Summary: socketConnect() is not invoked if we have TFO enabled. Need to explicitly disable it once new socket is created, and noTransparentTls flag is set.

Reviewed By: djwatson

Differential Revision: D7945457

fbshipit-source-id: 739b7ae5bc146b50255254e644fba2618147c4f4

Summary:
Enforce two aspects of folly::futures::Core<T>:

- that it is heap-constructed (prevent clients from constructing it other than via `new`)
- that it cannot be deleted early (prevent clients from explicitly deleting it)

Reviewed By: yfeldblum

Differential Revision: D7900761

fbshipit-source-id: c1c5a149e3d96c4cea04c17909c2ffd42383973e

Summary: It was spamming the MSVC build log with warnings.

Reviewed By: magedm

Differential Revision: D7954677

fbshipit-source-id: 1f2d1163483518382543aacb7e8fbc17e291e650

Summary:
clang whines about these to varying degrees. They seem like
braindead fixes.

Reviewed By: yfeldblum

Differential Revision: D7918701

fbshipit-source-id: a5aae179c2df89ef52148a500e95cd1cd25a1f0b

Summary:
There is a subtle race between thread destruction and task addition (ensureActiveThreads).

For an executor with a single thread, the race would be:
threadid
1 cpu thread returns from try_take_for by timing out.
2 someone from a different thread calls add()
2 add() finds lifosem empty, calls ensureActiveThreads
2 ensureActiveThreads grabs the lock, finds active=1, total=1, returns doing nothing, assuming thread is still running.
1 cpu thread decrements activeThreads_ to 0 in taskShouldStop.

There are now *no* threads running until the next call to add(), and one task waiting.

Fix: Grab lock in taskShouldStop (like the documentation says we should be doing).  Also double check that there are no currently pending tasks.

This probably only affects pools with size of 1, otherwise we would have added a second thread and there may temporarily be one fewer threads running.

Reviewed By: davidtgoldblatt

Differential Revision: D7943241

fbshipit-source-id: 502e5809ccf4ecca85205c14c4d97b508897de9b

Summary:
SerialExecutor used to have public constructors and could be
constructed on the heap or on the stack. In order to be able
to reference the shared stater later from the parent executor,
the task queue was implemented in a separate class, of which
the SerialExecutor would create one, managed by a `shared_ptr`.

With the recent move towards `folly::Executor::KeepAlive`, the
SerialExecutor itself always has to be constructed on the heap
(hence no public constructor anymore but factory functions).
However, that means that we can embed the task queue implementation
in SerialExecutor, using `KeepAlive` objects to access it from
the parent executor.

This change simplifies the implementation of SerialExecutor and
reduces the number of heap objects and thus memory allocations.

Reviewed By: andriigrynenko

Differential Revision: D7927260

fbshipit-source-id: d3b6a2376bfdf512a946f2f45c730ed68176a441

Summary: This turned out to only be useful in tests. Clean up the tests and get rid of it

Reviewed By: yfeldblum

Differential Revision: D7907221

fbshipit-source-id: f5e0263c1426e1af97c190e6b2bd47d7db37d403

Summary: When cleaning up the interface with r-value qualification for SemiFuture it seems we missed deferError.

Reviewed By: yfeldblum

Differential Revision: D7942259

fbshipit-source-id: c0cf05eb2b6357e6f6eb3fab6aef87ec6416caf7

Reviewed By: saifhhasan

Differential Revision: D7943135

fbshipit-source-id: 6b66b7b2ea6343b99ec113febca991eb840b1860

Summary: gcc-5.x does not allow brace elision in these std::array initializer lists.

Reviewed By: yfeldblum, mnv104

Differential Revision: D7923970

fbshipit-source-id: 807f2e0d376c9ac4f4f6988fb811590531b6334d

Summary:
Running the xlog test binary stand-alone revealed a bug that
was otherwise masked when running with buck test. The bug was that
2 tests ended up depending on the log-level associated with a default
category and would thus fail since the tests did not clean up
the log-level setting associated with those categories.
Attached diff fixes the issue by restoring the category's log-level in
a scope exit block.

Reviewed By: simpkins

Differential Revision: D7922086

fbshipit-source-id: 9d8ed917633ae214dc81075fbd38275160e64376

Summary: [Folly] Simpler SFINAE in variadic `F14BasicSet::emplaceItem`, without using `std::tuple_element_t` unsafely when `sizeof...(Args) == 0`.

Reviewed By: shixiao

Differential Revision: D7920929

fbshipit-source-id: d74d5b36ac683f72e21d83294aa890f49194280d

Summary:
Check if the compiler supports -Winconsistent-missing-override before trying to
disable this warning.

It looks like gcc wouldn't actually fail the build if
`-Wno-inconsistent-missing-override` was specified but was not supported.
However, if the build failed due to other compile errors the error messages
about this flag would make it confusing to identify the actual build failure
reason.

Closes issue #836.

Reviewed By: yfeldblum

Differential Revision: D7920577

fbshipit-source-id: 322444a9df3f4f36bb5420f0720c1e433c2856b7

Summary:
[Folly] Fix `ThreadPoolExecutor::getPendingTaskCount` decl return type to match defn.

Fixes #843.

Reviewed By: djwatson

Differential Revision: D7914071

fbshipit-source-id: a22658b1dd48a5e73fae3a0312f72007230e36c8

Summary:
One (artificial) test case creates ~1000 threads, resulting in lots of spinning contending for this lock.

Changing to folly::SharedMutex fixes the issue, allowing the actual working thread to finish.

Reviewed By: magedm

Differential Revision: D7847705

fbshipit-source-id: b2a8c3acdb4c62f347ef4ec761acbe7681ced9f3

Summary: Allows for conditional logging of messages.

Reviewed By: simpkins

Differential Revision: D7888834

fbshipit-source-id: d1fe47d82fe8889f6b597062f5542e7eb61ed9e6

Summary: Take now as input to improve performance when updating many buffered stats at once

Reviewed By: yfeldblum

Differential Revision: D7906107

fbshipit-source-id: cf09bc2c2673b0a336628b3e6b31ee2d55dbb10c

Summary: [Folly] Fix race condition in collectN where the future may be completed with too few values or by reading the values while they are in-process of being written.

Reviewed By: djwatson

Differential Revision: D7900639

fbshipit-source-id: 3eafa821abda4f93232db3b8114de54c6faf798a

Summary:
This reverts commit 5102e9c3de026c021bf322b5109aafcc115207a2

bypass-lint

An infra SEV is better than not reverting this diff.
If you copy this password, see you in SEV Review!
cause_a_sev_many_files

Differential Revision:
D7840699

Original commit changeset: 5102e9c3de02

fbshipit-source-id: 13e97d5497ac383c0eff18a4b947d5b3d82a19b7

Summary: [Folly] Fix outdated mention of `TSSLSocketFactory`.

Reviewed By: stevegury, knekritz

Differential Revision: D7892382

fbshipit-source-id: 7b267e5de857f9c5976b26d53d36b1545a46af2a

Summary:
[Folly] Use `size_t` for `ThreadPoolExecutor::getPendingTaskCountImpl`, as it is the canonical size type for counting the number of entries in in-memory data structures.

Prevents confusion when using values returned from this function with other sizes, especially in contexts where the two types are required to match as might occur in calls to `std::min` without explicitly specifying the type.

Fixes #843.

Reviewed By: djwatson

Differential Revision: D7898443

fbshipit-source-id: eec789f661351d09545bc7118f2e4b3587fd7302

Summary: Add thread local destroy benchmark

Reviewed By: yfeldblum

Differential Revision: D7892479

fbshipit-source-id: f16462820354a6cf27d9348a76b28be027cef20c

Summary:
SemiFuture::get() & Future::get() both render the `this` object in a consumed / moved-out state. However Future::get()'s postcondition was untestable by callers - it consumed the value within the core but left `this->valid()` true.

Goal: make Future::get():
- function like SemiFuture::get()
- testable by callers (see `valid()`)

Reviewed By: yfeldblum

Differential Revision: D7840699

fbshipit-source-id: 5102e9c3de026c021bf322b5109aafcc115207a2

Summary:
Lets callers determine whether a Future/SemiFuture/Promise has a shared-state.
`x.valid()` (where `x` is a Future, SemiFuture, or Promise) is is true if/only if the object has a shared state.

The `valid()` method will also be key in a near-term diff that will use it to clarify the contracts/rules.

Note: the concept of validity already shows up in the doc-blocks; this gives callers a conceptual oppty to check it.

Reviewed By: yfeldblum

Differential Revision: D7839038

fbshipit-source-id: f8390b3d1a8c3d5d89356ae48937e03c424d1174

Summary:
PIC is implied by BUILD_SHARED_LIBS in CMake, so the test is confusing.
Note that if one wants to build static and use it to create a shared library with it, he will need to compile either shared
or static with CMake option -DCMAKE_POSITION_INDEPENDENT_CODE
Closes https://github.com/facebook/folly/pull/833

Reviewed By: Orvid

Differential Revision: D7890111

Pulled By: yfeldblum

fbshipit-source-id: e3996cb83d30ae368dc5fdd4f55b721f7f89d0a0

Summary:
Add the ability to use folly::window without providing a vector ahead of time.
This allows window to scale to a large number of iterations.

Reviewed By: yfeldblum

Differential Revision: D7863481

fbshipit-source-id: 86c6b7926e9c3b7510233fac20a50c1fb38f5b57

Summary:
Pointer value may change when casting.

(Note: this ignores all push blocking failures!)

Reviewed By: lbrandy

Differential Revision: D7884048

fbshipit-source-id: aa55e1b3530bc28f7d2ffd356761d962d6029832

Summary:
I'm continue to try keeping the implementation simple, instead of copy-pasting EventBase/EventBaseLoopController code.
My understanding is that AsyncioExecutor/AsyncioLoopController will always be single threaded because they are tied to a unique asyncio event_loop, so I don't need an atomic counter.
Then it should be cheap to just always aquire/release the counter in runLoop.

Reviewed By: andriigrynenko

Differential Revision: D7875202

fbshipit-source-id: 900a03e6870c5e4828521a99785cfa976baac54a

Summary:
Copied the `const folly::fibers::FiberManager::Options& opts = {}` signature from main `getFiberManager`.
Discussed with Andrii if I should key the FiberManager cache by options' address, but he voted to keep the behavior consistent with FiberManagerMap.

Reviewed By: andriigrynenko

Differential Revision: D7873868

fbshipit-source-id: 0c7f6e6ef0ebe50d9658c2eee7ab1e098dd86ddd

Summary: [Folly] Fix Build: `folly/io/RecordIO.cpp` under MSVC 2015u3, which inexplicably gives `error C2872: 'detail': ambiguous symbol`.

Reviewed By: Orvid

Differential Revision: D7875032

fbshipit-source-id: 2ff5a60c941e20ef9a05cedb2e7d8224a9e83ed2

Summary: Use a single pointer to store Executor pointer and dummy flag.

Reviewed By: yfeldblum

Differential Revision: D7857087

fbshipit-source-id: 529fbfdc6ad14954f5bdd9bef3eb1e335b233ec9

Summary: It's timing out in open source test runs because each individual test was taking 10 seconds. They each now take 2.

Reviewed By: yfeldblum

Differential Revision: D7841938

fbshipit-source-id: eb81c67cc26e98fb0b4b0c736bea747e985f2fcf

Summary:
`{Future,SemiFuture}::poll()`:
- minor chg to impl from assignment to stateless ?:

Reviewed By: yfeldblum

Differential Revision: D7840466

fbshipit-source-id: 201dd7121b713324bb054b72c8ecdf6f4f163b97

Summary: SerialExecutor is always wrapping another Executor, so we shouldn't force users to coordinate their lifetimes.

Reviewed By: yfeldblum

Differential Revision: D7856146

fbshipit-source-id: ac7caaa0f181406dfb6b59d36ae4efe6d1343590

Summary:
This allows using KeepAlive token as a pointer to an specific Executor sub-type (e.g. SequencedExecutor).
getKeepAliveToken() has to be a free function to achieve this goal.

Reviewed By: yfeldblum

Differential Revision: D7847884

fbshipit-source-id: 95dc62433d30212ceb9b9b2fc7d6b38cb5425311

Summary:
This diff adds dynamic thread creation/destruction to ThreadPoolExecutor (mostly CPU).

1) Threads are lazily created when the first task is add()ed (or getEventBase is called), applies to both IO and CPU

and

2) after a timeout in LifoSem, threads are joined in CPUThreadPoolExecutor.

Most of the logic is in ThreadPoolExecutor.  A separate dynamic IOThreadPoolExecutor will be a follow on diff.

To keep the previous behavior, users can set minthread == maxthreads, or gflags --dynamic_cputhreadpoolexecutor=false

Reviewed By: magedm

Differential Revision: D7477390

fbshipit-source-id: 855a35380861111de33656e8fb1b681ae8cd2807

Summary: Fixes two tiny formatting issues with the folly::Function documentation.

Reviewed By: yfeldblum

Differential Revision: D7856507

fbshipit-source-id: dc82256a35f6c2b1eb66a868f419f2c800f1ddc5

Summary: [Folly] Avoid duplicating throw_exception instances taking string literal, which differ only in the length of the string literal.

Reviewed By: marshallcline

Differential Revision: D7853216

fbshipit-source-id: a4648952cc10661322e5e23c6317ba4b75c268fb

Summary:
This diff optimizes the operation of F14 set emplace when
the argument list happens to be a single reference to a key_type.
An intermediate key_type value is removed in that case, which avoids a
move in all cases and avoids a copy when the emplace finds that the key
is already present in the set.

Reviewed By: yfeldblum

Differential Revision: D7845720

fbshipit-source-id: 16a639f53993d9843ffe265edd58c0d74de0c1f7

Summary:
When used in a QuantileEstimator, some sliding window buckets can be sparse or all together empty. In this case, reserving memory in the constructor and in merge is wasteful. Don't reserve in the constructor (this is a complete waste) and call shrink_to_fit in merge.

It's worth noting that the benchmarks *are* shrinking the digest, because the digests are only 85 or so items big, so it is exercising this path.

Reviewed By: yfeldblum

Differential Revision: D7855242

fbshipit-source-id: 61bcbaaf17dd46f8b98d171af43516ceb8d84510