Add stress test for atomic_shared_ptr

Summary: None of the existing unit tests exercises the local -> global refcount transfer, which is the most critical part in `atomic_shared_ptr` in terms of correctness, so add a stress test that `load()`s the pointer enough times to trigger a transfer.

Reviewed By: dmm-fb

Differential Revision: D39927439

fbshipit-source-id: a8a50e33dd4afd1bd7bf0a2c309d92503bfb949c

In SingletonVault shutdown log, keep timestamps for post-mortem debug…

…ging

Summary:
It is difficult to understand which of the destructors take unreasonable time, e.g. P527005137.
Adding timestamps to shutdown log to enable that retroactively on a shutdown crash.

(Note: this ignores all push blocking failures!)

Reviewed By: Orvid

Differential Revision: D39331011

fbshipit-source-id: d0c464fcac89e9aa6b13d006f0a8701baa8324d9

Fixes necessary to build benchmark tests on OSX Monterey (facebook#1847)

Summary:
ToT (Top of Tree) will fail to build if benchmarks are enabled (i.e.  cmake ../ -DBUILD_BENCHMARKS=ON) on Monterey with clang++ 13.1.6.

This patchset will fix the build.

Pull Request resolved: facebook#1847

Reviewed By: Gownta

Differential Revision: D39107587

Pulled By: Orvid

fbshipit-source-id: 0bec33e64ee4d869fd00f8c77ae3aa8a0b0be9a7

folly::coro::timeoutNoDiscard

Summary: Add `timeoutNoDiscard`. It differs from `timeout` in that it doesn't rewrite the result in the case of timeout. This prevents "lost" operations.

Reviewed By: iahs

Differential Revision: D39370243

fbshipit-source-id: 154a26ecebb39ab1a3ef69b7e943d18536c82438

Inline a method to save two hot load instructions. (facebook#1855)

Summary:
Pull Request resolved: facebook#1855

Inline coalesceWithHeadroomTailroom into coalesce to save two loads. The
profile information shows that we usually don't enter the if inside the
'tailroom' function, and the compiler can't sink the two method calls, so the
two loads are wasted. Sinking the two loads/method-calls into the the body of
the IF should allow us to avoid the cost of the hot loads.

Reviewed By: ot, philippv

Differential Revision: D39221756

fbshipit-source-id: 98e4f222c04f3fbf98c1de32bb7cdb0f2d9719f1

cut exception_wrapper in-situ storage variant

Summary:
Cut `exception_wrapper` in-situ storage variant as a step in revising `exception_wrapper`.

This changes the semantics such that now all stored objects are stored with reference semantics and not value semantics, as opposed to the prior situation which tolerated both semantics in the same type.

This removes the optimization of using in-situ storage for certain small exception types, and this is noticeable in the benchmarks. For example, construction regresses since there is an extra allocation/deallocation pair and an extra refcount increment/decrement pair. However, exceptions should be sufficiently rare that this regression not be a problem in practice. There single variant remaining, the thrown variant, has creation accelerated with `std::make_exception_ptr` and access accelerated with `folly::exception_ptr_get_object`. Moreover, move-construction may improve since moving `exception_wrapper` instances containing such small exception types no longer requires an extra refcount increment/decrement pair.

The prior situation was reportedly confusing, where some instances would have their own private copy of the stored exceptions and access would follow value semantics, but where other instances would have a copy shared with further instances and access would follow reference semantics. For example, one case moved-out the stored exception of a private instance of an `exception_wrapper` under the mistaken assumption that the instance had value semantics, where actually the stored exception was shared across multiple instances of `exception_wrapper` related to each other by copy, and the next access to the stored exception via another instance of `exception_wrapper` failed.

Reviewed By: Gownta

Differential Revision: D26377018

fbshipit-source-id: 81c392a6326e0a7eb5a54927db5f166867e431e1

Update SSLContext API to have loadTrustedCertificates support lists o…

…f strings

Summary: Adding ability to load a whole vector of certs at once.

Reviewed By: xybu

Differential Revision: D38057061

fbshipit-source-id: a8415176522595cb2db9c935cf5fda40cf33c459

Fix lastWakeup_ accounting in ThrottledLifoSem

Summary: `lastWakeup_` should be updated with the timestamp *after* the sleep, otherwise the next sleep will be 0. This means that currently only every other thread in a chain of wake-ups waits for the wakeup interval.

Reviewed By: philippv

Differential Revision: D38684505

fbshipit-source-id: 02f6e033083b58a8d878c73b087395938bd78327

Link to github in dynamic header

Summary: Clickable links are best

Reviewed By: praihan

Differential Revision: D38466482

fbshipit-source-id: 789216005e02c4cd07b48eef2231778058baffd5

consolidate tool exclusion in cmake build

Reviewed By: Gownta

Differential Revision: D38265963

fbshipit-source-id: 1d0cda00ddb7036bc83d241dd1747806e743fda1