Summary:
Cut `std::hash` specialization in folly for `signed __int128` and `unsigned __int128`.

There are cases where folly does polyfill things in the global and standard namespaces, but in order for folly to play nicely with other libraries this practice is overall to be avoided.

Reviewed By: nlutsenko

Differential Revision: D26455137

fbshipit-source-id: 948140513b8daf6fe777d04ea42346d69c5eb03f

Summary: Avoid the use of fancy macros in the `folly::enumerate` test suite in order to facilitate cmake's gtest integration.

Reviewed By: Orvid

Differential Revision: D26477530

fbshipit-source-id: 1838709e76016c389b052a9523a0681c58b8af96

Summary: Cut the `EventBaseLocal::emplace` overload taking a pointer since it is unused and to consolidate on the approach of not letting internals leak.

Differential Revision: D26246689

fbshipit-source-id: 0d9dc1bfcc3d37dc980142b221ee4443b98205a3

Summary: Use the global immutable executor by default and remove the default entirely for the deprecated form.

Reviewed By: yfeldblum

Differential Revision: D26414638

fbshipit-source-id: b48fe4683165e7f29e8003560f2917e143138bb0

Summary: Let `File` tests not use `/etc/hosts` since that file is not portable across all systems. Likewise the fs-util tests using `/usr/lib` and `/usr/bin`.

Reviewed By: chadaustin

Differential Revision: D26447117

fbshipit-source-id: 669a3d7e0ea4efd54ca2efd5c15d8a21227b5c8c

Summary:
[Folly] Let some utility functions be invocable objects: `to_signed`, `to_unsigned`, `to_narrow`, `to_underlying`.

Additionally, force the template type param to be deduced.

Reviewed By: markisaa

Differential Revision: D21959729

fbshipit-source-id: eedd9f2f84efb62c99ed0d5f65123a7fe000cf0a

Summary:
Currently there is logic that keeps histograms which only have one value reported to it separately.

The multi level histogram keeps the histogram for several time buckets. Just in case we had only one value added, it will keep reporting the value even if it was reported long ago and shouldn't appear in the time window level we are querying.

This is inconsistent with non histogram metrics (`avg` for example) reported from the same histogram, which don't have this logic.

A scenario would look like this:
1. Report value 5 to histogram
2. Immediately query the 60 second level -> `avg` reports 5, `p50` reports 5.
3. Wait 2 minutes
4. look at the 60 second level -> `avg` reports 0 (no value in the last 60 seconds) but `p50` keeps reporting 5.

I think this does more harm than good, so I suggest to kill this behavior.

Reviewed By: simpkins

Differential Revision: D26339016

fbshipit-source-id: 120485f8b78876e60ce66a534edcf86e7d9a836d

Summary: [folly] no `shared_ptr` storage in `EventBaseLocal` since its refcount operations are unnecessary. Pare it down to `unique_ptr` with a straightforward fun-ptr deleter.

Reviewed By: iahs

Differential Revision: D26246619

fbshipit-source-id: 3d0214ccb6c4730ca7b573065b4468a0dce85d40

Summary:
Introduce `folly::detail::thunk`, a carefully curated collection of generic general-purpose thunk templates: `make`, `ruin`, `ctor`, `dtor`.

Initial uses in `StaticSingletonManager` and `to_erased_unique_ptr`.

Reviewed By: vitaut

Differential Revision: D26417824

fbshipit-source-id: 482f4154a957b3c3dbeede726ed870a8d10c2a32

Summary:
Define macro `FOLLY_CPLUSPLUS` on all platforms.

Move the initial C++ language version check below the `#include` group.

Reviewed By: Orvid

Differential Revision: D26420758

fbshipit-source-id: 33d156ae944e7112b007afc49e8fb929dc677b85

Summary: Move REGISTER_TYPED_TEST_CASE_P to cpp from h

Reviewed By: yfeldblum

Differential Revision: D26440651

fbshipit-source-id: 7efd63b2f4af3de206cf2abdba33c849145c7834

Summary: Avoid static init crash in testing::internal::TypedTestCasePState::VerifyRegisteredTestNames when compiling with clang 9.0 (works OK with clang-8.0)

Reviewed By: yfeldblum

Differential Revision: D26419675

fbshipit-source-id: c961ac0f73e3fc8d8852eb32c0b89634fb854c37

Summary: See title.

Reviewed By: yfeldblum

Differential Revision: D26419855

fbshipit-source-id: 2857c80eb2500d8d3d5f815a1f795ddeac31165f

Summary: This allows the test client to be used with session tickets enabled, if desired

Reviewed By: yfeldblum

Differential Revision: D25795104

fbshipit-source-id: 1adfb6c17c6da24db9c6367485a5cd0f8e534661

Summary:
Existing API allows user to specify a single preferred CPU for SQ_POLL workers,
but it also allows specifying multiple workers.

Add API to allow specifying a set of CPUs, not just one.

The specified set of CPUs are used in round-robin fashion, so it is not strictly
necessary that the number of CPUs and the number of workers be the same.

#forceTDHashing

Reviewed By: yfeldblum

Differential Revision: D26380317

fbshipit-source-id: ecfd8722670ffc3e3dc246fae1bcb7f4d8e6f3e4

Summary: Lower max submit so the test will not fail with EINVAL when SQ size is > CQ size

Reviewed By: kevin-vigor

Differential Revision: D26419746

fbshipit-source-id: f9cbd4bef06b314f43ab667646d9e5ef61999246

Summary: See title

Reviewed By: knekritz

Differential Revision: D26320388

fbshipit-source-id: 2796ae36cd8779f56474d21e89e80562f7bad22d

Summary: The `AsyncGenerator` test for `co_safe_point` immediately-invokes a temporary coroutine function to return a coroutine object, but the temporary coroutine function goes out of scope while the returned coroutine is live. Use `co_invoke` to fix it.

Reviewed By: lxfind

Differential Revision: D26408345

fbshipit-source-id: 4650ea57c01baffb1065560b661e3296cb0292cc

Summary:
The new version of googletest has a different internal API for registering test suites. This diff updates `CO_TEST_` to match the new version of googletest. I created this diff by copying the new version of `GTEST_TEST_` and adding the coroutine `co_TestBody`, just like in the old version of `CO_TEST_`.

~~There is a problem that I'd like reviewer's feedback on: how do we cleanly make the switch over to the new version of `CO_TEST_`? This change is tied to a specific version of googletest.~~

~~I have 2 ideas so far:~~

~~(Plan 1) Land this diff stacked on top of D25585021, so that we don't have to manage two versions at once.~~ Not possible

(Plan 2) Have two versions of `CO_TEST_` at the same time, but select which one to use based on a preprocessor macro defined in gtest.

Reviewed By: yfeldblum

Differential Revision: D26365504

fbshipit-source-id: 196c2277c7942163951f5781b334d0bff5c34de8

Summary:
Some use-cases of AsyncPipe may want to backpressure writers in the
case that readers start to slow down. Without backpressure, the queue
(and pending work) can essentially grow without bound, leading to OOMs
or starvation of other resources.

This implements a bounded channel primitive by layering a semaphore
in front of an existing AsyncPipe. It inherits similar cancellation behavior,
though pending writes must be canceled in the case that the read end is destroyed.

Reviewed By: iahs

Differential Revision: D26260278

fbshipit-source-id: a936cf021023b31a6db868e27e9855480a1dc405

Summary: Add io_uring SQ/CQ poll tests

Reviewed By: yfeldblum

Differential Revision: D26387632

fbshipit-source-id: 2b68fa4f9af41b76b90183a010f81a92537a6e3f

Summary: Remove `BaseFormatter::fbstr()` because it is virtually unused and those few cases that do use it are better off with `std::string`.

Reviewed By: yfeldblum

Differential Revision: D26358276

fbshipit-source-id: e96b4958735b3c9c03010e1e3dcdd01abd2d6a33

Summary:
Let the function passed to `EventBaseLocal::getOrCreateFn` directly return the value to be stored, rather than allocating it and returning a pointer.

Details of its storage should be internal to `EventBaseLocal` and should not leak out into the creator functions.

Differential Revision: D26246341

fbshipit-source-id: d72a62998985e42ef517eede3654324dc9f48709

Summary:
Remove `folly::writeTo` which is untested and virtually unused, replacing
```
folly::writeTo(
    stdout,
    folly::format(...));
```
with
```
fmt::print(...);
```
for better compatibility with C++20 `std::format`.

Reviewed By: yfeldblum

Differential Revision: D26326405

fbshipit-source-id: 598ec5d7906b2d36b20aa63f410a0c2621bc62fa

Summary:
removed double of word 'lock'

Pull Request resolved: https://github.com/facebook/folly/pull/1523

Reviewed By: yfeldblum

Differential Revision: D26367154

Pulled By: Orvid

fbshipit-source-id: 586647d28b9ad002ee975450426d2da967dc638e

Summary: There are lots of unnecessary uses of shared_ptr for type erasure where refcounting is not needed because it is trivial to convert to shared_ptr<void>, while it is more involved to make the equivalent unique_ptr. Adds a utility that hides the details to bring effort parity between the two pointer types.

Reviewed By: yfeldblum

Differential Revision: D26323443

fbshipit-source-id: e33c5378f7df3105c7cb97d2622dca351b866035

Summary: Add IoUringBackend support for openat, openat2, close, fallocate #forcetdhashing

Reviewed By: kevin-vigor

Differential Revision: D26325604

fbshipit-source-id: dbeea468ffbb324dc8c7afc1dbde1e451f72e06d

Summary:
Making it consistent with other *OnNextLine rules.

Also fixes really odd line breaks after the return type of some functions, for example:
```
template <typename M>
iterator
insert_or_assign(const_iterator /*hint*/, key_type const& key, M&& obj) {
```
becomes:
```
template <typename M>
iterator insert_or_assign(
    const_iterator /*hint*/, key_type const& key, M&& obj) {
```

Reviewed By: yfeldblum

Differential Revision: D25987579

fbshipit-source-id: cc24ea3db1b256f5563fb051fbd572adfccf556a

Summary:
On macOS, ninja only provides prebuilt Intel binaries, not arm64 ones. Since we
need to have a native arm64 ninja to be able to build binaries for the M1,
let's compile it from scratch on macOS too. Compiling it takes less than 30s so
it shouldn't add any notable cost to any builds.

Reviewed By: genevievehelsel

Differential Revision: D26322029

fbshipit-source-id: 1bbeeac2a3f0ba07c477a2b33f8878a91ccaec0e

Summary:
Combining CancellationTokens is a common pattern in user code, which requires the user to manage a CancellationSource and attach a callback per token to be merged:
```
    folly::CancellationSource source;

    folly::CancellationCallbacki cb1{
        std::move(token1), [&] { source.requestCancellation(); }};
    folly::CancellationCallback cb2{
        std::move(token2), [&] { source.requestCancellation(); }};
    ...

    co_await co_withCancellation(source.getToken(), ...);
```

This diff adds the functionality to folly:
```
    co_await co_withCancellation(folly::CancellationToken::merge(token1, token2, ...), ...);
```

This diff subclasses CancellationState for simplicity. If the extra vtable ptr is a concern, we can instead template CancellationState to combine them, but that would come with a compilation speed penalty.

Reviewed By: yfeldblum

Differential Revision: D26277772

fbshipit-source-id: 169f65b4cf8f7f10f2800fb7797bb09379ecc239

Summary: Fix the uncaught-exception termination message expectations in `folly/lang/ExceptionTest.cpp` under libc++.

Reviewed By: markisaa

Differential Revision: D26328622

fbshipit-source-id: 2e2f9fce93c65556508d99d1f7b80f0971f35c0c

Summary: Lower min capacity so the test succeeds when ulimit -l has a lower value

Reviewed By: yfeldblum

Differential Revision: D26321362

fbshipit-source-id: 9d2c41fbbc190f46a6174c3f847943e13f5bea6c

Summary: Removed deprecated folly format APIs.

Reviewed By: yfeldblum

Differential Revision: D26317507

fbshipit-source-id: 67f42d059088946a34e77c944befa4dcb24d8673

Summary:
I've been trying to get EdenFS to compile on an M1 mac, and openssl only
support these starting with 1.1.1i. From their changelog:

    *) Add support for Apple Silicon M1 Macs with the darwin64-arm64-cc target.
     [Stuart Carnie]

Let's start by upgrading openssl.

Reviewed By: fanzeyi

Differential Revision: D26318433

fbshipit-source-id: dfe1a06ee7061dfcd026b0347eb050155895ebd9

Summary: Reexport `coroutine_handle` from `folly/experimental/coro/Coroutine.h`, which wraps inclusion of `experimental/coroutine`.

Differential Revision: D26227588

fbshipit-source-id: 32cdb2b6d87e90065bf9f140716d63670c04db5f

Summary: Similar to XLOG_EVERY_MS but for formatted messages.

Reviewed By: yfeldblum

Differential Revision: D26296133

fbshipit-source-id: 92d2e240d70cf40b903f19b0350c6d9cc0daa1fb

Differential Revision: D26265450

fbshipit-source-id: 0cefa65feb3fc3b76c1b8b8a9059dc4d321c6751

Summary: ...to avoid additional `get()` call if we need the emplaced object right after.

Reviewed By: yfeldblum

Differential Revision: D26238594

fbshipit-source-id: a66396173a27ac684b403b3c0c576be50c39dde2

Summary: `getOrCreateFn()` is prone to undefined behavior as it returns `nullptr` reference if factory function returns a `nullptr` or empty `std::shared_ptr<T>`, or if `emplace(evb, nullptr)` is called. This diff adds `DCHECK`s to enforce non-nulls. Note that `getOrCreate()` is not affected as it uses `std::make_shared` under the hood that is guaranteed to have a new object constructed.

Differential Revision: D26246552

fbshipit-source-id: b7bc7164841b70fc0889c782438bf96a61857bbd

Summary: EdenFS has a use-case for this macro but with 10+ arguments. Raise the maximum number of arguments from 7 to 15 to permit this.

Reviewed By: yfeldblum

Differential Revision: D26268587

fbshipit-source-id: be0b0579d854a93dc2983706a253370fe05905df