Summary:
This diff adds support for optionally setting an obj batch such that objects retired to the batch bypass the batch and are directly pushed to the domain.
This is useful in cases where the benefits of earlier reclamation outweigh the benefits of amortizing the cost of pushing objects to the domain.

Reviewed By: davidtgoldblatt

Differential Revision: D18114444

fbshipit-source-id: a1af88dfa1b601827b2f71e700fb1dfcb08e513c

Summary:
Fix -Wshorten-64-to-32 compile warning treated as error

(Note: this ignores all push blocking failures!)

Reviewed By: nbronson

Differential Revision: D18493255

fbshipit-source-id: 10a2a576c7f587d02a61860285171f2c5f0534ad

Summary:
Allows a category to be configured to selectively propagate messages to its parent category.

I preferred this over a simple `true`/`false` filter because it can be useful to have one log file for all DBG logs from a component while still limiting it to `INFO` or whatever in the main application log file.

Reviewed By: simpkins

Differential Revision: D18360804

fbshipit-source-id: a45098c81fdc8b984e440748880e8943df95acd2

Summary:
Allow hazptr_obj_batch-es to outlive the default domain.

Background: Until now it was expected that batches are destructed before the destruction of the default domain. This may be restrictive to good use cases.

This diff removes the above mentioned restriction and gives hazptr_obj_batch::shutdown_and_reclaim() to decide whether or not to access the default domain by supporting a singleton bool in default_hazptr_domain_alive() that could be made to outlive the default domain by calling that function before constructing the user object that may outlive the default domain.

Reviewed By: aary

Differential Revision: D18297698

fbshipit-source-id: 6021479ab36ce86d2c411fd71974fde47f7502a2

Summary: In a coroutine returning a `coro::Task<T>`, make `co_return {};` return a default-constructed `T` rather than a default-constructed `folly::Try`.

Reviewed By: yfeldblum, andriigrynenko

Differential Revision: D18466785

fbshipit-source-id: ed20f0f5ad98a0ef8e693e490417a7e36f706822

Summary: Newer CRT versions have changed how certain definitions were gated, so we need to update to support them.

Reviewed By: akrieger, guangyfb

Differential Revision: D18435320

fbshipit-source-id: 40ffc57ed1fe0ad1b01ab834aed9762bde5563cb

Summary:
Not supported under ARMv6 and below.
Pull Request resolved: https://github.com/facebook/folly/pull/1176

Reviewed By: yfeldblum

Differential Revision: D18145021

Pulled By: Orvid

fbshipit-source-id: 21796e225f650038479d1f5a858313dab0477e66

Summary: As title.

Reviewed By: yfeldblum

Differential Revision: D18351375

fbshipit-source-id: 594270e7b6b7056f7223fa7575ee33fed8f7d266

Summary: [Folly] `is_trivially_copyable_v`, an inline variable helper for `is_trivially_copyable`.

Reviewed By: nbronson

Differential Revision: D18413171

fbshipit-source-id: e256e40351492ecce8cfd31151569544aadca4e1

Summary:
Synchronized::operator-> is dangerous and leads to accidentally
acquiring locks or accidentally releasing them too soon. Add a
deprecated attribute that suggests using explicit locking instead.

Reviewed By: yfeldblum

Differential Revision: D18351232

fbshipit-source-id: 7bd3702bdd9b58dfa0fc7e8b8b636db9b5d7af8f

Summary: From https://en.cppreference.com/w/cpp/thread/condition_variable/wait, predicate should return ​false if the waiting should be continued. In the current code, the logic is flipped.

Reviewed By: yfeldblum

Differential Revision: D18413606

fbshipit-source-id: 3b59efe49946a223204ab68fc9fbe3bb98e3b58e

Summary:
Found with modernize-deprecated-headers

Note that C++14 deprecated these.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1254

Reviewed By: Orvid

Differential Revision: D18369623

Pulled By: yfeldblum

fbshipit-source-id: e824bc5474f94fa4fe359ac6a4899eeaaddbb13b

Summary:
Fixes cases where control would reach the end of the setThreadName function without an explicit return
Pull Request resolved: https://github.com/facebook/folly/pull/1250

Reviewed By: Orvid

Differential Revision: D18222073

Pulled By: yfeldblum

fbshipit-source-id: 09cbefcc7e2ca80a95d61a4394ed74be8ea8ec55

Reviewed By: yfeldblum

Differential Revision: D18298354

fbshipit-source-id: 683da223956570304a9ba80dacfb4659ff9a114b

Summary:
We need the ability to dynamically change the priority of an executor when it's executing work. Specifically, we want to de-prioritize a query's work units while it is being executed if it has taken up too much resources.
Pass a callback function to ExecutorWithPriority which sets the priority upon every add.

Reviewed By: andriigrynenko

Differential Revision: D18344806

fbshipit-source-id: c0c59d3982637703e6e7c3793930b9d9fadce6fe

Summary:
Add folly::hardware_concurrency

(Note: this ignores all push blocking failures!)

Reviewed By: yfeldblum

Differential Revision: D18342219

fbshipit-source-id: 806ef7b46e52896fa289a79854e45447929ab632

Summary:
In preparation for adding a deprecation annotation to
Synchronized::operator->, ensure that folly builds deprecation-clean.

Reviewed By: yfeldblum

Differential Revision: D18351118

fbshipit-source-id: f90a230d24a95e3983989d2c3caf7a4383a8226a

Summary:
If the source IOBuf is not chained, we short-circuit the cloning logic and
instead share the existing buffer. This works when we implicitly require the
clone to have the same headroom and tailroom requirements as the source
IOBuf. However, the API allows setting other values. We need to test that we
meet those requirements in order to use the shortcut. This diff adds those
checks.

Reviewed By: yfeldblum

Differential Revision: D18180574

fbshipit-source-id: 31bfaabd7663cf83faf919d8cb242df0a7a2969d

Summary:
Using a ThreadPoolExecutor instead of calling getCPUExecutor() to get one.
getCPUExecutor() by default returns an inline executor, which doesn't ensure that sleep is triggered on the same executor as the fiber loop.

Reviewed By: andriigrynenko

Differential Revision: D18353758

fbshipit-source-id: 6810d11b12de3f62ac5286dba00d2b4a9d447eba

Summary: We removed this dependency in the code a while ago.

Reviewed By: saifhhasan

Differential Revision: D18366401

fbshipit-source-id: daf1c7865fe0c312694bdf395f9d7ace99f70395

Reviewed By: yfeldblum

Differential Revision: D18356211

fbshipit-source-id: ad2361c65260a1a09f8536d07e5447cfcf5cc552

Summary:
Add a fixed single-writer multi-reader hash map that supports:
- Copy construction with optional expansion
- Concurrent read-only lookup.
- Concurrent read-only iteration.

The map has fixed size. Higher-level users can manage instances of this map to build a more general unbounded map.

Reviewed By: davidtgoldblatt

Differential Revision: D17522603

fbshipit-source-id: b4fcfe427a343f7226d216670536f2594f187bf3

Summary: Change UserCounters to 64bit integer to fit big counters like cpu_cycle counting.

Reviewed By: yfeldblum

Differential Revision: D18249661

fbshipit-source-id: 7808ef51e7230aa929ed555ed837ba5dd1962944

Summary: Coro.Throw and Coro.FutureThrow are exactly the same. Delete one

Reviewed By: lewissbaker

Differential Revision: D18337140

fbshipit-source-id: 9cf479b3e33599b3c8c982cabc87d1227d4efb9b

Summary: Add microbenchmarks for request context operations.

Reviewed By: LeeHowes

Differential Revision: D17434204

fbshipit-source-id: 22c804d71b922dd6894bdf6d98449f0c60c83206

Summary: Adds `pwritevNoInt()` and `preadvNoInt()` to FileUtil.

Reviewed By: yfeldblum

Differential Revision: D18213230

fbshipit-source-id: 0dfcd19ecf4500aa2bba279800d65821309371c6

Summary: [Folly] `invoke_traits`, which can take an invoker and make a traits type that has member traits that look like the std's invoke traits.

Reviewed By: vitaut

Differential Revision: D18082562

fbshipit-source-id: 9cfaf1a43fd695a23f7eb30d28a3c8aaaf2b9b60

Summary:
Add timeout support to python executor. Executor uses a separate thread running a folly event base.
Whenever a timer operation is called, scheduleTimeout is executed in that event base.

Reviewed By: andriigrynenko

Differential Revision: D18201883

fbshipit-source-id: 337518bea52993461201d6706f438130673ef6ce

Summary: Add iouring folly support, refactor the async IO

Reviewed By: kevin-vigor

Differential Revision: D17834511

fbshipit-source-id: e20c876a32730549f305334fd5eed02cccf23638

Summary: BENCHMARK_COUNTERS tests need to set `using folly::UserCounters` or `using namespace folly`. Add absolute namespace path in the macro to get rid of this requirement.

Reviewed By: Orvid

Differential Revision: D18271700

fbshipit-source-id: 084041142b7c245e3c861943570416ac89ea6c34

Summary: [Folly] Permit multiple primary namespaces for free-invoke traits instead of just 0 or 1 primary namespace.

Reviewed By: vitaut

Differential Revision: D18113151

fbshipit-source-id: 57f7b76a89e6dbd88cc2584d72079fa7fd94fe53

Summary:
Using `shape[0]` as the length of the IOBuf means using the number of _elements_ along the first dimension, rather than the total number of bytes in the memoryview. Thus
- if the element size is not 1 byte, this will get only a fraction of the memory;
- if the memoryview is shapeless, the code will segfault;
- if the memoryview is multidimensional, it will only get a fraction of the memory (and, if the memory is C-contiguous, it will be an odd slice: in a 2x3 memoryview, it will get the entire first row and half of the second).

Perhaps the above was intentional, but I believe that when converting a shaped and typed object to an unshaped and untyped buffer, it's more intuitive to ignore the type and size and just look at the memory backing it.

I also added a check to ensure the view is contiguous. IOBufs can't represent a non-contiguous buffer, except by using a chain, which however in this case could have an arbitrary unbounded length, making it unadvisable. Alternatively we could copy the memory, but this is probably not what the user would expect. Disallowing this seems the safest option.

Reviewed By: yfeldblum

Differential Revision: D18006273

fbshipit-source-id: f21dbb462266eee68f1a5ce0ecea85b1a46e0a8b

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/1248

Reviewed By: yfeldblum

Differential Revision: D18094432

Pulled By: Orvid

fbshipit-source-id: 115afedc62be8881941cfb4d0e02748be349763f

Summary:
create a unorderedReduceSemiFuture method that returns a SemiFuture
instead of just a Future.

Reviewed By: mhlakhani

Differential Revision: D18248163

fbshipit-source-id: ba1f8ede93c95ea0b877ac306056e7e38c4452e6

Summary:
Build folly with -Werror and -Wunused-exception-parameter to prevent
them from sneaking back in.

Reviewed By: yfeldblum

Differential Revision: D18232483

fbshipit-source-id: 7154d3dfec6c672aa3aa46299456bd3c4d757709

Summary: ^

Reviewed By: mjoras

Differential Revision: D18256970

fbshipit-source-id: 4d07b961ab5ab145b620d9e10fa67116236bbf80

Summary:
Only do RAND_poll on older versions of OpenSSL.

RAND_poll is not required on OpenSSL 1.1.0+, as the RNG is implicitly seeded. On OpenSSL 1.1.1., RAND_poll became aliased to RAND_add. With some custom RNG implementations that did not implement RAND_add, (e.g. python-cryptography's osrandom implementation), this would manifest as a failure.

Reviewed By: ngoyal, yfeldblum

Differential Revision: D16786271

fbshipit-source-id: 4fb7876525006517c3e554d70736d7071bd3b4e3

Summary: The functions in QMS are all marked as const already.

Reviewed By: philippv

Differential Revision: D18240371

fbshipit-source-id: 113f5c968207f0a4ff9b780cd43de43cb90f758b

Summary: Floating-point scale conversion doesn't require a cast

Reviewed By: yfeldblum

Differential Revision: D18052505

fbshipit-source-id: 3d87a916da5cec03b23041908066a9862fab6fde

Summary:
Minor improvements to the CompressionContextPool. Although it should be
impossible, this diff checks that we do not retrieve `nullptr`s out of the
stack.

I also threw in a quick refactor to not store a copy of the deleter in the
RTPDeleter.

Reviewed By: terrelln

Differential Revision: D18131791

fbshipit-source-id: eda1856720fdb86a69ee0a2a867b25bb20ebc319