Reviewed By: yfeldblum

Differential Revision: D27253836

fbshipit-source-id: a72531e0deca9f0f6582339d483ed56f2eb37d87

Summary: Add a CHECK() to the IOBuf constructor to ensure the underlying buffer does not include any bytes that have been poisoned by AddressSanitizer.

Reviewed By: yfeldblum

Differential Revision: D25627483

fbshipit-source-id: 5bca59e545c32ba7b61430d7cd4787712eecb204

Reviewed By: Orvid

Differential Revision: D27559344

fbshipit-source-id: cf90be5a19da87bffe72fb6646296dd46889bc90

Summary:
Close-on-exec is an important feature to avoid leaking file descriptors to spawned processes.

This diff adds `File::dupCloseOnExec()` function which is equivalent to `File::dup` function, but sets close-on-exec flag where supported (i. e. everywhere except Windows).

Practically most users want to have `closeOnExec = true` by default, but we need to preserve backwards compatibility, thus this diff leaves `dup` function as is.

Reviewed By: yfeldblum

Differential Revision: D27495214

fbshipit-source-id: 540deb2bc6c8fec626a0120bc20c345950dc8af7

Reviewed By: Orvid

Differential Revision: D27555992

fbshipit-source-id: d68dc183f7c7b7c64975ede21e6f3018575fb7c9

Reviewed By: Orvid

Differential Revision: D27537202

fbshipit-source-id: d1aa8ad81f3749129dd6f1795e788a0e55d9f06a

Summary: Between gnu and msvc compilers, the builtins and intrinsics `__builtin_clzll` and `__lzcnt64` have different return types. They are wrapped in a portability function in `to_ascii` which currently forces a return type, which return type could trip implicit-conversion warnings. Avoid forcing the return type to avoid tripping implicit-conversion warnings.

Reviewed By: luciang

Differential Revision: D27535348

fbshipit-source-id: 3339d5d04e095f936c18d2b39ac1fb4b32f8a810

Summary: The ambiguity is that the name appears as both `folly::to_ascii_size` and `folly::detail::to_ascii_size`, so that anything which imports both `folly` and `folly::detail` or which is within `folly::detail` may, depending on the particular compiler, find an unqualified call ambiguous.

Differential Revision: D27535255

fbshipit-source-id: 05db44e6b153d75b691d67c878785cab1655051d

Summary: SingletonRelaxedCounter is templated over tag types, but this leads to heavy monomorphization. extract the common code.

Reviewed By: aary

Differential Revision: D27383671

fbshipit-source-id: e419579e4d9ab84b5107670bd1e94e1768dcebf4

Summary: No strict need for them to be functions.

Differential Revision: D27484949

fbshipit-source-id: f0fbc91aa9940720bbd3b685b730acc314bf122d

Reviewed By: philippv

Differential Revision: D27535007

fbshipit-source-id: 384235553b18320fa9728740f7f1b83b4c8b0a1a

Reviewed By: WillerZ

Differential Revision: D27489039

fbshipit-source-id: 5d397a24fcb165b7ea25e32f0c81435bce474b04

Summary:
Win32 does not support weak symbols.  This means the linker will fail due to the unresolved symbol `__asan_region_is_poisoned` when ASAN is not available.
Use `FOLLY_CREATE_EXTERN_ACCESSOR` macro when filtering for ASAN availability in `SanitizeAddress.cpp` to ensure the symbol `__asan_region_is_poisoned` is referenced only when ASAN is available.

Reviewed By: yfeldblum, luciang

Differential Revision: D27278826

fbshipit-source-id: 3c8efbfc687295c2fa70e3712402fd8dfe81bc45

Summary:
NestedCommandLineApp works by performing two passes over the argument vector:

1. Treat the entire argv as a single command, ignoring subcommands. Consume all options that "match" a global flag here.

2. Find the first positional argument in argv and treat it as a subcommand, then re-parse everything after it as options to the subcommand.

This allows constructs like `cli subcommand --subcommand-flag --global-flag`. However, as a side effect, if `subcommand-flag` happens to be even a prefix of a global flag name, boost "helpfully" matches it with the global flag in the first pass so that the subcommand never even sees it.

Fixing the issue of a subcommand flag conflicting with a global flag would break backwards compatibility; however, we can at least add an option to disable prefix matching so that there would need to be an exact flag name collision for this scenario to occur. It may be possible to extend NestedCommandLineApp in the future to detect flag collisions as well, instead of silently associating arguments with the wrong flag.

Reviewed By: yfeldblum

Differential Revision: D27419341

fbshipit-source-id: 941c0769e16d40c87f96620155f0ab09927cef3d

Summary:
If zero copy starts out enabled on a unencrypted connection, disable
it when the SSL handshake begins. Otherwise, AsyncSocket will receive send
notifications for buffers that it did not pre-register.

Reviewed By: yfeldblum

Differential Revision: D27145232

fbshipit-source-id: f07d006d095e1e6839eb2f4adb07d7c284786572

Summary: A mechanism to make it easier conditionally to access possibly-defined names, which can be a bit tricky especially without `if constexpr`.

Reviewed By: luciang

Differential Revision: D27434753

fbshipit-source-id: 4c9aaebe89b354aa94f4f370c8fbdc8456b0e7f7

Summary:
On darwin with Xcode, deleter for a shared_ptr is passed by copying the whole lambda, causing issues when building PrimaryPtr.

In this diff we try to work around this problem by putting the last promise value on the heap while sharing the control block with the shared_ptr exposed to user.

Reviewed By: yfeldblum

Differential Revision: D27354087

fbshipit-source-id: 9cb23912c9f1058505f3c20384eb998233eba2da

Summary:

Pull Request resolved: https://github.com/facebook/folly/pull/1553

Differential Revision: D27504501

Pulled By: yfeldblum

fbshipit-source-id: 291036c64050896d3873271722c84380fd2b3ec7
Co-authored-by: Zhiwei Chen <condy0919@gmail.com>

Summary: Let the `to_ascii` functions take and check the out range, not just the out begin. Add overloads which take arrays.

Reviewed By: WillerZ

Differential Revision: D27484478

fbshipit-source-id: 56a9ba1465204b58d9df018bdae707d39ef6525d

Summary:
Before this patch, it relies on the c++17 guaranteed copy elision, since `SemaphoreBase::Waiter` is NOT copy constructible due to `std::atomic<intptr_t>`.

This patch makes it compile even using `-std=c++14`.

Pull Request resolved: https://github.com/facebook/folly/pull/1554

Differential Revision: D27505938

Pulled By: yfeldblum

fbshipit-source-id: 6c26f82f77a24236a1233cff7921cd6b9f53bd09

Reviewed By: yfeldblum

Differential Revision: D27461546

fbshipit-source-id: 500ecd312b0b0ba15db7f373a3b6c7d633802846

Summary:
Tpx works well for all OSS jobs, we should migrate to Tpx instead of
relying testpilot.

Reviewed By: marcelogomez

Differential Revision: D27396539

fbshipit-source-id: dd1994be0eb3b4396764c1f3e68ddd6763b60139

Reviewed By: yfeldblum

Differential Revision: D27461452

fbshipit-source-id: 3e46ba71b10aea10d6006c3c6f145377ce6687da

Reviewed By: yfeldblum

Differential Revision: D27461464

fbshipit-source-id: 53facc1cc42f60ffa3083c12c32352bb78b16280

Summary: Persuade the compiler to emit a tail call in the `SingletonRelaxedCounter` inline slow path.

Differential Revision: D27383131

fbshipit-source-id: 6f58b1eb23477c00221515d7d6e55a91668f7c84

Summary:
C++17 permits use of `std::unique_lock` instead of `folly::SpinLockGuard`; just use that.

C++11 permits use of `std::unique_lock<SpinLock>`.

Differential Revision: D27390136

fbshipit-source-id: bcc34a043644289523ddfdfbf47a260c133db186

Summary: Avoids warnings with -Wterminate.

Reviewed By: yfeldblum

Differential Revision: D27452798

fbshipit-source-id: 129b66179c5ff9bbf3f8fe791224cd925a3b33f3

Summary:
This is towards upgrading to newer OpenNSA 6.5.21.

This diff will be landed after we have verified that OpenNSA 6.5.21 has not
regressed from OpenNSA 6.5.19 (HW test results)

sha256sum on the tarball, and use that in the manifest.

Reviewed By: jasmeetbagga

Differential Revision: D27470306

fbshipit-source-id: a56c1e012a790b0bc948a5804dd605cdf224916c

Reviewed By: yfeldblum

Differential Revision: D27461647

fbshipit-source-id: cf1f9747b535a9fb8c8aa851f2ce9cad8b9dce61

Summary: This is useful if your environment takes partial dumps of the process on crash - you can tell the process which does that to include the memory range of the flight recorder in what is dumped.

Differential Revision: D27206631

fbshipit-source-id: b3679004930486094e6030b5b39ee02d01bc7617

Summary:
jemalloc has background threads to purge unused pages, so stop purging
arenas used by idle threads since those arenas may be shared with active
threads which must page fault new pages in after the purge

Reviewed By: ot, davidtgoldblatt

Differential Revision: D26793042

fbshipit-source-id: 13617bfb0ace1c5b7aae52dfdc428b19ff99f7bf

Summary: This satisfies a new error in llvm11

Reviewed By: smeenai

Differential Revision: D27421459

fbshipit-source-id: 963d809a4f430a365223786ea7d2591fe239062f

Summary:
Enabling Velox/f4d getpdeps build on macOS, and adding the last
required dependency (libicu).

Reviewed By: amitkdutta

Differential Revision: D27316661

fbshipit-source-id: 60edf7d4a953a62f2b7b52d9b4f0827d12ac1367

Reviewed By: vitaut

Differential Revision: D27372973

fbshipit-source-id: fcbad2d9fceaecb4b02f5015a808f5744c7cb771

Summary: Add IOBuf multi create/destroy benchmark

Reviewed By: mokomull

Differential Revision: D27406019

fbshipit-source-id: 001cd92743892bb91111831044fc3a4d3b02aaf7

Summary: Rename, mark as kept, add a version taking an argument

Differential Revision: D27383132

fbshipit-source-id: 214af5b09e42159756e97b3b829ac9ecb237c17b

Summary: IOBuf allocate SharedInfo in the same allocation only if we do not cross into the next class size

Reviewed By: jrahman

Differential Revision: D27353914

fbshipit-source-id: 65be3641d7eaa0588bf26101ce02668fb827eee9

Summary:
`folly::observer::ReadMostlyAtomicObserver` docblock explains it well:
```
/**
 * A ReadMostlyAtomicObserver guarantees that reading is exactly one relaxed
 * atomic load. Like AtomicObserver, the value is cached using `std::atomic`.
 * However, there is no version check when reading which means that the
 * cached value may be out-of-date with the Observer value. The cached value
 * will be updated asynchronously in a background thread.
 *
 * Because there is no version check when reading, ReadMostlyAtomicObserver
 * does not capture observer dependencies when used from makeObserver. It is not
 * possible to create a dependent observer. Therefore, ReadMostlyAtomicObserver
 * models none of CopyConstructible, MoveConstructible, CopyAssignable, or
 * MoveAssignable. Dependent observers should be created using the underlying
 * observer.
 *
 * ReadMostlyAtomicObserver is ideal for fastest possible reads on a
 * trivially-copyable type when a slightly out-of-date value will suffice. It is
 * perfect for very frequent reads coupled with very infrequent writes.
 */
```
Other minor changes:
* Added `getUnderlyingObserver` to each of the special-purpose observers.
* Sprinkled `std::move` where neccessary.

Reviewed By: yfeldblum

Differential Revision: D27381490

fbshipit-source-id: fed039ae481af0c4e4148f80df03e30a125e333f

Summary:
This page says it can't be greater than 256:
https://docs.microsoft.com/en-us/windows/win32/api/winsock/nf-winsock-gethostname

POSIX one is equal to 255, so set it to 255.

Reviewed By: yfeldblum

Differential Revision: D27370716

fbshipit-source-id: 458ebc02d2914a0422862fafedd089e0c13c5e6d

Summary: Adding libicu manifest, which will be used by f4d/velox

Reviewed By: amitkdutta

Differential Revision: D25934666

fbshipit-source-id: 247d9961c90fcfc497498b556626e9ef1739b191