Summary: Replace inclusions of `folly/futures/helpers.h` with `folly/futures/Future.h` to avoid the cyclic include trap.

Differential Revision: D15600549

fbshipit-source-id: 19950be24a7437fb1fbec293e24058adf17343ca

Summary:
We sometimes see some flakeyness in our internal CI, so we can live
without it there.

Reviewed By: pkaush

Differential Revision: D15695124

fbshipit-source-id: 1d76ae89e245d9c95937e8818826c544c0ae8fc3

Summary:
Currently, AsyncSocket's `connect` API creates new socket, set various
options, bind it to passed argument address and used it for connecting to
server.

While most of the things are handled inside that single connect routine; sometimes
if the caller would like to do few more additional operations before actual
connect happens; caller would like to have hold on the underneath used socket.

In order to do that additional operation on the socket, this diff provides the
facility to provide custom pre-connect hook `preConnect(NetworkSocket fd)` in
the `ConnectCallback` which will be invoked just before the actual server
connect.

Reviewed By: yfeldblum

Differential Revision: D15282727

fbshipit-source-id: 5beac55c77a9537ee4f26856dd5f78c5224a1f49

Summary: Input and output types for the cast are the same (except const qualifier) and therefore it's not required.

Reviewed By: milend

Differential Revision: D15694981

fbshipit-source-id: d5a811d97cc145ba9c232b79f6c8fa52041b0cf0

Summary:
[Folly] `XLOG_EVERY_N_THREAD`.

Implemented with a single per-thread map for all `XLOG_EVERY_N_THREAD` collectively. This approach has better TLS overhad than a `thread_local` counter per `XLOG_EVERY_N_THREAD`, at the cost of integer-keyed map lookups.

While the `thread_local` counter approach is twice as fast as this approach for a noop log-handler, it is expected that the runtime overhead of real log-handlers will make this type of saving immaterial.

Reviewed By: simpkins, luciang

Differential Revision: D15293529

fbshipit-source-id: 7af535a0d329473126a0e4dcffc19d750e8b8b10

Summary:
Lifo sem exposes allocateNode and tryWaitOrPush as protected which allows inherited classes custom Node implementation, creation and enqueueing.
tryRemoveNode was private which prevented inherited classes from dequeueing a Node without signaling shutdown.
This diff fixes it

Reviewed By: shixiao

Differential Revision: D15647287

fbshipit-source-id: 5b26a9616f6de559224f9d8c38230780c78a7cf5

Summary: [Folly] Let `KeepAlive::copy()` allow dummy executors to expire, and not attempt to dereference them to perform an acquire - which, as is known in advance, would fail and return false were the executor still to be alive.

Reviewed By: andrewcox

Differential Revision: D15683240

fbshipit-source-id: e2026493af6936b8f0f1b8fe3c1cb97d2bf0f96e

Summary: [Folly] Split KeepAlive concepts of dummy and alias. A dummy KeepAlive is one for an executor which does not actually support keep-alive semantics. An alias KeepAlive is one for which there is another KeepAlive, with a surrounding lifetime, to the same executor.

Reviewed By: andrewcox

Differential Revision: D15683241

fbshipit-source-id: a5809b06c90ed4a655a6973fac67137b5e1981dc

Summary:
[Folly] Avoid gcc7 lambda-captures bug in `SerialExecutor` test.

```
folly/executors/test/SerialExecutorTest.cpp: In lambda function:
folly/executors/test/SerialExecutorTest.cpp:66:40: error: declaration of 'i' shadows a previous local [-Werror=shadow]
folly/executors/test/SerialExecutorTest.cpp:60:12: note: shadowed declaration is here
folly/executors/test/SerialExecutorTest.cpp: In function 'void simpleTest(const std::shared_ptr<folly::Executor>&)':
folly/executors/test/SerialExecutorTest.cpp:74:45: error: already captured 'i' in lambda expression [-Werror]
folly/executors/test/SerialExecutorTest.cpp: In lambda function:
folly/executors/test/SerialExecutorTest.cpp:74:62: error: 'i' is not captured
folly/executors/test/SerialExecutorTest.cpp:74:46: note: the lambda has no capture-default
folly/executors/test/SerialExecutorTest.cpp:60:12: note: 'int i' declared here
```

Fixes #1154.

Reviewed By: andriigrynenko

Differential Revision: D15629657

fbshipit-source-id: fec82e9d54b2c79a7f54f366ec8a19ad7dc198e9

Summary:
- Cut `FOLLY_HAVE_COMPARE_EQUIVALENT` as this is `1` for Clang 3.4
  onward (regardless of `libstdc++` or `libc++`) and also `1` for GCC
  versions we support.
Pull Request resolved: https://github.com/facebook/folly/pull/1157

Reviewed By: Orvid

Differential Revision: D15641688

Pulled By: yfeldblum

fbshipit-source-id: 0602913b375a406cce6337e43ba708362a94d444

Summary:
This is towards getting open source FBOSS to build using fbcode_builder.
libmnl is one of the dependencies for FBOSS. This patch adds a manifest file
to build the specific version of libmnl for FBOSS.

Reviewed By: wez

Differential Revision: D15633176

fbshipit-source-id: dd51997ae950f53842711fbf2d044e84cdf437e7

Summary:
This is towards getting open source FBOSS to build using fbcode_builder.
OpenNSL is one of the dependencies for FBOSS.

OpenNSL is github repository with headers and prebuilt opaque library. Thus, we
can't/don't need to build it. Thus, OpenNSL manifest uses NopBuilder.

fbcode_builder expects the library to be in installed/lib and headers to be in
installed/include, but OpenNSL lib structure is different.  Thus, add explicit
[install.fields] rule for bin/wedge => lib. Once a rule(s) is provided, we must
provide rule(s) for all, so add rule for include => include as well.

Reviewed By: wez

Differential Revision: D15593639

fbshipit-source-id: facb36ff7b9a31f5952ecbc1a9fdb834c0e68d11

Summary:
This is towards getting open source FBOSS to build using fbcode_builder.
iproute2 is one of the dependencies for FBOSS. This patch adds a manifest file
to build the specific version of iproute2 needed for FBOSS.

Additionally, the default git clone depth of 100 is insufficient for the
version of iproute2 FBOSS depends on. Thus, this patch extends the git SCHEMA
to add optional argument depth. The default remains 100.

The usual /configure --prefix does not work for iproute2. Thus, we need to add
a custom builder that:
  - copies sources to build directory, builds, and
  - installs to installed directory using DEST_DIR.
  - it must also explicitly copy include from build dir to install dir

Reviewed By: wez

Differential Revision: D15588809

fbshipit-source-id: ac5eab24134e078d88b85b4be433c78b05ef8ce5

Summary:
[Folly] Use std interface in fibers mutex types: `try_lock_for`, `lock_shared`, etc.

This makes fibers mutex types play well with `Synchronized`.

Reviewed By: mnv104

Differential Revision: D15440755

fbshipit-source-id: 068b0c0b973577729f76f78c9063ff2ec06f4d90

Summary:
Create pipe-adapters ("pipe fittings") that allow a `|`-based pipeline mixing {std-containers and/or range-v3} with folly::gen, e.g.,

```
auto result = myVec              // anything consistent with range-v3 or these adapters
    | ranges::view::filter(...)  // zero-or-more range-v3 pipe-elems
    | <adapter-goes-here>        // <==**one of the pipe-adapters provided by this task**
    | folly::gen::blah();        // trailing pipe-elems are folly::gen
```

This diff supplies only adapters that transition from {std-containers and/or range-v3} to folly::gen, but not the other way around.

Q&A:

* Why distinguish containers from range-v3's? E.g., container_to_gen vs. rangev3_to_gen?
    * Containers and range-v3's have different copy-costs: range-v3's have O(1) copy
    * Using two different names lets us separate the implementations so we can know which can be copied without significant overhead.
* Why `#include` range-v3 from inside folly::gen? Why the dependency?
    * That `#include` / dependency adds value to the process, and it is included only for the client-files that need it.
    * It adds value to the process since the subset of client-files to be migrated are precisely the same subset of client-files that _need_ to include range-v3.
    * The alternative would be to add the `#include` out in all the various client-files during the migration process, and we will do that as the last step in the migration process.
    * The migration process is sped up by including range-v3 for that specific subset of client-files _and_ it hurts no one since only those who need that `#include` actually get it.
    * Note: we limit the `#include` to the subset of files to be migrated via the `FOLLY_USE_RANGEV3` define; see folly-config.h.

Reviewed By: yfeldblum

Differential Revision: D15035294

fbshipit-source-id: 694979850d1f35dd382e3afde792ea51a2397af0

Summary:
Change the EventBaseTest code to request precise timing from libevent.
libevent 2.1+ uses a coarse monotonic clock by default on Linux.
This causes several of folly's timing-related tests for EventBase to fail most
of the time.

This changes the tests to request precise timing from libevent2, using the
`EVENT_PRECISE_TIMER`.  While we can request this programmatically, doing so
portably across libevent versions is awkward, so the environment variable is
the easiest solution for now.

Alternatively we could change the tests to use larger, coarser timeouts to
make them pass reliably.  However, this would make the tests slower.

Differential Revision: D15528322

fbshipit-source-id: 20dc55de36235e63910cf7fb5b67c6f2d509cc15

Summary: Put file-local classes and helper functions into an unnamed namespace.

Reviewed By: yfeldblum

Differential Revision: D15528323

fbshipit-source-id: 9e90aafe4382d9ba237823fea1e68dbea7fa1f2f

Summary: [Folly] Fix references to `TTransportException` in `folly/io/async/` to refer to `AsyncSocketException`.

Reviewed By: jmswen, knekritz

Differential Revision: D15628173

fbshipit-source-id: f6d36ede5e8b33d9407751733bcc9a7fa764815c

Summary:
- Clang 8 warns about appending integers to a string using `operator+`
  without a cast.

```
../folly/test/FixedStringTest.cpp:353:23: warning: adding 'unsigned int' to a string does not append to the string [-Wstring-plus-int]
  a.append("X world!" + 2u, 5u);
           ~~~~~~~~~~~^~~~
../folly/test/FixedStringTest.cpp:353:23: note: use array indexing to silence this warning
  a.append("X world!" + 2u, 5u);
                      ^
           &          [   ]
../folly/test/FixedStringTest.cpp:354:23: warning: adding 'unsigned int' to a string does not append to the string [-Wstring-plus-int]
  a.append("X world!" + 7u);
           ~~~~~~~~~~~^~~~
../folly/test/FixedStringTest.cpp:354:23: note: use array indexing to silence this warning
  a.append("X world!" + 7u);
                      ^
           &          [   ]
../folly/test/FixedStringTest.cpp:365:23: warning: adding 'unsigned int' to a string does not append to the string [-Wstring-plus-int]
  a.append("X world!" + 2u, 5u);
           ~~~~~~~~~~~^~~~
../folly/test/FixedStringTest.cpp:365:23: note: use array indexing to silence this warning
  a.append("X world!" + 2u, 5u);
                      ^
           &          [   ]
../folly/test/FixedStringTest.cpp:366:23: warning: adding 'unsigned int' to a string does not append to the string [-Wstring-plus-int]
  a.append("X world!" + 7u);
           ~~~~~~~~~~~^~~~
../folly/test/FixedStringTest.cpp:366:23: note: use array indexing to silence this warning
  a.append("X world!" + 7u);
                      ^
           &          [   ]
```

- Fix this warning by creating a local char[] and using that to append
  to the fixed string
Pull Request resolved: https://github.com/facebook/folly/pull/1148

Reviewed By: ericniebler

Differential Revision: D15618465

Pulled By: yfeldblum

fbshipit-source-id: 7f72b3597f51d99665da85744aeb8805eb2e8f00

Summary: [Folly] Fix `PriorityUnboundedBlockingQueue` priority translation from `Executor` priorities to offsets as used by `PriorityUnboundedBlockingQueueSet`.

Reviewed By: andriigrynenko

Differential Revision: D15619435

fbshipit-source-id: 1c5d4d68dfde7f9db07d64f40e73717b48a05ed1

Summary: If performRead is called with bufLen > 2GB then overflow occurred with the value passed to SSL_read.  We'll clamp it here. This should be fine since performRead will be called again as there is more data to read.

Reviewed By: andriigrynenko, knekritz, mingtaoy

Differential Revision: D15625000

fbshipit-source-id: 8633686a76baebd796139f3b20c6f7286339fd96

Summary: Exposed by D15610738.

Reviewed By: ngoyal

Differential Revision: D15622920

fbshipit-source-id: f3058c69098ae740d43f27e8161e99b3c5a6b995

Summary:
The main change is to grab a wlock in doSetContextData, which enables to make the code more readable, and (insignificantly) more efficient.

Grabbing the wlock directly is also strictly better for both `set` and `override`.
`setIfAbsent` is the only one to potentially suffers from the lock change, in the case where it already exists. But `setIfAbsent` isn't used by either of the guards, which are the recommended way of changing values in RequestContext. So it seems incorrect to optimize for it, and overkill to fork the code.

Reviewed By: yfeldblum

Differential Revision: D15604768

fbshipit-source-id: 44f564b09ff50e8bfe0e1c4cf5ee2d77b654e929

Summary: A very basic function that if done incorrectly will get optimized away as undefined behavior. For now keep things simple and only support addition of unsigned values.

Reviewed By: yfeldblum

Differential Revision: D15609542

fbshipit-source-id: 70d5fde784f57d3d52e6122352a885d7a835f104

Summary: [Folly] Fix SerialExecutor task destruction order where tasks are destroyed outside of the request-context scope.

Reviewed By: andriigrynenko

Differential Revision: D15618395

fbshipit-source-id: e2ec8e42137d14f549f92dbc96a904eadcc0a587

Summary:
A couple of changes to support building folly in MSVC/win32 for x86 and arm platforms.
Pull Request resolved: https://github.com/facebook/folly/pull/1147

Reviewed By: Orvid

Differential Revision: D15592309

Pulled By: yfeldblum

fbshipit-source-id: 97860ab9309e5492bfadac56079af735741d4a2c

Reviewed By: yfeldblum

Differential Revision: D15610738

fbshipit-source-id: 367919db6465536e215735400652d33290930ca0

Summary:
There are two variants of `strerror_r` function, one returns
`int`, and another returns `char*`. Selecting proper version using
preprocessor macros portably is extremely hard.

For example, on Android function signature depends on `__USE_GNU` and
`__ANDROID_API__` macros (https://git.io/fjBBE).

So we are using C++ overloading trick: we pass a pointer of
`strerror_r` to `invoke_strerror_r` function, and C++ compiler
selects proper function.

Reviewed By: yfeldblum

Differential Revision: D15484722

fbshipit-source-id: fc0bd45fa67dc401631ffc185c9fad231e4a48a0

Summary: Title, unecessary cost in case of conflict

Reviewed By: yfeldblum

Differential Revision: D15604536

fbshipit-source-id: ad40c84b822a96a688f810b24b539ac162fdcc59

Summary:
As part of trying to get our internal usage of folly off a fork, we have various compiler warnings set ridiculously high.

This gets rid of a couple of warnings that we hit in folly code.
Pull Request resolved: https://github.com/facebook/folly/pull/1145

Reviewed By: LeeHowes, Orvid

Differential Revision: D15586209

Pulled By: yfeldblum

fbshipit-source-id: dc3ebffbddd0b82f2c2ee719a95b7fcfe551e074

Summary: swap the order of getLocalAddress() and getPeerAddress(), so we are more likely to log the local address after exception happens.

Reviewed By: knekritz

Differential Revision: D15559814

fbshipit-source-id: 50b7a964d6c3add2e4ae1da9fde8e9e4e67e5715

Summary:
Grabbing the mutex of the newly created context is unecessary cost.
folly::Synchronized's constructor support this optimization, so it's an easy change.

Reviewed By: yfeldblum

Differential Revision: D15600165

fbshipit-source-id: e92570a7ac7ac1e908d7c797bed6de22f940d387

Summary:
The shallow copy code path currently iterates and copies all values manually. Instead of benefiting from the underlying data collection's implementation of self-copy.
At the very least, it's simpler looking code.

Inheriting unique_ptr is usually discouraged, but in this specific (internal impl detail) use case, I can't foresee any risk.

Reviewed By: yfeldblum

Differential Revision: D15599260

fbshipit-source-id: 73edea27a27e8fbc8b057cf3e974ee3d83a6e760

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/1141

Reviewed By: LeeHowes

Differential Revision: D15560029

Pulled By: yfeldblum

fbshipit-source-id: 5f0d5ee8d09be834d1216d4c39244e0855d2dc0f

Summary:
- Old versions of GCC (specifically, GCC 4.7) warn about uninitialized memory
  in `Expected.h`
- Newer versions of GCC do not complain, so remove the pragma push/pop which
  was disabling the warning.
Pull Request resolved: https://github.com/facebook/folly/pull/1143

Reviewed By: ericniebler

Differential Revision: D15586279

Pulled By: yfeldblum

fbshipit-source-id: 7b01934c7601ecf035f1e05f2ed3e5f799cf25b9

Summary:
This can be used to check if ChaCha20 is available for use.
Signed-off-by: Rosen Penev <rosenp@gmail.com>

related: https://github.com/facebookincubator/fizz/pull/27
Pull Request resolved: https://github.com/facebook/folly/pull/1146

Reviewed By: knekritz

Differential Revision: D15583706

Pulled By: yfeldblum

fbshipit-source-id: 1195d623098018992811cd7d6cef72e2366ace19

Summary: [Folly] Fix a regression in per-task allocations in `SerialExecutor`.

Reviewed By: andriigrynenko

Differential Revision: D15589007

fbshipit-source-id: 4461798435833cb317b41cfea932762eb7ac8662

Summary: [Folly] Cut outdated comment in `F14Table`, referring to the return value of a function which returns `void`.

Reviewed By: nbronson

Differential Revision: D15586980

fbshipit-source-id: b75f7b989c39b52e589ce8051663085201939433

Summary: [Folly] Move `SerialExecutor` test-specific helper types into the tests which need them.

Reviewed By: andriigrynenko

Differential Revision: D15587721

fbshipit-source-id: 603e5381b2e00b4317f7e9836c6102238158c5d0

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/1142

Reviewed By: shixiao

Differential Revision: D15574540

Pulled By: yfeldblum

fbshipit-source-id: 2ce91f7510f0b278677c15b3a9f1f80b43ff8998