Summary:
The current predicate yields false positive for NDK r17. NDK r18 has launder without defining `__cpp_lib_launder`. NDK r19 finally gets it right by defining `__cpp_lib_launder`.

r17: launder? no,  `__cpp_lib_launder` undefined, `_LIBCPP_VERSION == 6000`
r18: launder? yes, `__cpp_lib_launder` undefined, `_LIBCPP_VERSION == 7000`
r19: launder? yes, `__cpp_lib_launder == 201606L`, `_LIBCPP_VERSION == 8000`

Reviewed By: yfeldblum

Differential Revision: D14098189

fbshipit-source-id: 315dba44c6474407a8f4c1d25c81fff057d5f7db

Summary:
[Folly] Use `StaticSingletonManager` for `LoggerDB::get()` inline fast path.

This also simplifies the singleton setup overall.

Note that the inline slow path here is not quite as small as doing the strategy directly would be since three pointers need to be passed to the outline slow path.

Reviewed By: simpkins

Differential Revision: D14016983

fbshipit-source-id: ea2646f8efd5c2dbcdece4ebd3b88949bc3d4294

Reviewed By: yfeldblum

Differential Revision: D14109224

fbshipit-source-id: 4bcd40b8692478540f5b5b4ee2751f1d311b3ac5

Summary:
- AtomicHashMap.h has an unneeded include for
  `<boost/type_traits/is_convertible.h`.
- Remove the include and add `<type_traits>` in `AtomicHashMap-inl.h`
  since it uses `std::is_convertible`.
Pull Request resolved: https://github.com/facebook/folly/pull/1018

Reviewed By: Orvid

Differential Revision: D14078944

Pulled By: yfeldblum

fbshipit-source-id: f1450166a5bedb279db71e9b4f915811c71e5f1a

Reviewed By: yfeldblum

Differential Revision: D13603520

fbshipit-source-id: 08365b3723c024697a546223683e774f18b61a94

Summary:
Use C++11’s override and remove virtual where applicable.
Change are automatically generated.

Reviewed By: shixiao

Differential Revision: D14089910

fbshipit-source-id: 12c776ea06094ee77c7e07263397dab3419c5309

Summary:
On ARM gcc, the unwind header defines exception_class as a char[8] rather
than a uint64_t.  Consequently compilation of exception_tracer fails on ARM
with the following error:

    .../folly/experimental/exception_tracer/ExceptionTracer.cpp: In function 'bool folly::exception_tracer::{anonymous}::isAbiCppException(const __cxxabiv1::__cxa_exception*)':
    .../folly/experimental/exception_tracer/ExceptionTracer.cpp:107:45: error: invalid operands of types 'const char [8]' and 'unsigned int' to binary 'operator&'
        return (exc->unwindHeader.exception_class & 0xffffffff) == cppClass;
                ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~

Construct a uint64_t for this platform.

Reviewed By: yfeldblum

Differential Revision: D13763587

fbshipit-source-id: 011cf13cdfcd1fffdeab8c384f7274f20faecbe5

Summary: Wrap Executor in ExecutorWithPriority if custom priority is passed.

Reviewed By: yfeldblum

Differential Revision: D14079222

fbshipit-source-id: 56917827f54115b90ca237c71321f7db327355e1

Reviewed By: yfeldblum

Differential Revision: D14062156

fbshipit-source-id: 2cdc32fc9834bd165fa8011ca1d336edaf05f653

Summary:
delayedUnsafe was deprecated.

All callers have been removed. This removes the core code so that it will not sneak back in.

Reviewed By: yfeldblum

Differential Revision: D14029673

fbshipit-source-id: 0384364a8e0ab8953e9746b91e3b6e7b2e5d7c67

Summary: [Folly] A test verifying `Function` behaves as expected with max-align callables.

Reviewed By: aary

Differential Revision: D14077496

fbshipit-source-id: 8a9efaee3cc4c5d69c6a94eb5ad1928d241eb4e0

Summary: Clone sleep as sleepUnsafe as a codemod target to migrate callsites to a SemiFuture-returning sleep.

Reviewed By: yfeldblum

Differential Revision: D14072020

fbshipit-source-id: 3dc88378732ca203c168ea2e4371508ee376cc43

Summary: SemiFuture implementation assumes that it's always safe to read Executor (if Executor is DeferredExecutor - then SemiFuture can never be completed until via is called, otherwise Executor is nullptr and we should never overwrite it). doCallback() however was unconditionally overwritting Executor, which resulted in TSAN reports.

Reviewed By: yfeldblum

Differential Revision: D14063047

fbshipit-source-id: 7553c8a2d0cbda05ad07b87297a334a69ee77f9c

Summary:
range.pxd needs to be exported; so that fbtrhift-py3 can compile against it
Pull Request resolved: https://github.com/facebook/folly/pull/1013

Reviewed By: yfeldblum

Differential Revision: D13987018

Pulled By: calebmarchent

fbshipit-source-id: 7b7e6f62e5bc23e70a3c263fb3b33cf1d439e439

Summary: [Folly] Remove disabled code from `FOR_EACH_RANGE`.

Reviewed By: aary

Differential Revision: D14062987

fbshipit-source-id: 6d01e76debc46398ceae3be435691c20595d393e

Summary:
[Folly] `pretty_name`, for getting static type pretty-names without RTTI.

Available as constant expressions - no runtime overhead.

Depends on per-platform non-specified magic symbols. May not work everywhere.

Reviewed By: aary

Differential Revision: D14019723

fbshipit-source-id: e50557261024298cd37957b94dad3a10c927ee97

Summary: [Folly] Cut macro `FOR_EACH_RANGE_R`.

Reviewed By: aary, danobi

Differential Revision: D14058731

fbshipit-source-id: d31c0673785d467665598e2138b6b6082d9629ab

Summary:
- Use `std::is_trivially_destructible` instead of
  `boost::has_trivial_destructor` in a few places.
Pull Request resolved: https://github.com/facebook/folly/pull/1003

Reviewed By: lbrandy

Differential Revision: D13754694

Pulled By: yfeldblum

fbshipit-source-id: b32dbc901fd7c9cae681ac3436cc71ff075f30ec

Summary: The file descriptor overload will be going away.

Reviewed By: yfeldblum

Differential Revision: D13628834

fbshipit-source-id: bc81ffdf011113479c95c1e8ae7deb4d0ff91e41

Summary: [Folly] `StaticSingletonManager` without RTTI as a variant, but without support for dynamic loading either.

Reviewed By: andriigrynenko

Differential Revision: D14018865

fbshipit-source-id: d4eadc8924e6fec2cbb1dae957e311d80e9ce289

Summary: [Folly] Fix StaticSingletonManager inline fast path under gcc - code size for default-initialized `static std::atomic<...>` local is large, but when value-initialized is trivial.

Reviewed By: andriigrynenko

Differential Revision: D14018861

fbshipit-source-id: e767ea3e46c3e3317f7b60008bc77f64edb2dea0

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/1014

Reviewed By: lbrandy

Differential Revision: D14019229

Pulled By: yfeldblum

fbshipit-source-id: dd07d567a7f176a3bb0e18f7a19c3babfe2d431d

Summary: Allows erase without a re-hash into index_ when we already have an iterator

Reviewed By: aary

Differential Revision: D13928691

fbshipit-source-id: d662dc72cd8a07cdbd874cb6a3517f68ef320d80

Summary: Right now we invoke ValueType ctor (e.g. `ValueType(bucket.count)`) to calculate range adjust for long values, but ValueType can be any type and it should not have a ctro which accepts long

Reviewed By: yfeldblum

Differential Revision: D13986728

fbshipit-source-id: e5b8d826d14fb4c889047b97d88df35654119849

Summary: [Folly] Always define `FOLLY_HAS_RTTI` to either `0` or `1`.

Reviewed By: LeeHowes

Differential Revision: D14018830

fbshipit-source-id: 5212f3f6269067fc6326e293cedb243e650f71db

Summary:
Currently udp socket send packet to listener in a round robin fashion. This will cause issue if there are more than one packet in one "session". Because packets from one session could be routed to two different listeners.

This diff added a packet dispatch option which use clientAddress (address family + ip + port) hashing. In this way, the packets from same client will always routed to same listener when the total number of listener is consistent.

Reviewed By: yfeldblum

Differential Revision: D13994665

fbshipit-source-id: e450c2fcbc95c55cb37cb5cda13c5df6d8119812

Summary: [Folly] Make Init non-moveable to break code that could trigger early invocation of the moved-from instance's destructor when copy elision is not applied.

Reviewed By: markisaa

Differential Revision: D14002508

fbshipit-source-id: faf8de587c0573d14d9b1082b7f531e147c48ed7

Summary: [Folly] Detect libc++ std version in `folly::chrono` backports, since libc++ does not define the recommended feature macros.

Reviewed By: wez

Differential Revision: D14002296

fbshipit-source-id: 022085658aee3a453c5aea714f75f27013825c6f

Summary: It's dead now.

Reviewed By: yfeldblum

Differential Revision: D13629451

fbshipit-source-id: f4a79919ba20a13b6afd59b5e4605bf8faa3c6ad

Summary: The file-descriptor backed API will be going away.

Reviewed By: yfeldblum, ryantimwilson

Differential Revision: D13603038

fbshipit-source-id: d89d3a3836ab551ec92c2fb1f74248cc2119ba03

Summary: Add insert method to mimic std::unordered_map::insert. Allows for insertion without overwriting an existing element.

Reviewed By: yfeldblum

Differential Revision: D13928583

fbshipit-source-id: 54f0eb5069b753bcaca55b04c09968d12dceb5e3

Reviewed By: lewissbaker

Differential Revision: D14002394

fbshipit-source-id: 8f0422d66c6b6859ec42a6ee5ea58c97e9a15de9

Summary:
This manifested once D13765262 landed, but is likely
more closely related to the changes in D4375603:

```
In file included from /Users/facebook/wez-fbsource/fbcode/watchman/external/folly/folly/io/async/TimeoutManager.cpp:21:
/Users/facebook/wez-fbsource/fbcode/watchman/external/folly/folly/Chrono.h:153:32: error: call to 'floor' is ambiguous
  return detail::round_impl(d, floor<To>(d));
                               ^~~~~~~~~
/Users/facebook/wez-fbsource/fbcode/watchman/external/folly/folly/io/async/TimeoutManager.cpp:77:44: note: in instantiation of function template specialization 'folly::chrono::round<std::__1::chrono::duration<long
      long, std::__1::ratio<1, 1000> >, long long, std::__1::ratio<1, 1000000>, void>' requested here
  timeout_type timeout_ms = folly::chrono::round<timeout_type>(timeout);
                                           ^
/usr/local/fbprojects/packages/xcode/104/xcode_10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/include/c++/v1/chrono:444:1: note: candidate function [with _ToDuration =
      std::__1::chrono::duration<long long, std::__1::ratio<1, 1000> >, _Rep = long long, _Period = std::__1::ratio<1, 1000000>]
floor(const duration<_Rep, _Period>& __d)
^
/Users/facebook/wez-fbsource/fbcode/watchman/external/folly/folly/Chrono.h:127:14: note: candidate function [with To = std::__1::chrono::duration<long long, std::__1::ratio<1, 1000> >, Rep = long long, Period =
      std::__1::ratio<1, 1000000>, $3 = void]
constexpr To floor(std::chrono::duration<Rep, Period> const& d) {
             ^
```

this diff resolves the ambiguity in favor of `folly::chrono::floor`.

Reviewed By: bolinfest

Differential Revision: D14000897

fbshipit-source-id: 669dbf3573c3799cb21e14f247acc8d053afe746

Summary:
Move `FatalHelper.cpp` to the new folly/logging/test/helpers directory created
in D13984692.

Reviewed By: yfeldblum, dmaone

Differential Revision: D13994517

fbshipit-source-id: e2b7e4b4f649fed5b882df47424d24e4fc4f717a

Summary:
Fix a problem where `LoggerDB::get()` could crash when called after `main()`
has returned.

The `LoggerDBSingleton` object may have been destroyed already, so calling
`LoggerDBSingleton::getDB()` was not allowed.  This updates the code to just
store the singleton in a simple raw pointer, and only use the singleton helper
object for flushing log handlers.

Reviewed By: dmaone

Differential Revision: D13984692

fbshipit-source-id: a0c8550af367458ca39fefa9090e3165ad6a82bb

Summary: Judging from the number of things that needed fixing, pushmi has been busted on gcc-5 for some time.

Reviewed By: yfeldblum, kirkshoop

Differential Revision: D13979146

fbshipit-source-id: 95def0ef6289b41e116a55359476ffec91787706

Summary: It's not needed

Reviewed By: yfeldblum

Differential Revision: D13993615

fbshipit-source-id: 7e53224d39502b96e13bff352dab778df4782a1c

Summary:
[Folly] Fix `SingletonRelaxedCounter` in `T::~T()` of `ThreadLocal<T>`.

Given a type `T` which has a dtor which increments some `SingletonRelaxedCounter`, and given some use of a `ThreadLocal<T>` where, within some given thread, the only time `SingletonRelaexCounter` is incremented is in the `T` dtor, there would be a leak of the `LocalLifetime` and this leak would cause subsequent `SingletonRelaxedCounter::count()` to segfault with access to a deallocated local counter.

C++ `thread_local` destructors run before `pthread` thread-specific destructors. If a `pthread` thread-specific destructor triggers initialization of a C++ `thread_local`, then that `thread_local` will be leaked in that its destructor will never be run.

Fix this with `ThreadLocal` by detecting when `ThreadLocal` per-thread destructors have begun running. This is a larger issue with `pthread` thread-specific objects overall, but as long as we stick to using `ThreadLocal`, it can be papered over.

Reviewed By: davidtgoldblatt

Differential Revision: D13970469

fbshipit-source-id: a4c7f36d2c0d63f8f0e363ddb9d35a44b027aea6

Summary: Executor::KeepAlive<> is now implicitly constructible from Executor*.

Reviewed By: capickett

Differential Revision: D13954744

fbshipit-source-id: 523a860a337429a995152dc7171c0f6e8be8cecb

Summary: Deprecate delayedUnsafe to stop people adding it to code.

Reviewed By: yfeldblum

Differential Revision: D13949215

fbshipit-source-id: ffc4eec113964a0b247f16f3c37f3f8c69fbb0df