Summary: Mcrouter still enables these by default.  Tracking down a perf regression in another code path that uses fibers.

Reviewed By: @JohnRambo

Differential Revision: D2320772

Summary: Add a golden struct for checking the size of Core.
It makes this test pass on OSX
(now failing because of the different size of std::function).

Reviewed By: @​hannesr

Differential Revision: D2315429

Summary: [Folly] ScopedEventBaseThread should support an EventBaseManager context.

Reviewed By: @Gownta

Differential Revision: D2314049

Summary: [Folly] Use static const keys in AtomicHashArray::Config.

From https://github.com/facebook/folly/pull/264/, MSVC may have a hard time with certain expressions. D2284130 factored them out as `static constexpr` members, but Clang had a hard time with that in HPHP. This adds a test case that triggers the same failure to the Folly build. Not quite sure how this impacts MSVC though.

Reviewed By: @Gownta

Differential Revision: D2304346

Summary: Because `.native()` returns a wide string on MSVC, but a normal string is expected by the rest of the code.
Closes #283

Reviewed By: @yfeldblum

Differential Revision: D2307488

Pulled By: @sgolemon

Summary: MSVC has some odd accessibility rules,
so this makes a couple of things public rather than private.
Closes #284

Reviewed By: @yfeldblum

Differential Revision: D2307578

Pulled By: @sgolemon

Summary: Make RetryingTest more robust by giving it multiple chances to
meeting timing criteria.

Reviewed By: @yfeldblum

Differential Revision: D2306245

Summary: [Folly] Refactor extract FOLLY_SYNCHRONIZED_HAVE_TIMED_MUTEX.

The same complex check is done in two places. Ick. Extract it to a single place.

Reviewed By: @​mzlee

Differential Revision: D2302885

Summary: Fix incorrect using of pthread in folly::ThreadLocal.

The root of the trouble (see man pthread_key_create):
"At thread exit, if a key value has a non-NULL destructor pointer,
and the thread has a non-NULL value associated with that key,
the value of the key is set to NULL, and then the function pointed to is called
with the previously associated value as its sole argument."

At thread exit we need to recover the thread-specific threadEntry
otherwise the subsequent calls of getThreadEntry may recreate it
(what actually DOES  happen in the test ThreadLocalPtr.CreateOnThreadExit)
and the newly created threadEntry will "leak". It will live longer than the corresponding
thread violating the internal invariant of StaticMeta and also it will break
the code of the method onThreadExit. In particular this bug causes the failure
of the test ThreadLocalPtr.CreateOnThreadExit on OSX.

Reviewed By: @lbrandy

Differential Revision: D2304950

Summary: It was originally unconditionally useing `__attribute__` syntax, so this makes it conditionally use the `__declspec` syntax.
Closes #268

Reviewed By: @yfeldblum

Differential Revision: D2283909

Pulled By: @sgolemon

Summary: [Folly] Fix ASAN failure in folly/io/async/test/NotificationQueueTest.cpp.

Reviewed By: @djwatson

Differential Revision: D2299112

Summary: D2284130 broke the HHVM clang build.

Reviewed By: @mxw

Differential Revision: D2303914

Summary: Temp remove EventBaseLocal on mobile.

Reviewed By: @djwatson

Differential Revision: D2302322

Summary: Do not use these if __ANDROID__ is defined as well.

Reviewed By: @​yangchi

Differential Revision: D2301782

Summary: I have no idea why this specific declaration would error, or if there is a difference in meaning, but this fixes MSVC's absurd parser error.
Closes #277

Reviewed By: @yfeldblum

Differential Revision: D2284096

Summary: Improves MSVC support.
Closes #274

Note that this diff has been heavily modified from @Orvid's original PR by @yfeldblum and @sgolemon

Reviewed By: @yfeldblum

Differential Revision: D2284035

Pulled By: @sgolemon

Summary: Closes #264

Modified by @sgolemon from the original PR to declare MSVC2015-final as the official minimum version, making the defines in the original PR unnecessary.

Reviewed By: @yfeldblum

Differential Revision: D2284130

Pulled By: @sgolemon

Summary: There are a bunch of methods in `folly::dynamic` that return a
reference to a sub-object, e.g. `getString()`.
These methods are a bit unsafe because it allows you to write code with dangling
references when the `folly::dynamic` object is an rvalue:

  auto& obj = parse_json(some_string).at("key");

However, the bigger win is that when we have an rvalue `folly::dynamic`, such as
returned by `getDefault(...)`, we can move the sub-object out:

  auto obj = parse_json(some_string).at("key");
  auto str = obj.getDefault("another-key", "").getString();

In the first line, the subobject is copied out when it could be safely moved out.
In the second line `getDefault(...)` copies the dynamic sub-object out, and then
`getString()` copies the string out, when the string could be moved.

Also in the case of `getDefault()` being called on a rvalue, we can move the
sub-object out.

Reviewed By: @marcinpe

Differential Revision: D2267588

Summary: remove unnecessary stuff

Reviewed By: @siyengar

Differential Revision: D2284095

Summary: Add accessor to estimate a FiberManager's run queue size

Reviewed By: @sarangbh

Differential Revision: D2293367

Reviewed By: @yfeldblum

Differential Revision: D2300047

Summary: Various fixes to get tests passing using buck's dev mode.

Reviewed By: @yfeldblum

Differential Revision: D2299981

Summary: [Folly] futures::retrying.

Reviewed By: @fugalh

Differential Revision: D2201630

Summary: enum iterators operator!= didn't use the iterator parameter for comparison
because they assumed you only ever compare to end. Change the operator!= to
look at the iterator parameter to check if it was end.

Thrift generated code for reflection initializers for some types that weren't
declared before definition because they were never called outside of the file.
Marked those initializers as 'static'.

Reviewed By: @alandau

Differential Revision: D2284371

Summary: This has come up a couple times, and the implementation is never nice.  Just like we have thread-local storage, it will be useful to also have evb-local storage.  Provides at pretty simple get/set/delete interface (see unittest).

Reviewed By: @djwatson

Differential Revision: D2203063

Summary: folly only accepts "Infinity" while deserializing but writes "infinity" when serializing.
This means folly cannot deserialize what it serialized before.
If we agree on this, we could update ##fbcode/common## as well: https://fburl.com/136793901

Reviewed By: @marcinpe

Differential Revision: D2293627

Summary: This `typename` qualifier was unneeded, and MSVC errors if it's there, so remove it.
Closes #275

Reviewed By: @yfeldblum

Differential Revision: D2284056

Pulled By: @sgolemon

Summary: MSVC doesn't support constructor priorities, so use the same code as OSX.
Closes #272

Reviewed By: @yfeldblum

Differential Revision: D2283998

Pulled By: @sgolemon

Summary: Support for this under mingw was already present, this just uses that for MSVC as well.
Closes #273

Reviewed By: @yfeldblum

Differential Revision: D2284007

Pulled By: @sgolemon

Summary: Specifically, MSVC doesn't define `std::__ostream_insert`, so just write to the stream instead.
Closes #270

Reviewed By: @yfeldblum

Differential Revision: D2283960

Pulled By: @sgolemon

Summary: Because MSVC doesn't support `__thread`, and `FOLLY_TLS` is already setup to use the MSVC syntax when needed.
Closes #269

Reviewed By: @yfeldblum

Differential Revision: D2283919

Pulled By: @sgolemon

Summary: This adds `asm_volatile_memory`, which goes along with the same style used by `asm_volatile_pause`.
This also switches the two places in `RWSpinLock.h` that were using inline assembly for this to use the new functions instead.
Closes #260

Reviewed By: @yfeldblum

Differential Revision: D2283541

Pulled By: @sgolemon

Summary: This adds `FOLLY_ALIGNED` to `Portability.h`, and adjusts the places that were previously using the raw aligned attribute to use this instead.

Closes #262

Reviewed By: @yfeldblum

Differential Revision: D2283639

Pulled By: @sgolemon

Summary: With MSVC, `seedData.begin()` does not return a `void*`, so get a pointer to the data, and explicitly cast that to a `void*` instead.
Closes #251

Reviewed By: @yfeldblum

Differential Revision: D2282929

Pulled By: @sgolemon

Summary: This implements it via a function called on startup. The mechanism used is roughly equivelent of `__attribute__((__constructor__))` in that it doesn't have the ability to set the order of invokation.
Closes #249

Reviewed By: @yfeldblum

Differential Revision: D2282860

Pulled By: @sgolemon

Summary: Before this change you would get an error like:
"Expected to get greater than or equal to1 for value 0"

Now it add in the space correctly:
"Expected to get greater than or equal to 1 for value 0"

Reviewed By: @snarkmaster

Differential Revision: D2288454

Summary: exit() is evil, let's make it easier for programs that want to exit
successfully.

Reviewed By: @meyering

Differential Revision: D2290201

Summary: This diff adds partial specializations of folly::hasher for enum types, std::pair and std::tuple.
We also restrict the specialization for folly::Range to POD value types.

Reviewed By: @ot, @ddrcoder

Differential Revision: D2285554

Summary: [Folly] Fix Build: FOLLY_SSE and preprocessor floats.

Float values for preprocessor symbols are not a thing.

Reviewed By: @​fyan

Differential Revision: D2289436

Summary: This diff fixes the helper template ReturnValue to enable using of applyTuple with
constant refs.

Reviewed By: @ot, @ddrcoder

Differential Revision: D2284733