
a single iovec array with this data, and then perform a single writev()
call.

Test Plan: Updated the unit tests to exercise appendToIov()

Reviewed By: jasmeetbagga@fb.com

Subscribers: trunkagent, doug, exa, net-systems@, folly-diffs@

FB internal diff: D1783090

Tasks:

Signature: t1:1783090:1421450476:d5c89d6e1adbd2f2239057238e020b302df09f23

Blame Revision:

Summary:
Introduce ProgressableExecutor, which is an Executor that can be driven somehow. Examples include EventBase and ManualExecutor
Then introduce Future<T>::getVia(ProgressableExecutor*) and Future<T>::waitVia(ProgressableExecutor*) that drive the given executor until the Future is complete, with the usual semantics of get and wait respectively
This is a really common pattern in tests and you can see in the various changes to other projects lends sopme nice redness and cleanliness

Some notes:
1. I don't like ProgressableExecutor::makeProgress() that much. Too verbose. Maybe DrivableExecutor::drive()? Something else? Thoughts?
2. I still need to integrate this with some stuff in Zeus (SessionFuture) and Zookeeper (ZookeeperFuture) but I'm going to do that in a separate diff because it's going to be a little more intrusive
3. These APIs take a raw ptr so that they are consistent with via()
4. See inline note on ManualExecutor
5. See inline note in added unit tests

Test Plan: add unit for new API, wait for contbuild

Reviewed By: hans@fb.com

Subscribers: trunkagent, dresende, pzq, tdimson, fbcode-common-diffs@, targeting-diff-backend@, alandau, apollo-diffs@, bmatheny, everstore-dev@, zhuohuang, laser-diffs@, mshneer, folly-diffs@, hannesr, jsedgwick

FB internal diff: D1789122

Tasks: 5940008

Signature: t1:1789122:1421868315:6ea2fc2702be1dc283c24a46d345fb5da3935b32

Summary: as above. nice redness. couple of comments inline though

Test Plan: contbuild

Reviewed By: hans@fb.com

Subscribers: trunkagent, zeus-diffs@, cold-storage-diffs@, targeting-diff-backend@, apollo-diffs@, micha, folly-diffs@, jsedgwick

FB internal diff: D1787776

Tasks: 5940008

Signature: t1:1787776:1421776843:13772348538a0af27ceb5a363b818ece0cdfffc6

Summary:
see task. also adjust all callsites to get() or wait() as appropriate. this is a lot better, especially in various tests

Facebook:

cc @rushix, @wez as this was quite nice for docstore and novak, and you might want to be aware of new conventions

Test Plan: futures unit, wait for contbuild

Reviewed By: hans@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, hero-diffs@, cold-storage-diffs@, adamsyta, zhuohuang, darshan, micha, folly-diffs@, lins, tingy, hannesr, jsedgwick, wez, rushix

FB internal diff: D1785572

Tasks: 5940008

Signature: t1:1785572:1421866794:a879de4d0bc14e96c434f623ed2a74361e25f28c

Summary:
Start of codec framework.  Copied the frame based codecs almost exactly from netty, but made to fit our pipeline management.

BytesToMessageCodec is slightly different: Netty preprocesses all the available data to a List<Message>, while this codec a) Only does one message at a time to avoid queueing issues, and b) doesn't template the message type, and just passes an IOBuf

I'm fighting the type system to get the pipelines to play nice to each other:  I'd rather template the message type, but it prevents stacking outgoing handlers on top of it, since you ahve to specify both the input/output type for each handler, even if you only care about the output type.  Suggestions to fix?   Netty gets around this by lots of dynamic casting to Object type, but we can't do that in C++ since we don't have a base object type

Test Plan: Includes lots of tests

Reviewed By: jsedgwick@fb.com

Subscribers: jsedgwick, doug, fugalh, folly-diffs@

FB internal diff: D1758189

Tasks: 5002361, 5002316

Signature: t1:1758189:1421170225:6bc8cc6a0bb461a965665bc88f7009033b215da9

Test Plan:
fbconfig folly/experimental/test:singletonvault_c_test && fbmake runtests
buck test folly/experimental/test:singletonvault_c_test

Reviewed By: andrii@fb.com

Subscribers: trunkagent, sdwilsh, folly-diffs@

FB internal diff: D1780054

Signature: t1:1780054:1421187023:b18c1d5575b2864b3d3df001a0024b91eb321c46

Summary:
IdleTime unit test has been failing randomly (#5996886). The
event base loop calculates an exponentially moving average of loop busy
time in each loop. In the test, the busy time is initialized to 5900 and
each loop runs more than 8000 us. In normal case decaying of the
previous loop busy time will take 6 loops to  move this average to >6000.
But if the test is running on a heavily loaded machine the busy+idle time
a loop is longer than it should be, causing the decaying happen faster.
This diff skips this test if this happens.

Test Plan:
It won't affect anything if running on my devserver. It only
helps on a heavily loaded contbuild host.

Reviewed By: alandau@fb.com

Subscribers: folly-diffs@

FB internal diff: D1788450

Tasks: 5996886

Signature: t1:1788450:1421460774:1ec575c50f881e10b5a0208717fe68164f0d0f57

Test Plan: checked this is the right function name

Reviewed By: ngbronson@fb.com

Subscribers: folly-diffs@

FB internal diff: D1788386

Signature: t1:1788386:1421453582:82d0c4685c1bd2d888f5974fd2a998ebbc27a9b5

Summary:
The big comment block at the top of MPMCQueue had a non-sense
sentence resulting from a merge conflict, and also didn't include
details about ordering or recommended patterns for handling consumer
shutdown.

Test Plan: comment change

Reviewed By: lesha@fb.com

Subscribers: folly-diffs@

FB internal diff: D1788230

Signature: t1:1788230:1421451643:7c768c2b127499e3f2b06ee241fba851b4ed2a29

Summary:
We want all socket options to be applied to the listen sockets too.

The old code here that only applied SO_KEEPALIVE is a bit superstitious.
Not all options will be inherited by accept()ed sockets, but that's
fine.  We set them again post-accept().

Applying all these options to the listening sockets will fix at least
one long standing bug, which is that QoS is not set correctly for the
first few packets in a flow.  This also will let us do the right thing
when we gain the ability to set TCP_MAXSEG in D1741753.

Test Plan: fbmake runtests & canary in prod

Reviewed By: afrind@fb.com, davejwatson@fb.com

Subscribers: jsedgwick, satadru, trunkagent, fugalh, exa, folly-diffs@, agartrell

FB internal diff: D1745182

Tasks: 2911597

Signature: t1:1745182:1418860157:045680a91b153482bcd4a014894fb28059955d06

Summary:
My initial change to avoid the -Wsign-compare error neglected
the fact that the difference of two unsigned types is also
unsigned. That broke some unicorn tests, as seen in D1776343,
which worked around the problem by changing the type of
bitOffset_ like this: s/unsigned int/size_t.

The real problem lay in the distance_to function, which computed
the signed difference of two bitOffset_ values.
To work properly with the narrower unsigned type,
we could either first widen each operand to ssize_t
before computing their parenthesized difference, or
(as Philip suggested and I prefer), just drop the
parentheses and let the signedness of the preceding
operands ensure that each successive intermediate
result is also signed.

Test Plan:
Choose one of the failing tests, and ensure it now passes:

fbconfig -r unicorn/diskindex4/test:doc_map_test
fbmake runtests

Also run all unicorn tests, for good measure (still waiting):
fbconfig -r unicorn && fbmake runtests

Reviewed By: philipp@fb.com, andrei.alexandrescu@fb.com

Subscribers: trunkagent, net-systems@, folly-diffs@

FB internal diff: D1776624

Tasks: 5941250

Signature: t1:1776624:1420872874:b1ea8a3f23f21269aa65cbe1647e86a8f622a710

Blame Revision: D1770613

Summary:
it's redundant and we shouldn't encourage it, so i'm changing all callsites
i changed docs as well

Test Plan: wait for contbuild

Reviewed By: hans@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, hero-diffs@, cold-storage-diffs@, targeting-diff-backend@, adityab, everstore-dev@, zhuohuang, darshan, micha, folly-diffs@, wch, lins, tingy, jsedgwick

FB internal diff: D1784667

Tasks: 5936469

Signature: t1:1784667:1421364620:83169739320e5342d28a744e3689794f16108fea

Summary: grep around for wangle, adjust where appropriate

Test Plan: contbuild

Reviewed By: hans@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, ruibalp, zeus-diffs@, vikas, targeting-diff-backend@, alandau, mwa, jgehring, zhguo, jying, darshan, fuegen, mshneer, folly-diffs@, lins, maxwellsayles, jsedgwick

FB internal diff: D1777581

Tasks: 5960242

Signature: t1:1777581:1421361395:2ee8bfeca76dafe7376a217c66e0a4eaf3ef416a

Summary:
* folly/MPMCQueue.h (waitForTurn): My changes to accommodate
-Wsign-compare introduced a bug that could cause lots of wasted CPU.
(See @scottf's comments on the blamed patch.)
Subtraction involving an unsigned type will always yield an unsigned
result, and there was at least this one instance where the result
would sometimes have to be a small negative number.  Instead, we'd
get approximately UINT32_MAX.  The fix is to explicitly cast the
difference to a signed type.

Test Plan:
Run this and note there is no error:
(this ensures it remains -Wsign-compare-clean with gcc-4.9)

fbconfig --platform-all=gcc-4.9-glibc-2.20 tao/server && fbmake dbgo

Before this change, running the mpmc_queue_test test would get
timeout errors ~8 in 100 runs.  Confirm that with this patch, now
there is no type of failure. I.e., run these commands and ensure
the grep prints nothing:

fbconfig folly/test:mpmc_queue_test
for i in $(seq 100); do fbmake runtests_dbgo >& makerr-$i; done
grep -E '(TIMEOUT|FATAL|FAIL).* [1-9]' makerr-*

Reviewed By: hans@fb.com

Subscribers: trunkagent, scottf, folly-diffs@

FB internal diff: D1783672

Tasks: 5941250

Signature: t1:1783672:1421350260:d7d306e8cf3ed3ee636d06228e1f24a2c2796b54

Blame Revision: D1766722

Summary:
configure seems to think ubuntu 14.04 doesn't have pthread_atfork, which seems wrong?! But this test seems to fail if we don't have it on some platform anyway.  Not sure the right fix, just disable it for now?

Test Plan: watch https://ci-builds.fb.com/job/folly/

Reviewed By: hans@fb.com

Subscribers: doug, folly-diffs@

FB internal diff: D1783046

Signature: t1:1783046:1421348575:d279877abcc597a04c285092a5e5e53406922d5a

Blame Revision: D1768669

Summary:
i) FOLLY_HAVE_INT128_T definition is duplicated and is getting set to 1 on android. deleted that.
ii) we dont have pthread_spinlock_t and pthread_atfork on android, the former is absent apple as well
iii) a condition got flipped in D1768669. __ANDROID__ implies !FOLLY_HAVE_PTHREAD_SPINLOCK_T. fixed that

Test Plan:
- fbconfig -r folly && fbmake
- ran folly build with buck as well
- synced to fb4a and it builds

Reviewed By: seanc@fb.com

Subscribers: doug, seanc, folly-diffs@

FB internal diff: D1783579

Signature: t1:1783579:1421350138:9301303f77784e7bdb851e3f643751078457b025

Summary:
unfortunately, can't use atomics now that a weak_ptr is stored instead of a raw ptr, hence the additional singleton locks
might want to make an overload of via() that takes shared_ptr to integrate more easily? or change via() to store a shared_ptr (and make it non-owning if a raw ptr is passed)

Test Plan: unit

Reviewed By: davejwatson@fb.com

Subscribers: jsedgwick, trunkagent, fugalh, folly-diffs@

FB internal diff: D1764359

Tasks: 5002442

Signature: t1:1764359:1420845340:349ea88091d7ca4ee386b54aec599647341fadd4

Summary: http://www.cplusplus.com/reference/map/map/at/

Test Plan: unittest

Reviewed By: ldbrandy@fb.com

Subscribers: trunkagent, jkong, folly-diffs@

FB internal diff: D1773996

Signature: t1:1773996:1421176834:33c8b39f03bf085ca2dd45df113cba948d0f2dd0

Summary:
This is a bootstrap script for building folly on osx. I attempted to make it Just Work™ for anyone who uses Homebrew (http://brew.sh/) but even if it doesn't work out of the box it should be instructive for the motivated developer on OSX.

Facebook:
In #5134151 Wangle used to not build, because something was using a futex. But that seems to have been rectified (or the build problems are hiding in headers that aren't used in the libfolly build).

I'll also update https://our.intern.facebook.com/intern/wiki/Folly/BuildingOnMac to point people to this script instead.

Test Plan:
Tested on my laptop which is running Yosemite (10.10.1)
Build, then inspect `otool -L .libs/*dylib`

Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, exa

FB internal diff: D1765248

Tasks: 5134151

Signature: t1:1765248:1420500905:bc5da97a0446f502cb2badc85c9004ff0a62027a

Summary:
SpookyHashTest now needs libgtestmain to be linked
(https://phabricator.fb.com/D1775555).

https://ci-builds.fb.com/job/folly/1570/console

Test Plan:
Don't have Ubuntu setup, but this should fix it. Will kick off jenkins
build.

Reviewed By: andrewjcg@fb.com

Subscribers: folly-diffs@

FB internal diff: D1779604

Signature: t1:1779604:1421174608:de865529d728e943288b52c0d63ebb712eda094f

Blame Revision: D1775555

Summary: Namespace change. Next up is to extract the Executors into folly/executors/

Test Plan: wait for contbuild

Reviewed By: davejwatson@fb.com

Subscribers: jsedgwick, trunkagent, fbcode-common-diffs@, chaoyc, search-fbcode-diffs@, lars, davejwatson, ruibalp, hero-diffs@, zeus-diffs@, andrewcox, vikas, mcduff, cold-storage-diffs@, unicorn-diffs@, ldbrandy, targeting-diff-backend@, netego-diffs@, abirchall, fugalh, adamsyta, atlas2-eng@, chenzhimin, mpawlowski, alandau, bmatheny, adityab, everstore-dev@, zhuohuang, mwa, jgehring, prometheus-diffs@, smarlow, akr, bnitka, jcoens, benj, laser-diffs@, zhguo, jying, darshan, micha, apodsiadlo, alikhtarov, fuegen, dzhulgakov, jeremyfein, mshneer, folly-diffs@, wch, lins, tingy, hannesr, maxwellsayles

FB internal diff: D1772779

Tasks: 5960242

Signature: t1:1772779:1420751149:bc0b9220be25f1d46d9cef3fdeaa9c3681217aff

Summary: Also add the `@buckify` tag to keep these automatically synced.

Test Plan: built and tested with buck and fbconfig/fbmake

Reviewed By: robbert@fb.com

Subscribers: trunkagent, sdwilsh, fugalh, folly-diffs@, jsedgwick

FB internal diff: D1775555

Signature: t1:1775555:1420840948:e9a2e86379be71b7d953fa53a727e51d2541b011

Summary: D1770613 made bitsPerBlock 32 bit. Sadpanda.

Test Plan: _bin/unicorn/diskindex4/test/doc_map_test --gtest_filter=DocMap/DocMapTest.Deleted/23

Reviewed By: meyering@fb.com

Subscribers: net-systems@, folly-diffs@

FB internal diff: D1776343

Tasks: 5968993

Signature: t1:1776343:1420852923:f35166dd9023a701ff16467710131f821810215a

Blame Revision: D1770613

Summary:
* folly/dynamic-inl.h (format): Add an explicit
int-to-size_t cast (that is ok because we've just ensured
it is non-negative) to avoid this error from gcc-4.9:

folly/dynamic-inl.h:953:29: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]

Test Plan:
Run this and note there are fewer errors than before:
fbconfig --platform-all=gcc-4.9-glibc-2.20 -r folly && fbmake dbgo

Reviewed By: delong.j@fb.com

Subscribers: trunkagent, folly-diffs@

FB internal diff: D1773242

Tasks: 5941250

Signature: t1:1773242:1420822301:b1e3c065cd25728c77957ce9e1f52e884ba8693b

Summary:
There were many templates instantiated with signed types.
Changing them to unsigned avoids the compilation failures
noted below.  However, we'll have to watch arc-diff-spawned
tests for new failures, just in case.

This is by far the largest undecomposable patch of those
I've written for this task so far.  It is possible that
some subset of these changes accomplishes the same goal,
but once I managed to get a successful compilation, I have
only re-reviewed it once and did not try to shrink it.

Test Plan:
Without these changes, we would get the following 32 errors from gcc-4.9:
https://phabricator.fb.com/P19692642
With them, (and all of the other patches attached to the parent task,
all of fbcode compiles and passes its tests (modulo preexisting
failures and aborts).

Reviewed By: delong.j@fb.com

Subscribers: trunkagent, folly-diffs@

FB internal diff: D1773461

Tasks: 5941250

Signature: t1:1773461:1420821964:61635f3a2efc95e8d4457c36e7a5ee3a0a50df23

Summary:
First step of moving futures out of wangle into folly

Namespace change will be another diff

This was just a straight codemod of "wangle/futures" to "futures"

Test Plan: wait for contbuild

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, chaoyc, search-fbcode-diffs@, lars, ruibalp, hero-diffs@, zeus-diffs@, vikas, mcduff, hitesh, cold-storage-diffs@, unicorn-diffs@, ldbrandy, targeting-diff-backend@, netego-diffs@, fugalh, adamsyta, atlas2-eng@, alandau, apollo-diffs@, bmatheny, adityab, everstore-dev@, zhuohuang, sweeney, mwa, jgehring, prometheus-diffs@, smarlow, akr, bnitka, jcoens, zhguo, jying, darshan, apodsiadlo, alikhtarov, fuegen, dzhulgakov, alihussains, jeremyfein, mshneer, folly-diffs@, wch, lins, tingy, maxwellsayles, hannesr

FB internal diff: D1772041

Tasks: 5960242

Signature: t1:1772041:1420739169:36126b1264c5d1747d241921d02b13e306c73fe1

Summary:
* folly/Bits.h (BitIterator) [bitOffset_]: Change type of this
member from ssize_t to "unsigned int".
(bitsPerBlock): Change return type from size_t to "unsigned int".
These changes avoid some -Wsign-compare-highlighted issues, e.g.,
folly/Bits.h:456:18: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]

Test Plan:
Run this and note there are fewer errors than before:
fbconfig --platform-all=gcc-4.9-glibc-2.20 -r folly && fbmake dbgo

I've also run "fbmake runtests", and it failed precisely the way it does
without this patch.

Reviewed By: andrei.alexandrescu@fb.com

Subscribers: trunkagent, net-systems@, folly-diffs@

FB internal diff: D1770613

Tasks: 5941250

Signature: t1:1770613:1420757397:7ee7e5de500c8991af2597a6245d12d8f286b4b7

Summary:
With the upgrade to gcc-4.9 fbcode's default is to enable
-Wsign-compare by default.  That exposes nontrivial technical debt.
This is part of that clean-up.  Some of these changes fix the problem
where there's an "int" for-loop index, yet an unsigned limit.
The fix is to use an unsigned type when the limit is also unsigned.
I usually choose size_t because of the general recommendation (when
writing portable code) to avoid size-tied types like uint32_t and
uint64_t unless you have a very good reason to require them.

Test Plan:
Run this and note there are fewer errors than before:
fbconfig --platform-all=gcc-4.9-glibc-2.20 tao/server && fbmake dbgo

Also, run this, compiling and running tests with gcc-4.8.1.
Verify that there are no more failures than without this patch:
fbmake opt -k && fbmake runtests_opt
Here's the tail of that process, (same with and without the patch):
Summary (total time 47.45s):
PASS: 1949
FAIL: 1
SKIP: 0
FATAL: 9
TIMEOUT: 0

Reviewed By: hans@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@

FB internal diff: D1766722

Tasks: 5941250

Signature: t1:1766722:1420762240:a8545648ddb4fd0b2adf0d3147a84409c0f706a8

Summary: Apparently there is some dispute about the correct thing for `snprintf(buf, len, "bogus%", "foo")` to return. OSX says 5, Linux says -1. The code in `folly/String.cpp` depends on -1 to throw the exception this test was expecting.

Test Plan: make check on OSX

Reviewed By: chip@fb.com

Subscribers: exa, folly-diffs@

FB internal diff: D1770970

Tasks: 5936226

Signature: t1:1770970:1420680750:bafae2aa8327a52013e68af64a163323244ca03b

Summary:
This switches from gcc-4.9 provided by Homebrew to the default g++ (clang) from Yosemite.

$ g++ --version
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 6.0 (clang-600.0.51) (based on LLVM 3.5svn)
Target: x86_64-apple-darwin14.0.0
Thread model: posix

Absolute path for double-conversion, so make in `test/` works.

gtest-1.7.0 (Using 1.6.0 causes compilation errors in Yosemite)

Various portability fixes.

Enhancements/fixes to `configure.ac` and associated source code for better portability.

Test Plan:
It builds and runs (yay)
Two tests fail, not sure whether that's specific to osx. Will investigate and make any fixes in a follow-up diff

Reviewed By: pgriess@fb.com

Subscribers: trunkagent, folly-diffs@, exa

FB internal diff: D1768669

Tasks: 5936226

Signature: t1:1768669:1420728580:dd6919c21e0edf30788e523b16e3a5b923e2d7f0