Summary: [Folly] Wait on the specified child in the rcu fork test, v.s. waiting on any child. Makes the test more exact.

Differential Revision: D14869269

fbshipit-source-id: 2bb0110e5097f1dfe21602a09893a33e92421555

Summary:
As an alternative to backend to BucketTable, introduce SIMDTable, which mimics the use of SSE intrinsics to filter tags as found in F14 code.

On synthetic benchmarks, SIMDTable outperforms BucketTable by between 1.1x and 2.6x operations per second when the map does not fit in cache. However, when the map fits in cache SIMDTable executes around 10% fewer operations per second compared with BucketTable.

BucketTable, the existing backend, remains the default.

Reviewed By: djwatson

Differential Revision: D14458269

fbshipit-source-id: 5b6b01db5eb2430bdfc6f3500458f25971a6ad3d

Summary: [Folly] Handle Clang on Windows in `pretty_name`.

Reviewed By: Orvid

Differential Revision: D14844690

fbshipit-source-id: 1f995604d5be0cd571caa344e5b18f877188390d

Summary:
When building without Microsoft Visual C++, `do_nothing` is not used
at all, potentially triggering unused function warnings, which makes it useless
to compile with strict flags such as `-Wall -Werror`.

Reviewed By: yfeldblum, aary

Differential Revision: D14843227

fbshipit-source-id: 28f0041ccae8bf07dc1f3e41e1cf67291c3d8447

Summary:
TLDR; The element type of a utf8 string literal was changed from a `char` to a `char8_t`, which are incompatible. There is a paper that aims to resolve some of these issues, but the next version of MSVC has already implemented p0482, so we need to add a workaround for now.

Fixes: https://github.com/facebook/folly/issues/1102

Reviewed By: yfeldblum

Differential Revision: D14856705

fbshipit-source-id: 983548cb0cb210341ec55d9dcdc63e726bfe4f62

Summary: `collect` short-circuits on an exception, so we need to wait for all the callbacks before deallocating the captured state

Reviewed By: yfeldblum

Differential Revision: D14818841

fbshipit-source-id: e7af2965b7641feff1e4230a6a66c7860994f826

Summary:
F14Map::erase(it) returns ++it for compatibility, but it
is often discarded. erase(const_iterator) was marked always-inline to
enable the computation of the return value to be optimized away, but the
annotation was missing from erase(iterator).  This also adds always-inline
annotations to a couple more methods that participate in the optimization,
but they are very good candidates for inlining regardless.

Reviewed By: yfeldblum

Differential Revision: D14854278

fbshipit-source-id: ad686b2dc8996b6a7c7038653a7c9ea1567e554d

Summary: Appear to be leftover from D8683752

Reviewed By: yfeldblum

Differential Revision: D14854252

fbshipit-source-id: bef87438201f9292fb0e5c62222d1a248163eda2

Summary: Reusable mask/iter code moves into folly/container/detail/F14Mask.h. This is in preparation for reuse by SIMDTable in ConcurrentHashMap.

Reviewed By: yfeldblum

Differential Revision: D14729170

fbshipit-source-id: fbf5e10a13eeb5e964f52d4fe262f9e5ae2e8f73

Summary:
This sets the stage for D14458269, which adds a 2nd possible ImplT class.

Move code shared between ImplT's into shared area.

Work done in *_internal functions and functions that access member fields other than `batch_` now goes into this impl_ member.

ConcurrentHashMapSegment basically is now just a wrapper that knows how to translate things like emplace, try_emplace, assign, etc. to various flavors of insert_internal. Changes to erase control flow are analogous.

Reviewed By: davidtgoldblatt

Differential Revision: D14458241

fbshipit-source-id: 41d103141ead1eb306b6fe863932f00f827f68c3

Summary:
- `Replaceable` defines a deduction guide if the feature test macro for
  it is met, or if we are on MSVC. However, the MSVC version that folly
  supports also defines the feature test macro for deduction guides. As
  such, remove the extra check for MSVC.
- Add `ReplaceableTest` to the  cmake build.
- Add test for deduction guide.
Pull Request resolved: https://github.com/facebook/folly/pull/1097

Reviewed By: Orvid

Differential Revision: D14699047

Pulled By: yfeldblum

fbshipit-source-id: ee62bae0f4b05f1a110c2bd90470fecbb0666d24

Summary: [Folly] `kIsClang`.

Reviewed By: Igman

Differential Revision: D14844691

fbshipit-source-id: 80a738d5925d6d61e69f38ef5b43bf47691d233a

Summary:
It was broken, fix it.

Fixes https://github.com/facebook/folly/issues/1092

Reviewed By: yfeldblum

Differential Revision: D14842180

fbshipit-source-id: 899c95fdac3622c78c9c6f326cd14898007ddf6c

Summary: They're dead

Reviewed By: yfeldblum

Differential Revision: D14838871

fbshipit-source-id: 4b136d7a5914a1564f32bb87e6bec2b8d1411829

Summary: Because of the guard against constraint recursion in the definition of `struct awaitable_sender_traits` (awaitable/concepts.h, line 75), there is no need to be so conservative with the constraints of the generic `operator co_await` for `SingleTypedSender`s.

Reviewed By: kirkshoop

Differential Revision: D14816672

fbshipit-source-id: b2f49588ae98ea63a24f4044692470c59be15e29

Summary: Things have been ported, time to remove the fd version.

Reviewed By: yfeldblum

Differential Revision: D14703143

fbshipit-source-id: 13498d5a4c20d19c30c0d0fe942c5d62bb99aa08

Summary:
I only enabled this speculatively, but since it quite a heavyweight
test, it is timing out in the open source build, so let's turn it off again.

Reviewed By: lewissbaker

Differential Revision: D14817162

fbshipit-source-id: 343b021890974e5b92975bdfc7b4fa554d20a703

Summary:
When pulled in as an external project into an environment
that has changed the default cmake policy, the trick we're using
to extract the full library path errors out.  This works around
that issue by saving the policy settings and then setting them
how we need them before restoring them after we're done.

refs https://github.com/facebook/folly/issues/1100

Reviewed By: yfeldblum

Differential Revision: D14766991

fbshipit-source-id: e2506bc0aebd9e66bc70b12c4c538c9e034f9f73

Summary:
Shift AsyncPipe to be backed by NetworkSocket, as the only things this can be used for are listenable to by libevent. At the moment, the APIs for what are listenable to by libevent are all described in terms of NetworkSocket.
While, by name, this may seem incorrect, in practice, it is correct in function, due to the limitations of how NetworkSocket's are obtainable on Windows, and all existing calls to these functions on posix platforms are already functional by being file descriptor backed.
The plan is to have this, slightly badly named for the location, interface for the moment, and to clean it up in the future.

Reviewed By: yfeldblum

Differential Revision: D14702972

fbshipit-source-id: 785281a7bb9fafc63f5295e33f5bc136b9bc70f1

Summary: [Folly] Lower quadratic iteration count in an elias-fano test to make it take less wall time to run.

Reviewed By: philippv

Differential Revision: D14821341

fbshipit-source-id: 726fe5bd17442661196593ea6279314509622832

Summary:
- Mark `swap(Replaceable&)` noexcept iff `Replaceable` is
  noexcept-swappable. As the comment aludes to, this is a C++17 feature,
  but we have it backported in `folly/Traits.h`
Pull Request resolved: https://github.com/facebook/folly/pull/1070

Reviewed By: LeeHowes

Differential Revision: D14518839

Pulled By: yfeldblum

fbshipit-source-id: 966cca1733cf2a397c2406577baf941992f362d6

Summary: Prefer `FOLLY_HAS_COROUTINES` from `<folly/Portability.h>` over directly testing `__cpp_coroutines` because of MSVC.

Reviewed By: yfeldblum

Differential Revision: D14816497

fbshipit-source-id: 1f569824d5a79beca5255dfd43da8baf1556f8fe

Summary: Original commit changeset: e533bfa6b9d2

Reviewed By: heslami

Differential Revision: D14818275

fbshipit-source-id: 06ecb8058538ca150b941df56a050cf572f7eeea

Summary:
[Folly] Use wide multiplication for ints in UnboundedQueueTest

Fixes #1101.

Reviewed By: magedm

Differential Revision: D14794551

fbshipit-source-id: 223f9bb957eebb2fa065fe4945ce5a4339710823

Summary: All calls have been converted to the NetworkSocket overload

Reviewed By: yfeldblum

Differential Revision: D13566829

fbshipit-source-id: c6a3b155ffe35534c0297eba7819413e4ce77282

Summary: [Thrift] Cut `fbcode_builder` dep for `thrift` on `krb5`. In the past, Thrift depended on Kerberos and the `krb5` implementation for its transport-layer security. However, Thrift has since migrated fully to Transport Layer Security for its transport-layer security and no longer has any build-time dependency on `krb5`. Clean this up.

Reviewed By: stevegury, vitaut

Differential Revision: D14814205

fbshipit-source-id: dca469d22098e34573674194facaaac6c4c6aa32

Summary:
Add functions for concurrently awaiting a collection of SemiAwaitable objects.

* `folly::coro::collectAllRange()` takes an InputRange of SemiAwaitable objects and concurrently awaits them, producing a `std::vector` of the results. This will throw an exception if any of the input operations failed with an exception.
* `folly::coro::collectAllTryRange()` takes an InputRange of SemiAwaitable objects and concurrently awaits them, producing a `std::vector<folly::Try<T>>` containing the results. Failure of individual operations can be determined by inspecting the corresponding `Try` value.

This diff also refactors the existing implementations of `folly::coro::collectAll()` to make use of a new `Barrier` and `BarrierTask` implementation detail rather than using `DetachedInlineTask`, `Baton` and an atomic counter. This allows delaying the start of the child corouines until after all coroutines have been created which was necessary for cases where we didn't know the number of elements in the input range to allow storage for the results to be allocated in a vector before the tasks start producing results.

Reviewed By: andriigrynenko

Differential Revision: D14430147

fbshipit-source-id: 5ba4b7cdc3c9b65d1736f14a2d39f7beb6cb82b0

Summary: It's dead.

Reviewed By: yfeldblum

Differential Revision: D14191959

fbshipit-source-id: 8cfb68fa8f10b1f847bf6601aec8d869cb9c008d

Summary: To reflect the fact that the concept `SingleSender` is a refinement of `Sender` (there is no "`ManySender`"), rename `(any_)?(flow_)?many_sender` to `(any_)?(flow_)?sender`.

Reviewed By: kirkshoop

Differential Revision: D14803648

fbshipit-source-id: 5987ebe5ba877d0a3ab38e84b036f97302bcc70b

Summary:
The pushmi library has grown to the point where a little more structure would aid comprehensibility.

Also, finer-grained headers and build targets would make it easier for downstream projects to include only what they absolutely need.

Reviewed By: kirkshoop

Differential Revision: D14802895

fbshipit-source-id: 8d253f97c5d0529530cc3b7110505000f513ce83

Summary:
This is a performance optimization to avoid an executor re-queue for the collect[All] continuations which just move the result.

The biggest problem with this now is that these futures are often on a thread pool backed with a blocking MPMCQueue which means the add can potentially deadlock. This diff is not meant as a true fix to the problem though; just an improvement.

Reviewed By: yfeldblum

Differential Revision: D14754908

fbshipit-source-id: e533bfa6b9d2d820d50422b612c1489b560ec9d5

Summary: [Folly] `PriorityUnboundedBlockingQueue`, paralleling `UnboundedBlockingQueue` and `PriorityLifoSemMPMCQueue`.

Reviewed By: andriigrynenko

Differential Revision: D14536886

fbshipit-source-id: b3ea04c3f67d3e1e9d80626d963be70de1ae79ba

Summary: [Folly] Cut traits for old-ABI `std::basic_string` under libstdc++.

Reviewed By: pixelb

Differential Revision: D14764155

fbshipit-source-id: fa4991f4b70fb407672212886c84f1be4c33581f

Summary: For consistency adds thenExValue and deferExValue.

Reviewed By: yfeldblum

Differential Revision: D14793864

fbshipit-source-id: 700f02cfb27b93e1739c511f01142c30c1896990

Summary:
D14531266 defaulted static linking on macOS because of a link error in
folly/stats. A better fix for the issue is instantiating the
subclasses of BufferedStat, so add those.

Reviewed By: andrewjcg

Differential Revision: D14767345

fbshipit-source-id: eaa597dec468264595547f338348d7d74b49dd69

Summary: Renames defer taking an executor-taking continuation to deferExTry and modifying to take a KeepAlive for consistency with thenExTry.

Reviewed By: yfeldblum

Differential Revision: D14761230

fbshipit-source-id: af95afabb662197edcf2eeb432bc64c57011c776

Summary:
Add coroutine support to pushmi by making all sender types that satisfy the `SingleTypedSender` concept awaitable; that is, you can get their result by `co_await`-ing them in a coroutine.

In addition, all Awaitable types automatically satisfy the `SingleTypedSender` concept; that is, you can pass them to `pushmi::submit` along with a `Receiver` that is invoked with the result of `co_await`-ing the awaitable.

Reviewed By: kirkshoop

Differential Revision: D14688373

fbshipit-source-id: 8598b303aebcaa070033868abd440245c165bd73

Summary: Types opt in to receiver-ness with pushmi::receiver_traits, analogous to sender_traits (and std::iterator_traits). Alternatively, they can simply define a nested ::receiver_concept typedef that is an alias for one of "sink" or "flow".

Reviewed By: kirkshoop

Differential Revision: D14668335

fbshipit-source-id: 230abfcffef0cd231196b5f6a7ed05e1ac86ef0c

Summary:
* The sender concepts are now defined in terms of a `sender_traits` class template, in which the user can specify additional sender metadata like what values and errors it sends to the Receiver. This will make it possible to automatically make Senders awaitable from within coroutines.

* Remove the now-unused is_sender<>, is_single<>, and is_many<> properties.

* Define a `SingleSender` concept and a `TypedSender` concept.

* Define a `pushmi::sender` class template from which a Sender can inherit to declare the sender category (single/many/flow), and the types of the values and errors.

Reviewed By: kirkshoop

Differential Revision: D14631770

fbshipit-source-id: afcfd10bcd4e0e7ba224ea2c713d7fa92249ab7f

Reviewed By: lbrandy

Differential Revision: D14766712

fbshipit-source-id: 78c84117aa1c35825c72eb4bcd04e0bebb20c67f