Summary:
In the vast majority of cases in fbcode, the sanitizer macros are checked with
`#ifdef`, so do the same in some places to fix `-Wundef`.

Reviewed By: luciang

Differential Revision: D32501662

fbshipit-source-id: 123e22d78476c9db0e328ccb01a734a06a163594

Summary:
Users may set `FOLLY_USE_LIBCPP` or `FOLLY_USE_LIBSTDCPP` to `0` or
`1`, so make sure everything checks it via `#if`.

Reviewed By: luciang

Differential Revision: D32496561

fbshipit-source-id: 5eee660b540af0bf1eaee4949576acf11c9be8d0

Summary: Cast the result of shifts to the integer type parameter, rather than the shift width.

Reviewed By: aary, luciang

Differential Revision: D32160950

fbshipit-source-id: 0100a1a92c48bfa1cd957cf42ed84d01302ea4f7

Summary: Previously update() could be run both directly from within the update notification callback (that can happen in any context while holding some application locks), but also while observer is initially created from the observable. This change moves the update logic to be called on the observer internal thread, making the implementation even more resilient to various locks that may be held by the application.

Reviewed By: yfeldblum

Differential Revision: D32338088

fbshipit-source-id: 557f47c98469a951ce4b5513ce96e2b3a5d8ba6b

Reviewed By: igorsugak

Differential Revision: D32415926

fbshipit-source-id: b62ac87215800eed8e6cf8acf92f0518b9a56da5

Reviewed By: ot

Differential Revision: D31763384

fbshipit-source-id: ef666193f83a16d4eedf65a499361d967809163f

Differential Revision:
D32405920 (https://github.com/facebook/folly/commit/6b58b9e81875d92ba53add248d113593b24d2c8b)

Original commit changeset: b0b5c55149ae

fbshipit-source-id: f5a6d0dd6bb4434cebe02b5cde357d94e28253f0

Summary: Provies single-await co_nothrow wrapper

Reviewed By: capickett

Differential Revision: D32405920

fbshipit-source-id: b0b5c55149aebe4246422fabeffb7ed433b4e20d

Summary: Have constructor do init() instead. I can't think of any valid reason to have this API other than confusing everyone who decides to use it.

Reviewed By: praihan

Differential Revision: D32302228

fbshipit-source-id: 136e0d2919291acd60be953337309cd006bd674b

Summary: Now we'll ignore all the known sanitizers: ASAN, TSAN, MSAN, UBSAN.

Reviewed By: yfeldblum, jdonald

Differential Revision: D32364717

fbshipit-source-id: f71a294dd9ebaa8521f0487ebc65c6f7c1e7bf04

Summary:
Inlining `getStackTraceInPlace` fails on Fedora with

```
/builddir/build/BUILD/folly-2020.12.21.00/folly/experimental/symbolizer/StackTrace.cpp: In function 'folly::symbolizer::(anonymous namespace)::getStackTraceInPlace(ucontext_t&, unw_cursor&, unsigned long*, unsigned long)':
/builddir/build/BUILD/folly-2020.12.21.00/folly/experimental/symbolizer/StackTrace.cpp:88:9: error: function 'folly::symbolizer::(anonymous namespace)::getStackTraceInPlace(ucontext_t&, unw_cursor&, unsigned long*, unsigned long)' can never be inlined because it uses setjmp
   88 | ssize_t getStackTraceInPlace(
      |         ^~~~~~~~~~~~~~~~~~~~
```

Reviewed By: yfeldblum

Differential Revision: D29639503

fbshipit-source-id: 823692d95171d5fb3968761c00b4cc5650db171b

Differential Revision: D32394449

fbshipit-source-id: 2c2bc54c2d71cc677c35195adc73f579b5d0c101

Summary: exception_wrapper no longer rethrows during this construction

Reviewed By: yfeldblum

Differential Revision: D32371738

fbshipit-source-id: 0d5d51cdde4115aaf8774950e53bfac0b37f0b88

Summary: Breaks rescheduling on error

Differential Revision: D32398162

fbshipit-source-id: 582db297e0b700617c2e455e1f6d2fd1fa40e143

Summary:
- For Windows in xplat we use [Clang9/12 + MSVC 2017](https://www.internalfb.com/code/search?q=filepath%3Afbsource%2Ftools%2Fbuckconfigs%2Fcxx%2Fwindows%20repo%3Afbsource%20vs2017_15.9) and it's [<experimental/coroutine>](https://www.internalfb.com/phabricator/paste/view/P467028969)
- When upgrading to MSVC 2019 we also get [<coroutine>](https://www.internalfb.com/phabricator/paste/view/P467029007) in addition to  [<experimental/coroutine>](https://www.internalfb.com/phabricator/paste/view/P467029069)
- The MSVC2019 ```<coroutine>``` does not work with our current Windows [coroutine solution](https://fb.workplace.com/groups/rp.co.await/permalink/748274059391415/) / [D26877884](https://www.internalfb.com/diff/D26877884)
- Hence I am *re-using* an existing flag to opt-out of ```<coroutine>``` as MSVC 2019 left for backwards compatibility also ```<experimental/coroutine>```

Reviewed By: yfeldblum

Differential Revision: D32282445

fbshipit-source-id: d2eeb3057518e95135cf054ab6156dac5aa15065

Summary: Provies single-await co_nothrow wrapper

Reviewed By: capickett

Differential Revision: D31450124

fbshipit-source-id: e8d18c49248d85ef864abad2eecffe69de58a546

Summary: Extended coroutine handle interface for attaching new features

Reviewed By: capickett

Differential Revision: D31450126

fbshipit-source-id: 2aa6b8a4fb72e5007d08f38fcde5c7abd8eabd67

Reviewed By: AjanthanAsogamoorthy

Differential Revision: D32328268

fbshipit-source-id: e2816aa89d0ffc96e0377603fca6ca4a39830def

Summary:
We're hiding internal frames during dump, but forget to adjust
the reported count, so the output looks clowny.

Reviewed By: luciang

Differential Revision: D32340094

fbshipit-source-id: 86a02c54178a03a83fa64c29faf6d82abfa45bf8

Summary: I use it in the next diff.

Reviewed By: xavierd

Differential Revision: D32280314

fbshipit-source-id: 32d86780ae3187449fc7ceb1dc474cc7ac2d1ac1

Summary: Let their types be determined by the atomic parameter only so that conversions may happen at the call-site. This is what the `std::atomic_compare_exchange` functions do. Those functions are specified to use `typename std::atomic<T>::value_type` rather than `T` but, curiously, not all implementations of `std::atomic` have member type alias `value_type`. So instead of using member type alias `value_type` to get dependency, we use `type_t`.

Reviewed By: ot

Differential Revision: D32259208

fbshipit-source-id: 413e3580aff3dd0ece5051b7a62dfd16f49eb5a8

Summary: Use invokers and rename the invokers for clarity.

Reviewed By: luciang

Differential Revision: D32160395

fbshipit-source-id: 97052fc42c23db3e315bd721dcf7b3f29669bcaa

Summary: Choose `_native` and `_fallback` suffixes nstead of `_x86` and `_deafult` suffixes. `_native` instead of `_x86` since the intention is broader than x86; `_fallback` instead of `_default` since fallback is a clear word.

Reviewed By: luciang

Differential Revision: D32160396

fbshipit-source-id: 08d064aae670ea7e93a098b16a511ff43ca8a7e1

Reviewed By: praihan

Differential Revision: D32297368

fbshipit-source-id: a58c7805d396cc30c0d82a49d5998e2f9d8a799a

Summary: The bit param is 0-based. So, for bit 1, the mask is `0b10`.

Reviewed By: luciang

Differential Revision: D32160394

fbshipit-source-id: cfef250b3035d8c7e29b2f36848482888801781c

Summary: This fixes mvfst OSS tests on Mac.

Reviewed By: lnicco

Differential Revision: D32227732

fbshipit-source-id: 1375799046803a5183f83441b93deb0ab2e0961f

Summary: As titled

Reviewed By: ivmaykov

Differential Revision: D31974990

fbshipit-source-id: 03b60c4e9d4a7464c27cd53f1989cda878132933

Reviewed By: Gownta

Differential Revision: D32197071

fbshipit-source-id: 19ead9d39777efa73750346e59dc69ef57ce4961

Summary: We currently call closeReceiver() on a receiver in receiversBySubscriptionId_ that has already been removed from receivers_ (and freed). The fix is to check to see whether the receiver in receiversBySubscriptionId_ is still present in receivers_ before invoking any functions on it. If the receiver is not present in receivers_, that means it has already been closed, so we do not need to close it.

Reviewed By: aary

Differential Revision: D32233062

fbshipit-source-id: 0aa763b509897db7de970e12e2f01cffc1b49fc3

Summary: macos build is failing, disable to get clean internal CI

Reviewed By: HarveyHunt

Differential Revision: D32238516

fbshipit-source-id: dad2d15d65a9ee971c254775fcc8e6e80fa97ab0

Reviewed By: mingtaoy

Differential Revision: D32218077

fbshipit-source-id: 5a966ced07387340d02ad2a85c095939c9e7f530

Summary: Fixing an incorrectly formatted link in https://github.com/facebook/folly/blob/main/folly/docs/Synchronized.md.

Reviewed By: yfeldblum

Differential Revision: D32227087

fbshipit-source-id: 773ed8f1a0651a177fd306b63b514a917a8934bd

Summary: Currently, when we get a cancelled notification on the output sender for a transform, we schedule the cancellation of any running transform callback on an executor. CancellationSource is thread-safe, so we don't need to do this. This diff ensures that cancelling the output receiver will immediately trigger cancellation of any running transform callback, allowing the transform callback to deterministically know that it has been cancelled.

Reviewed By: aary

Differential Revision: D32191796

fbshipit-source-id: aabd58be9b58b0f560b8127b336e626c3482a644

Summary: This ensures that closing a FanoutChannel will synchronously cancel its input receiver. We don't need to run this code on the provided executor, because it is not running as part of a callback (and because we already have a lock).

Reviewed By: aary

Differential Revision: D32191795

fbshipit-source-id: 9ec2a2303a12580e052de7af8651ebf5b773b8b6

Summary: To ease writing and reading IO code.

Reviewed By: iahs

Differential Revision: D32220718

fbshipit-source-id: db695eb33a60aae0db1b34991bd7d5ce33489f86

Summary: The memory orders seem backwards but are correct. Add a comment explaining the surprising situation.

Reviewed By: ot

Differential Revision: D31594601

fbshipit-source-id: 61a3f5a5700cfe640e025ad5d5388a5dda1f97eb

Summary:
Add an implementation of `WorkerProvider` to `IOThreadPoolDeadlockDetectorObserver`.

- On thread start, the observer creates a `TaskStallDetector` and provides it a `ThreadIdCollector` containing the TID of the thread that started
- The thread will be blocked from exiting until any existing `WorkerProvider::KeepAlive`s are destroyed

Differential Revision: D31963855

fbshipit-source-id: cd749d76a433684a8280d95a4dbff441e32bfda9

Summary: Move the hg build to fbthrift_socket, it was referencing the removed fbthrift_socket-02 from some cargo files

Reviewed By: mzr

Differential Revision: D32159904

fbshipit-source-id: 58db0fd53cf859d19a237c0a1d50370d405132b8

Summary:
rpm names needed updating as boost-contract wasn't found.

The rpm names now include the boost version which should be handy when boost dependency is updated next

Unfortunately the centos boost 1.69 rpms aren't found automatically by the centos cmake boost rules, so this also adds builder.py logic to point to it using  BOOST_INCLUDEDIR and BOOST_LIBRARYDIR

Reviewed By: mzr

Differential Revision: D32140834

fbshipit-source-id: 3e2dd822613957ae4d7be5b73623ff581f11d02b

Summary: Add the system dependencies so getdeps can install them if not present

Reviewed By: krallin

Differential Revision: D32152976

fbshipit-source-id: 29311f37126e1eac8dfbe2f46f6bc3516848d48b