Summary:
- Change folly::coro::co_viaIfAsync to be a customization-point object.
- Modify co_viaIfAsync() to call a .viaIfAsync() member function if one exists.
  Otherwise it calls the co_viaIfAsync() function found by ADL.
- Add FOLLY_DEFINE_CPO() helper macro to folly/Portability.h
  Borrows some techniques from range-v3 to support CPOs in different
  compilers and C++ versions.
- Redefined co_viaIfAsync() CPO in terms of this macro.
  This allows it to be callable as folly::coro::co_viaIfAsync() while
  still allowing other types in folly::coro namespace to define friend
  functions that customise the behaviour of this customisation point
  without causing conflicts.
- Add folly::coro::is_semi_awaitable<T> metafunction.
- Add folly::coro::semi_await_result_t<T> metafunction.

Reviewed By: yfeldblum

Differential Revision: D14016389

fbshipit-source-id: 63998993b2fa8560700f81666a910dee31f3ba72

Summary:
The behavior of fiber Baton is currently inconsistent around posting of a timed out baton, and subsequent calls to try_wait().  As stated in the post, only when the waiter is a thread will a post overwrite a timeout internally, and this is observable externally because try_wait returns true only if a baton is in the POSTED state, but not the timed out state.  Example code:

```
folly::fibers::Baton b;

b.try_wait_for(std::chrono::milliseconds(1));
b.try_wait(); // returns false

b.post();
b.try_wait(); // Returns true on threads, false on fibers
```

Other options we could consider for the fix:
 * Update postThread to leave baton in a timeout state, and have waitThread move the baton into the timeout state on a failed wait.
 * Update try_wait() to return true if the state is either POSTED or TIMEOUT.

Reviewed By: andriigrynenko

Differential Revision: D14324532

fbshipit-source-id: 51b993a6153cba0fc3cf5701f765fff8ad1c8e3c

Summary:
Some codecs may not want counters enabled. Right now that will
`std::terminate`. Allow disabled counters, and add a test.

Reviewed By: felixhandte

Differential Revision: D14327170

fbshipit-source-id: 37858518c80e192e351e8578e088b59732d2a0cb

Summary:
- Some constant expressions are wrapped in a type carrying a bool, but
  are only used in value contexts. Remove the type wrapper
  (bool_constant) use and just use `constexpr bool` directly. This makes the
  usages not have to use `::value` everywhere.
Pull Request resolved: https://github.com/facebook/folly/pull/1039

Reviewed By: Orvid

Differential Revision: D14293828

Pulled By: yfeldblum

fbshipit-source-id: 78399e4fb401ab680a4f1ecef63fb4c2a82927c2

Summary: [Folly] Remove extraneous uses of keyword `typename` in fibers `collectN`.

Reviewed By: ericniebler

Differential Revision: D14317102

fbshipit-source-id: 1ae04081258bf2398865a1cc5461618648623158

Summary: [Folly] Canonical SFINAE in fibers, where the `enable_if` is on the outside.

Reviewed By: andriigrynenko

Differential Revision: D14294993

fbshipit-source-id: 498c9bfcd346edb5e4b5b65941e2a09c93e870b6

Summary:
- On some platforms, `CMAKE_SYSTEM_ARCHITECTURE` may resolve to an empty
  string.
- When this value is used without quotes to the `string` command, it is
  ill-formed with error:

```
  CMake Error at CMakeLists.txt:174 (string):
  string sub-command FIND requires 3 or 4 parameters.
```

- Explicitly wrap the value to the `string` operation in quotes to treat
  `CMAKE_SYSTEM_ARCHITECTURE` as a string, which is allowed to be empty.
  The result is that the `string` operation will not result in a hard
  error.
Pull Request resolved: https://github.com/facebook/folly/pull/1040

Reviewed By: calebmarchent

Differential Revision: D14293784

Pulled By: yfeldblum

fbshipit-source-id: cd5924fa62277aa07e930c7b088197ef596be977

Summary: The property_set_insert_t<PS1, PS2> template metafunction had a bug that meant that it would fail to replace elements in the base set that had the same category as elements from the set being inserted.

Reviewed By: ericniebler, kirkshoop

Differential Revision: D14268145

fbshipit-source-id: ccc64e455bc6ff2073229e028a19d64dc7050092

Summary:
[Folly] Mark `BitIterator<BaseIterator>` with the iterator category of `BitIterator`.

As one benefit, using `std::distance` to find the distance between two `BitIterator`s gotten from a random access container such as `std::vector` is now constant-time.

Fixes #1026.

Reviewed By: ot

Differential Revision: D14294852

fbshipit-source-id: 2345bc73dec169803ae41b9391687e89ad77207b

Summary: [Folly] Support C++11 in `folly/IPAddress.h` w.r.t. `aligned_union` by avoiding `aligned_union_t`, which is C++14.

Reviewed By: spalamarchuk, mengz0

Differential Revision: D14278604

fbshipit-source-id: 474ac0aa0110f4939d1042d334b69feb6b2644cb

Differential Revision:
D14114164

Original commit changeset: e227609e9214

fbshipit-source-id: 99d0fde58e512224915d7a00893ddf15c547dfba

Summary: [Folly] Remove include of `glog` from `Range.h`.

Reviewed By: lbrandy, Orvid

Differential Revision: D14114164

fbshipit-source-id: e227609e9214ab39ff272e44519d34f3047fe025

Summary: The reinit() method was removed but some comments mentioning it remained. Remove them.

Reviewed By: yfeldblum

Differential Revision: D14277614

fbshipit-source-id: 389ef3e2afe1fdc60f37aaec2f8a48dfa0b1436c

Summary:
[Folly] Make `EventBase` enqueue `noexcept`.

It cannot really fail anyway in correct usage besides allocation failure, unless in the `EventBase` destructor and while draining and the `AlwaysEnqueue` variant is called.

Theoretically if a caller attempts to enqueue concurrently with `EventBase` dtor while in `consumeUntilDrained`, but either *not* in the `EventBase` thread or in the `EventBase` thread and using the `AlwaysEnqueue` variant, there is a race which can lead to termination.

Reviewed By: andriigrynenko

Differential Revision: D14114678

fbshipit-source-id: 9a0128d207f86ca34eb8a1d417766c095ed5e137

Summary:
The `TypeError` special member functions - copy and move constructors and assignment operators - were outlined to reduce inline code size of `throw TypeError(...);` statements. This is no longer necessary, since these sites have been replaced by `folly::throw_exception<TypeError>(...);` statements, which have minimal inline code size. As a benefit, the special member functions may all now be declared as defaulted within the class body, and the conditional `noexcept` calculations are no longer required to be written - the compiler will do them. (Note: as of the libstdc++ which ships with GCC 5, library exception special member functions are `noexcept`; previously, they are not.)

Pull Request resolved: https://github.com/facebook/folly/pull/1034

Reviewed By: nbronson

Differential Revision: D14255166

Pulled By: yfeldblum

fbshipit-source-id: 2f795a2b937fee58f243a9d374fc01829c674fe6

Summary: [Folly] Avoid explicit `memset` in `IPAddress` storage union by having an explicit storage field and value-initializing it in the default constructor.

Reviewed By: stevegury

Differential Revision: D14270246

fbshipit-source-id: f1f135f20709d8225e984ee3d6b397d4b04e0d76

Summary:
Added LtHash, a cryptographic homomorphic hash, to folly/experimental/crypto.
This has a soft dependency on libsodium and the code will not be compiled if libsodium is not detected by cmake.

Reviewed By: djwatson

Differential Revision: D13390825

fbshipit-source-id: f7597ced7bcc7b403e8bbaa733837b795128f1b3

Summary: struct crypto_generichash_blake2b_state was made opaque in libsodium 1.0.17. This change is needed to make blake2xb code compile.

Reviewed By: djwatson

Differential Revision: D14141454

fbshipit-source-id: eee887e8300229ff568d12ef19b46472edfc56d2

Summary: Implemented the Blake2xb XOF. See the specification at https://blake2.net/blake2x.pdf.

Reviewed By: djwatson

Differential Revision: D13403920

fbshipit-source-id: 38afd6d3b782c2f6649365d01ca3d83b55c90c4b

Summary:
This was treating a socket as a file descriptor, so make it explicitly be a file descriptor instead.
The full fix for this requires a *lot* of refactoring and splitting of a lot of concepts within Folly, so this short-term fix should be good enough for now.

Reviewed By: boguscoder

Differential Revision: D14266198

fbshipit-source-id: bafb3a403ab8555f865f4541323c5d0ad81f5052

Summary: Original commit changeset: 1e15d312d1a8

Reviewed By: Orvid

Differential Revision: D14265880

fbshipit-source-id: 8c2ad3e15334015dadc19bf77b061f66086adcd6

Summary: It's dead.

Reviewed By: yfeldblum

Differential Revision: D14192000

fbshipit-source-id: 3cecbc8e9f656080a023d676883de9943e7d0004

Shift calls of the file descriptor overload of AsyncSocket::AsyncSocket() to the NetworkSocket overload

Summary: The file descriptor overload will be going away.

Reviewed By: yfeldblum

Differential Revision: D13998628

fbshipit-source-id: 7a1d3bae03d4742d874e1e7d733f7733b5a1c110

Summary: Deprecate sleepUnsafe to avoid use growing and to clearly communicate the change.

Reviewed By: yfeldblum

Differential Revision: D14255404

fbshipit-source-id: cb2116910a15b9835fef9bf4273242457a5e96bd

Summary:
InlineFunctionRef is a semantically the same as folly::FunctionRef but has the
additional benefit of being able to store the function it was instantiated
with inline in a buffer of the given capacity.  If there is not enough in-situ
capacity for the callable, this has the same semantics as FunctionRef.

This helps give a perf boost in the case where the data gets separated from
the point of invocation.  If, for example, at the point of invocation, the
InlineFunctionRef object is not cached, a remote memory/cache read might be
required to invoke the original callable.  Customizable inline storage helps
tune storage so we can store a type-erased callable with better performance
and locality.  A real-life example of this might be a folly::FunctionRef with
a function pointer.  The folly::FunctionRef would point to the function
pointer object in a remote location.  This causes a double-indirection at the
point of invocation, and if that memory is dirty, or not cached, it would
cause additional cache misses.  On the other hand with InlineFunctionRef,
inline storage would store the value of the function pointer, avoiding the
need to do a remote lookup to fetch the value of the function pointer.

To prevent misuse, InlineFunctionRef disallows construction from an lvalue
callable.  This is to prevent usage where a user relies on the callable's
state after invocation through InlineFunctionRef.  This has the potential to
copy the callable into inline storage when the callable is small, so we might
not use the same function when invoking, but rather a copy of it.

Also note that InlineFunctionRef will always invoke the const qualified
version of the call operator for any callable that is passed.  Regardless of
whether it has a non-const version.  This is done to enforce the logical
constraint of function state being immutable.

This class is always trivially-copyable (and therefore
trivially-destructible), making it suitable for use in a union without
requiring manual destruction.

Reviewed By: yfeldblum, ot

Differential Revision: D14029799

fbshipit-source-id: 2cff3ce27d564f3d524095189f847c14911f9402

Summary:
Don't assume a jemalloc target exists (even if it would be a no-op, as
in on macOS). Instead compute the correct set of external_deps in
jemalloc.bzl.

This fixes the folly build on mode/mac after D14179337.

Reviewed By: andrewjcg

Differential Revision: D14235815

fbshipit-source-id: d1bfbb91897de9557524bbb4b5736ca7b03cbf9b

Summary:
- If a type is not marked explicitly with `IsRelocatable`
  as `std::true_type`, we infer its relocability only based on the
  property of whether the type is trivially copyable.
- Extend the default relocability to also be true for trivially move
  constructible types.
Pull Request resolved: https://github.com/facebook/folly/pull/1035

Reviewed By: Orvid

Differential Revision: D14240127

Pulled By: yfeldblum

fbshipit-source-id: 1e15d312d1a8340417bba2beb1db30ce4c543b26

Summary: It is dead

Reviewed By: yfeldblum

Differential Revision: D14235873

fbshipit-source-id: bafc323898f2a17aedf6644f9c2dd7319b8e6420

Summary: Merge two mutexes together and get rid of atomics.

Reviewed By: yfeldblum

Differential Revision: D14236244

fbshipit-source-id: 0b97d519e72322377e225a0cc8975e3a2a86ec31

Summary: I missed one of the functions that needs a NetworkSocket overload in past iterations.

Reviewed By: yfeldblum

Differential Revision: D14237211

fbshipit-source-id: 6d0bd151ff636206a44940c88a2d48c339aa67de

Summary:
Make this metafunction available for use outside of the
folly::coro::blockingWait() implementation.

I plan to reuse this metafunction for other operators such as `when_all()`.

Reviewed By: yfeldblum

Differential Revision: D13778687

fbshipit-source-id: f766f93da21bf19e89262cd60048b0b7b1b76940

Summary:
- `folly/experimental/Bits.h` has a section where we disable two
  uninitialized-flavors of warnings that were needed for GCC 4.8
- Since we do not support GCC 4.8 anymore and GCC 5.1 does not have
  issues with this, do not disable the warnings anymore.
Pull Request resolved: https://github.com/facebook/folly/pull/1033

Reviewed By: Orvid

Differential Revision: D14223285

Pulled By: yfeldblum

fbshipit-source-id: 0088341b13e26dff2dc90a768319cccc45e942da

Summary: [Folly] `DefaultKeepAliveExecutor::reset` in case of restartable executors.

Reviewed By: andriigrynenko

Differential Revision: D14236173

fbshipit-source-id: 14f71af6bd54777770fdfe6036a4137af4e884bc

Summary: [Folly] Cut deducible template arg in `DefaultKeepAliveExecutor` init - let it be deduced instead.

Reviewed By: andriigrynenko

Differential Revision: D14236171

fbshipit-source-id: 2de043efe541c13c59764715516b08d77efcb99b

Summary:
If multiple leaves are updated at the same time (or a new observer is created concurrently that depends on the updated leaf) it's possible for refresh to be called concurrently with force=false which will not update the leaf value, but also ignore the refresh with force=true.
This is fixed by making force to be the property of the leaf observer itself, which is set when the version is bumped.

Reviewed By: yfeldblum

Differential Revision: D14222400

fbshipit-source-id: 657c2c273002f576dbe48a232eaef2d530cea07d

Summary:
Angled brackets inside angled brackets is not displayed.
This makes the document misleading.
Pull Request resolved: https://github.com/facebook/folly/pull/1032

Reviewed By: shixiao

Differential Revision: D14207772

Pulled By: yfeldblum

fbshipit-source-id: 4f2e58145a5473a7b887ef1da4efbeb16d5330c6

Summary:
futures::sleep returning a Future leads to continuations easily being run on
the Timekeeper's callback. The goal is to change the return type so that
futures::sleep returns a folly::SemiFuture.

Reviewed By: yfeldblum

Differential Revision: D14069549

fbshipit-source-id: b98e2c76ccbf0a80aed28f90a1f3c63f15eb2e5c

Summary:
- Existing code disables `-Warray-bounds` warning for GCC due to a bug in
  GCC 4.9.
- As seen in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59124, the bug
  has been resolved for GCC 5.1.
- Since the bug has been resolved, we no longer need to disable the
  warning.

yfeldblum I suspect that we have to hold off on this PR until the internal use cases relying on GCC 4.9 upgrade, but I figured it is better to have some of these cleanup opportunities ready for when the time arises.
Pull Request resolved: https://github.com/facebook/folly/pull/999

Reviewed By: ot, Orvid

Differential Revision: D13729723

Pulled By: yfeldblum

fbshipit-source-id: da33704c2e2022c2dad0681e4d6152105a1cf54d

Summary:
[Folly] Address MSVC C4686 in `folly/lang/Pretty.h`.

Resolves:

```
folly\lang\pretty.h(119): warning C4686: 'folly::detail::pretty_name_zarray<T>::zarray_': possible change in behavior, change in UDT return calling convention
```

Reviewed By: Orvid

Differential Revision: D14214290

fbshipit-source-id: c5e0ebd7c100413737bdcba157ec8b5ce42895de

Summary:
Infer AL has reported valid warnings on some folly and thrift code:

```
fbcode/folly/io/async/AsyncUDPServerSocket.h:228:19:
WARNING Extra Copy: Potentially unnecessary to copy var `client` at line 228, column 19. Use 'const auto&' or 'auto&' if possible.
fbcode/folly/io/async/AsyncUDPServerSocket.h:230:19:
WARNING Extra Copy: Potentially unnecessary to copy var `socket` at line 230, column 19. Use 'const auto&' or 'auto&' if possible.
fbcode/thrift/lib/cpp/TProcessorEventHandler.h:148:15:
WARNING Extra Copy: Potentially unnecessary to copy var `ew` at line 148, column 15. Use 'const auto&' or 'auto&' if possible.
fbcode/thrift/lib/cpp2/server/ThriftServer.h:705:20:
WARNING Extra Copy: Potentially unnecessary to copy var `sockets` at line 705, column 20. Use 'const auto&' or 'auto&' if possible.
```

As these are transitively included in many services, they are fairly prevalent/annoying warnings. Let's fix them.

Reviewed By: yfeldblum

Differential Revision: D14168260

fbshipit-source-id: 12e537fe5aa73e653c15c4fef01ac618e51cd4d0