Summary:
'Nuf said.

Depends on D4413112

Reviewed By: yfeldblum, spacedentist

Differential Revision: D4425605

fbshipit-source-id: a7c98867cfa5c859a1b4d2eb78724710c4f1115f

Summary: It was killed in an earlier diff.

Reviewed By: yfeldblum

Differential Revision: D4418832

fbshipit-source-id: c9a5134fa100b24a3394607dea20e6e8a8d341e1

Summary:
Pthread is currently a dependency of Folly that is not really necessary on Windows, or even with standard C++ for the most part, so start work on killing it in Folly.
This switches EventBase to using `std::thread::id`'s instead, which also means we aren't reliant on the implementation detail that thread id 0 is invalid. Well, we are, but it's now the standard library's fault not ours.

Reviewed By: yfeldblum

Differential Revision: D4418128

fbshipit-source-id: a9c95ac6c7305c960156a4ad684b6db89b5856d9

Summary: It's possible via some hackery, so wrap it up nicely into the API.

Reviewed By: yfeldblum

Differential Revision: D4418728

fbshipit-source-id: f00aed95cdbdc3cc83a0ab96565117cd01141c2e

Summary:
Our internal version of boost contains a patch that reverts the following commit to boost filesystem module:
https://github.com/boostorg/program_options/commit/88dea3c6fdea8c9ea894911897b1770599c383e4

It was done due to a breaking change that disallows the use of "--key value" for implicit options, allowing only "--key=value" style syntax. Since there are too many tests to fix, and it's probably going to be reverted in the upstream as well (see below), removing the `override` keyword here to fix folly build.

https://github.com/boostorg/program_options/issues/25

Reviewed By: meyering

Differential Revision: D4415347

fbshipit-source-id: c215331338b3aa218d26890a83aa0b363514fe79

Summary:
While running a UBSan enabled binary, I got:

  folly/io/IOBuf.cpp:671:15: runtime error: null pointer passed as argument 2, which is declared to never be null

This change protects all calls to memcpy from passing `nullptr`.

Reviewed By: pixelb

Differential Revision: D4415355

fbshipit-source-id: a27ba74244abcca8cd4e106967222890a67f5b6d

Summary:
This started when many fbvector-related tests began failing with reports of
ASAN-detected heap abuse (usually invalid free of an address 16 bytes
into an already-freed buffer).

It was very specific, though. The bug struck only when:
  - gcc-5-based platform
  - compiling with clang (~3.8) and GCC5's libstdc++, not gcc
  - strings were short enough to reside within an in-situ std::string

Why? because FBVector.h:1588's conditional:

  if (folly::IsRelocatable<T>::value && usingStdAllocator::value) {

was true for clang (erroneously), but false for gcc. The difference
was in `IsRelocatable<std::string>::value`.  However, manual tests showed that
each component from the definition of `IsRelocatable` were the same for both
gcc and clang. Then Jay Feldblum realized that we'd probably specialized
that trait for `std::string`, and sure enough, I found the culprit:

```
FOLLY_ASSUME_FBVECTOR_COMPATIBLE_3(std::basic_string)
```

Reviewed By: pixelb

Differential Revision: D4414159

fbshipit-source-id: 60e3fb4b096ec77cbd2b48c561e5c6ec8f128fff

Summary:
prefer creating an exception_wrapper with a reference to the active exception

Depends on D4410421

Reviewed By: spacedentist

Differential Revision: D4410455

fbshipit-source-id: d6b6aeb5fa72782e31d754a0b853514af5fdb8cd

Summary: Folly parts of D4383906.

Reviewed By: plapukhov

Differential Revision: D4387254

fbshipit-source-id: 3c039720c88c91b7292d60a85272dd1978510296

Summary: [Folly] Slight simplification of `exception_wrapper` constructor.

Reviewed By: ericniebler

Differential Revision: D4391899

fbshipit-source-id: ddb066723bcd10abb0dbbaeab12b1e9be4f39acc

Summary:
Normally clearing out the error before calling into an openssl method
is a bad idea. However there might be other code outside AsyncSSLSocket
calling into openssl in the same thread that doesn't use openssl
correctly. This allows users of AsyncSSLSocket to safeguard themselves
from such code.

Reviewed By: anirudhvr

Differential Revision: D4389970

fbshipit-source-id: 12da254d6b281c2b9d522ba19999b2489c0083a2

Summary:
[Folly] Fix an inefficiency in json-serializing a `dynamic` with sorted keys.

The inefficiency is that, at each level of the `dynamic` which is map-typed, we effectively make a full copy of the `dynamic` object. This can be expensive.

Reviewed By: Gownta

Differential Revision: D4389671

fbshipit-source-id: 223739397f913d3e65a421a9a4dcb089ec757ec6

Summary:
`unitialized` seems to be typo.
Closes https://github.com/facebook/folly/pull/538

Reviewed By: igorsugak

Differential Revision: D4397318

Pulled By: yfeldblum

fbshipit-source-id: a2e6c665b4dfe95d769430405886b0725423aac2

Summary:
[Folly] Rename `exception_wrapper::getExceptionPtr` to `to_exception_ptr`.

Make it clear that this will sometimes be an expensive operation. Using the word `to` evokes the possibility of an expensive conversion, while using the word `get` implies cheap access.

Reviewed By: djwatson

Differential Revision: D4391621

fbshipit-source-id: 33ca051d9be5a6050ba9f30b20faee35b7e58afb

Fix dynamic MPMCQueue tryObtainPromisedPushTicket() to prevent tryWriteUntil() and writeIfNotFull() from blocking indefinitely for a matching read.

Summary:
The bug was reported by Alexander Pronchenkov in https://fb.facebook.com/groups/560979627394613/permalink/837052843120622/

Under certain conditions a `tryWriteUntil()`--and also `writeIfNotFull()`--operation may block indefinitely awaiting a matching read. This could happen because in each dynamic MPMCQueue expansion, typically one or two tickets are associated with the closed array not the new one. In the incorrect code, a `tryWriteUntil()` operation that induced expansion but gets a ticket associated with the closed array, incorrectly assumes that because the expansion succeeded then there is space for it. However because the ticket is associated with the closed array, the operation needs to wait (possibly indefinitely) for space to open in the closed array.

The fix: Changed the code in tryObtainPromisedPushTicket() such that the operation tries to acquire a ticket only if there is promised space in the array associated with that ticket. If there is no space, an expansion is attempted if the ticket is not associated with a closed array. If not or if expansion fails because of reaching maximum capacity or for being out-of-memory, then the operation returns false without attempting to acquire the ticket.

Other changes:
- Added a note about this difference in semantic between the dynamic and non-dynamic version to the main comment about the dynamic version.
- Changed `oldCap` to `curCap` because the value is actually current not old.
- Added two tests for checking that tryWriteUntil() never blocks indefinitely for both dynamic and non-dynamic versions.
- Removed all the `never_fail` tests for the dynamic version, because such operations may fails as described above.
- Added `asm_volatile_pause` when spinning on the seqlock.

Reviewed By: djwatson

Differential Revision: D4389347

fbshipit-source-id: c46dbefc9fe08e146250d2ad8ba68b0887f97436

Summary: This was accidentally changed in D4375539.

Reviewed By: philippv

Differential Revision: D4393574

fbshipit-source-id: 0420223575bad0c6b0e89f8cf5ed138e063462b9

Summary: This allows something else (ie fizz) to read data from a socket, and then later decide to to accept an SSL connection with OpenSSL by inserting the data it read in front of future reads on the socket.

Reviewed By: anirudhvr

Differential Revision: D4325634

fbshipit-source-id: 05076d2d911fda681b9c4e5d9d3375559293ea35

Summary: Similar to the bio write method. This should be essentially a no-op, bioRead has the same functionality as the openssl version. This is needed in the next diff.

Reviewed By: siyengar

Differential Revision: D4325622

fbshipit-source-id: a90b9ec06bee920a1a73551a3ea4c77e1ee0ab08

Summary: Closes https://github.com/facebook/folly/pull/537

Reviewed By: yfeldblum

Differential Revision: D4391871

Pulled By: Orvid

fbshipit-source-id: 80401dd65ae376f9ebbb0b079577d0a1fd15475b

Summary:
This constructor is unsafe. The check it uses before doing the comparison isn't a safe enough check to see if the cast is valid. For example, this is broken in the presence of multiple inheritance because it doesn't adjust the pointer offset to the correct vtable.

e.g.
  struct A {
    virtual ~A() {};
    virtual void doSomething() = 0;
  };
  struct B {
    virtual ~B() {}
    virtual void doSomethingElse() = 0;
  };
  struct C : public B, public A {
    virtual ~C() {}
    void doSomething() override {
      std::cout << "Something!" << std::endl;
    }
    void doSomethingElse() override {
      std::cout << "Something Else!" << std::endl;
    }
  };
  int main (int argc, char **argv) {
    auto c = folly::makeFuture<std::shared_ptr<C>>(std::make_shared<C>());
    folly::Future<std::shared_ptr<A>> a = std::move(c);
    a.get()->doSomething();
    return 0;
  }
This code will print "Something else!" when run.

Reviewed By: siyengar

Differential Revision: D3679673

fbshipit-source-id: dcbf40ca82d458f17ee11191591f8b8daf58c919

Reviewed By: yfeldblum

Differential Revision: D4386258

fbshipit-source-id: 23499267eb4390f0f40b3643760514cae1ca7838

Summary:
gcc-5 -Wlogical-op reports:
```lang=bash
folly/io/async/AsyncSSLSocket.cpp: In member function 'void folly::AsyncSSLSocket::invalidState(folly::AsyncSSLSocket::HandshakeCB*)':
folly/io/async/AsyncSSLSocket.cpp:418:35: error: logical 'or' of collectively exhaustive tests is always true [-Werror=logical-op]
```

Reviewed By: yfeldblum

Differential Revision: D4391317

fbshipit-source-id: c0f5ce748f1fc21678e4080c7f6351e7fada1e2b

Summary: A one-liner from C++17, but a useful one. Const-qualifies an lvalue reference, avoiding the need for an awkward const_cast in some cases.

Reviewed By: yfeldblum

Differential Revision: D4389929

fbshipit-source-id: 1650c4901489eb0dd62fd9fa633b4a0da9f01954

Summary: It was using `std::chrono::milliseconds` to represent a point in time, so use a time point instead.

Reviewed By: yfeldblum

Differential Revision: D4378116

fbshipit-source-id: f0b10bb7894dda44d78b31672d2b6735f3e1cbf4

Summary: Because modern is good, and this allows for cleaner interaction with the other APIs being `std::chrono`ized.

Reviewed By: yfeldblum

Differential Revision: D4377773

fbshipit-source-id: 42ad5ad8476a3678e02e9314f9592f5449102f00

Summary: Onward towards more modern code!

Reviewed By: yfeldblum

Differential Revision: D4377678

fbshipit-source-id: 6ca5ecd902be9028fb55f139374f7919fa522899

Summary:
[Folly] Remove `<glog/logging.h>` from `folly/Indestructible.h`.

This makes the header much lighter. And abort on invalid access (the use-case for `glog`) only in `!defined(NDEBUG)` mode.

Anti-Pattern Combined Mega-Diff Description:
* Add non-`explicit` default ctor.
* Disable ctor explicitly when underlying ctor would fail to compile. Not sure how useful it would be, but it makes type-traits work correctly for whatever that is worth.
* Switch boolean flag to `erased_{false}` so that it is zero-initializable, because that's better.

Reviewed By: ericniebler

Differential Revision: D4380469

fbshipit-source-id: a39cb7470f7ee678fa722778da587409f468d985

Summary:
C++17 is adding std::void_t. It's handy for controlling class template partial specialization. Folly should have it too.

See http://en.cppreference.com/w/cpp/types/void_t

Reviewed By: yfeldblum

Differential Revision: D4387302

fbshipit-source-id: 85f955f3d8cfacbd6c9e61fb3f5cf53c056459bb

Summary:
[Folly] Skip `defined` checks in `folly/Memory.h`.

Undefined symbols, when evaluated by the preprocessor in preprocessor integer contexts, evaluate to `0`. That works for these checks.

Reviewed By: ericniebler

Differential Revision: D4385026

fbshipit-source-id: 3162f8b83431ebab2de53f9ba202e7ea8c339210

Summary:
__throw* functions exist in master branch but are not included in both
libc++ 3.9.0 and 3.9.1. Expect them to appear in next LLVM release
(which is 4.0).
Closes https://github.com/facebook/folly/pull/536

Reviewed By: yfeldblum, Orvid

Differential Revision: D4377002

Pulled By: smeenai

fbshipit-source-id: 5dd311ca3ec43955f29dd1197fd8fbeb9564a7f6

Summary: Modernization is good.

Reviewed By: yfeldblum

Differential Revision: D4377612

fbshipit-source-id: adb03d8a92f25c8a792c8e7240a93ab20180b038

Summary: Modernizing for the future.

Reviewed By: yfeldblum

Differential Revision: D4375539

fbshipit-source-id: e3edf8dd6b77d3338f8cf58ed128bc5bce62da32

Summary:
```lang=bash
folly/test/SmallLocksBenchmark.cpp:212:18: error: use of undeclared identifier 'sqrt'
  double stdev = sqrt(accum / (results.size() - 1));
                 ^
folly/test/SmallLocksBenchmark.cpp:221:20: error: use of undeclared identifier 'sqrt'
  double stddev2 = sqrt(variance);
                   ^
2 errors generated.
```
Add the missing header.

Reviewed By: yfeldblum

Differential Revision: D4386343

fbshipit-source-id: 79c3dcae32c62f9b6d162862deabccf5fea7eaef

Summary:
Clang with libgcc-5 reported:
```lang=bash
folly/experimental/test/BitsTest.cpp:249:16: error: use of undeclared identifier 'pow'
  auto value = pow(2, bits) * (negate ? -2.0 : 2.0) / 3.0;
               ^
```
Add missing include statement, and elaborate with the namespace.

Reviewed By: yfeldblum

Differential Revision: D4385807

fbshipit-source-id: a847dd439cd4c9f28ea8f222aa4ab60876949d13

Summary: Missing compatibility case for Android

Reviewed By: yfeldblum

Differential Revision: D4384944

fbshipit-source-id: 0b3e43fb284b6a71c6ae2a2162d1ea7cb99bc817

Summary: Critical tests in benchmarks should not be in asserts since they will be stripped out in release mode, which is the only mode interesting for benchmarks.

Reviewed By: yfeldblum

Differential Revision: D4384411

fbshipit-source-id: 5237ab48f99ddcd2bce7a159fcf82900303efec2

Summary:
[Folly] Namespace the `UBSAN_DISABLE` symbol.

Symbols defined by Folly should typically be namespaced to Folly.

Reviewed By: igorsugak

Differential Revision: D4378401

fbshipit-source-id: ff72f5f44e7c9d1bbf08efcec24bd3dc05d10ee7

Summary:
[Folly] Backport `ceil`, `floor`, `round` from C++17 `std::chrono`.

These functions for operating on `std::chrono::duration` and `std::chrono::time_point` values are useful

Note: The implementations are derived from cppreference.com, which has a notice listing all content there as licensed under CC-BY-SA.

Reviewed By: ericniebler

Differential Revision: D4375603

fbshipit-source-id: 36a098bf5f75db071c1670518fc42bbc6df2817d

Summary:
[Folly] Keep the `std::exception*` or `std::type_info*` in `folly::exception_wrapper`.

In the case of construction with a `std::exception_ptr`. Instead of keeping a pair of `std::string`s, strictly losing information (not that we use any more information, but theoretically we could).

Of note:
* Shrinks the size of `folly::exception_wrapper`, in all cases, to 48 bytes. Down from 32 bytes + 2 S bytes on 64-bit architectures, where S is the size in bytes of `std::string`. In particular, `libstdc++`'s implementation is 32 bytes, while `libc++`'s and Folly's implementations are 24 bytes on 64-bit architectures, for total original sizes of 96 bytes or 80 bytes.
* Allows to avoid rethrowing in `with_exception` in the case of an instance constructed with an `std::exception_ptr` plus an `std::exception&`.

Reviewed By: ericniebler

Differential Revision: D4369935

fbshipit-source-id: 35155e0b271959a4878fe077fc911b17767a2358

Summary:
libc++ 4.0 and above define the throw* functions in `stdexcept`, so
Folly doesn't need to define them itself. In case those definitions are
backported to 3.9, add a Folly macro to skip the definitions regardless
of the libc++ version number.

Reviewed By: yfeldblum, Orvid

Differential Revision: D4378102

fbshipit-source-id: 674d98d7c448aa7e75bdd8b6e9fe965cd34a83d7