Summary:
Replace Future::onError with Future::thenError:
 * to remove ambiguous typing
 * to ensure that the executor is not lost and the returned Future is still bound to an executor

This diff finally removes the legacy onError.

See:
https://fb.workplace.com/groups/fbcode/permalink/2002251863144976/
for details.

Reviewed By: yfeldblum, Orvid

Differential Revision: D15034702

fbshipit-source-id: 17faf37e79c0a31ad25c02542f09a8293e693899

Summary:
The folly::coro functionality depends heavily on the symmetric-transfer capability of coroutines which MSVC does not yet support.

The folly/Portability.h header was previously defining FOLLY_HAS_COROUTINES to true for MSVC even though you would get compile errors if you tried to use any of the folly::coro facilities.

Also further restrict defining FOLLY_HAS_COROUTINES only when C++17 or later is available as folly::coro makes use of C++17 language features.

Reviewed By: yfeldblum, andriigrynenko

Differential Revision: D14811921

fbshipit-source-id: f3da2d3c07fc9ebe08f566a7707c62e74e412df9

Summary:
Remove legacy logging API (`XLOGC` and friends) because it is not used anywhere
and is not fully safe being varargs-based. The code that needs to be converted
from `printf`-based logging can use fmt's `printf` API which is fully type-safe.

Reviewed By: simpkins

Differential Revision: D15036094

fbshipit-source-id: 3919e5d4eb7994ed0db4fa47bada111ab0dca505

Summary: temp

Reviewed By: yfeldblum

Differential Revision: D14882681

fbshipit-source-id: 3017cb9743c81396764e6a61bdca183ae2692401

Summary:
- Most of `IntToDouble` test is commented out as it previously did not
  work with GCC 4.6.1.
- Enable the rest of this test as it works fine on GCC 5.1
Pull Request resolved: https://github.com/facebook/folly/pull/1112

Reviewed By: yfeldblum

Differential Revision: D15014999

Pulled By: Orvid

fbshipit-source-id: a26a0fcacc9a22c03d0d67c35cabf8ec0ca91813

Summary: The only fix in this diff to an error currently being seen by the contbuilds is the change to concurrent hash map. The rest would have been seen when building the slow tests is enabled.

Reviewed By: yfeldblum

Differential Revision: D14898822

fbshipit-source-id: b28cd697496bb97b0e35b189a9bb31b666da3b17

Summary:
- Both GCC and Clang define the macro `__GNUC__`, `__GNUG__` and
  friends. Simplify some macros whose intent is to check "if we are GCC or
  Clang" but do so via checking `__GNUC__` in addition to `__clang__`. It
  is sufficient to just check `__GNUC__`.
Pull Request resolved: https://github.com/facebook/folly/pull/1095

Reviewed By: Orvid

Differential Revision: D14697480

Pulled By: yfeldblum

fbshipit-source-id: cd6ec82c4235a1321ecdf240afaad7d5c0e99437

Summary: [Folly] Consistent `void` return-type checks in `DistributedMutex`. `const void` is a different type from `void`, but both match the `std::is_void` predicate.

Reviewed By: aary

Differential Revision: D14638815

fbshipit-source-id: 96d1e18dfe1026b3881dafe5ceb816a8dd3add08

Summary: Since C++17 `vector::emplace_back()` returns a reference to the new element, support that in `small_vector` too.

Reviewed By: terrelln, luciang

Differential Revision: D15007059

fbshipit-source-id: 3873fd0097e3d50f7967f47fbecfa6ab20cc72c5

Summary: The virtual functions marked `protected` in `ThreadPoolExecutor` should be marked `protected` in `EDFThreadPoolExecutor`.

Reviewed By: yfeldblum

Differential Revision: D15008767

fbshipit-source-id: 6bad51068a8b64f623643e0d0deb684ed6e2be22

Summary: as in title

Differential Revision: D15005737

fbshipit-source-id: 6ac0ff3d7498e505f0437dd5be76735d66eb953e

Summary:
Remove getSelfCert and getPeerCert APIs from
AsyncTransport.

This removes an explicit dependency on the X509 type
from openssl

Reviewed By: yfeldblum

Differential Revision: D14970297

fbshipit-source-id: b727789d3a37a57d4728709ab66a575d40176172

Summary:
- Older versions of GCC required defining private member function
  `emplace_back_aux` out-of-line. That is, declare the member in-class and
  then provide the definition fully out of line.
- These older versions of GCC are not supported anymore, so just define
  it inline as you would any other member function.
Pull Request resolved: https://github.com/facebook/folly/pull/1077

Reviewed By: meyering

Differential Revision: D14591465

Pulled By: yfeldblum

fbshipit-source-id: 0ebfe55cec4e230c36c98ad595ec8dcfc98a211e

Summary: StorageTriviallyDestructible needed a constexpr ctor.

Reviewed By: yfeldblum, ot, luciang

Differential Revision: D14944492

fbshipit-source-id: cd9d39d7cb0719e7830887f548eabcc0defa1aee

Summary:
Update the folly spec to use `fb_github_project_workdir()` rather than just
`github_project_workdir()`.

This shouldn't have any impact for external github-based builds, but for our
internal continuous integration builds this should make builds for downstream
projects pick up the correct local version of folly that matches the commit of
the downstream project that is being built.

This behavior was unintentionally changed to not use the `fb_*` version in
D10520960.

Differential Revision: D14969380

fbshipit-source-id: 000b10525821fb5e0312d1869198a68befa0bd91

Summary: The current order of retire and reclaim_nodes in clear() introduces a subtle bug which allows the hazptr library to destroy the Chunks array from reclaim_nodes, freeing the memory at chunks_, then continue reclaiming nodes which involves further access to chunks_.`

Reviewed By: magedm

Differential Revision: D14968860

fbshipit-source-id: 1bc8e543ae54c3d0e26affb1fd352c58cc5ce897

Summary:
Recently, IOBufQueue started "packing" smaller IOBufs (i.e. <= 4KB) when serializing.
This was helpful to avoid unnecessarily IOBuf chaining.

That said, we did not update our serializedSizeZC() functions to emit an accurate
memory estimate in the case that we did need to pack the IOBuf data. This changes
to pass though the IOBuf size, if under the threshold.

Also, add a few basic comments in the Protocol implementations about what
serializedSize() means, because it's not entirely obvious to a new user that
the results aren't intended to be accurate. And document what "ZC" is intended for.

Reviewed By: yfeldblum

Differential Revision: D14919060

fbshipit-source-id: 18e9bd0c6b9e40ccd1006cffba36363c6f95ea41

Summary: [Folly] Fix `StaticSingletonManagerSansRaii` cache.

Reviewed By: aary

Differential Revision: D14928194

fbshipit-source-id: 6800c96ff40ec35e18a06b9c6a9f150a50695e37

Summary:
The uint32_t futex specialization for atomic_notify_all didn't remove
the Integer template parameter, causing calls with atomic<uint32_t> to not
specialize properly.

Reviewed By: yfeldblum

Differential Revision: D14957311

fbshipit-source-id: 063e0f6d18360d7e2ce8dd5bbf5c46c52a6fc833

Summary: change the macros to work on MSVC

Reviewed By: yfeldblum

Differential Revision: D14942488

fbshipit-source-id: 12d6814230bd9b92b54603cd711d03cd8b5f8086

Summary: reduces folly::Executor dependencies

Reviewed By: yfeldblum

Differential Revision: D14939208

fbshipit-source-id: bc354e838608eb8b25373ed5179e72acef813edd

Summary:
fixed some compile errors when building SMC with pushmi as a dependency

some _v traits do not exist

a specialization of ManualLifetime<void> was required

Reviewed By: yfeldblum

Differential Revision: D14939161

fbshipit-source-id: 76a1f05d7ce889d00fdf4a264e5e719fdb12d4ac

Summary: `reference_type` is unused

Reviewed By: yfeldblum

Differential Revision: D14944273

fbshipit-source-id: f4c968735fcb7173e5ee256381455c81492e02df

Summary:
In zstd-1.4.0 we've renamed one function used by folly. Add compatibility
macros to work with both function names.

Reviewed By: Cyan4973

Differential Revision: D14941863

fbshipit-source-id: deaf63b28ac8e4c3e6f6c0c192ed2d0e452ddca7

Summary: Under heavy contention, lock will recurse several times. This can lead to fiber stack overflow.

Differential Revision: D14936384

fbshipit-source-id: e934569a9f742859655ba3ea888ebabd4d959651

Summary: Support for capturing metadata while doing folly json parsing

Reviewed By: yfeldblum

Differential Revision: D13626597

fbshipit-source-id: 8e7a26e23d966240d35c4f6d6104d9334a675a12

Summary: Let `ThreadLocalPtr` member `reset()` synchronize with member type `Accessor` lifetime, preventing data races that could happen if `reset()` is called while another thread is within the `Accessor` critical section.

Reviewed By: yfeldblum

Differential Revision: D14878801

fbshipit-source-id: 4090c5749d08bee3fd9b50b23df6e716bc78c9c7

Summary: add support for flow to `on()` - requested by Songqiao

Reviewed By: ericniebler

Differential Revision: D14814588

fbshipit-source-id: 3989d0aa554a32482dcc15c1ba706d5604c95eb8

Summary: Fixes bug in transform reported by rhodo

Reviewed By: ericniebler

Differential Revision: D14814574

fbshipit-source-id: fdc3866ad56ced11e858869a24ca95b08295477d

Summary: The Impl param is last so its a pain to switch between BucketTable and SIMDTable in the common case where only KeyType and ValueType are specified. This diff enables ConcurrentHashMapSIMD<key, value>.

Reviewed By: djwatson

Differential Revision: D14916671

fbshipit-source-id: 980518179b87316ae344cbf96be4e6bcc032424e

Reviewed By: yfeldblum

Differential Revision: D14813810

fbshipit-source-id: b6294f0b2e52bd062ad7b1d5977fbe68dec82d1b

Summary: [Folly] Use `tag_t` instead of `TypeTuple` for singletons.

Reviewed By: andriigrynenko

Differential Revision: D14927424

fbshipit-source-id: fdd33a30d5abde67c50df009a0cacc89462278b7

Summary:
`moveObjectsRightAndCreate()` did pointer arithmetic outside of the object,
because it had a one before the beginning pointer. Instead, decrement the
pointers before use.

Reviewed By: ot

Differential Revision: D14887926

fbshipit-source-id: 198151d1eaef17e86cda800367480e0a8f7b2f49

Summary:
Instead of default constructing in `moveObjectsRight`, insert the objects
directly into the hole. This has the added benefit of avoid a default
construction. The creation function has to be careful to return references
(rvalue, or lvalue) to avoid extra constructions without guaranteed NRVO.

I've constructed the elements in reverse order since it is strange to copy
backwards, then construct forwards, but don't have strong opinions, since
either way is correct.

Reviewed By: ot

Differential Revision: D14883034

fbshipit-source-id: bc0d19bf94705718500b80e2d6c202784134e7a5

Summary:
In order to set the verified chain of a X509 store we need to be able to use X509_STORE_CTX_set0_verified_chain both for openssl 1.1.0 and previous versions.

Proxygen needs this to complete the clang/platform007 upgrade.

Reviewed By: Orvid

Differential Revision: D14894384

fbshipit-source-id: 378eb834b077469b666ad303f28f54af5caff990

Summary:
This ensures that we can use any tuple type in maps/other places where
hashes are useful.  Since there's only one value for an empty tuple, just
return 0.

Reviewed By: ot

Differential Revision: D14898947

fbshipit-source-id: 8fefab46df32a6cc56e43b02346bbefe4d4ec567

Summary:
Split out enable IPv6 on docker into its own script.
Do not do this on MacOS hosts; where we would not be running in a docker container.
use apt add-on rather; which automatically disabled for non-apt supporting OSes (MacOSX)
Pull Request resolved: https://github.com/facebook/folly/pull/1108

Reviewed By: vitaut

Differential Revision: D14899358

Pulled By: calebmarchent

fbshipit-source-id: 342e7a37e27317018d98d77e8603c896f34b3bc6

Summary:
Hello Facebook,

This small PR to fix the redundant std::move usage detected by gcc (GCC) 9.0.1 20190225 (experimental)

```
/folly/BUILD/folly-2019.04.01.00/folly/executors/task_queue/PriorityLifoSemMPMCQueue.h:92:30: error: redundant move in return statement [-Werror=redundant-move]
   92 |         return std::move(item);
      |                              ^
```

Stac
Pull Request resolved: https://github.com/facebook/folly/pull/1107

Reviewed By: Orvid

Differential Revision: D14874144

Pulled By: yfeldblum

fbshipit-source-id: c65a422c500606103192b3d9377a52b4abf5a653

Summary: This patch introduces an earliest-deadline-first (EDF) executor: `folly/executors/EDFThreadPoolExecutor`.

Reviewed By: interwq

Differential Revision: D13983430

fbshipit-source-id: 0df9bc4a1cdbe9059489d1e1abad88b9e6d17ad9

Summary:
Needs to be enabled by option PYTHON_VENV in the config.

shell_builder.py sets up the venv and uses it once; calling activate

For docker we set ENV; resulting in the virtual environment being present
when the resulting container is run as well as at build time. This is also cleaner
and easier to follow than re-asserting on each RUN step.

For Lego builder we need to source activate on each command as environment
will not persist between commands.

While man on the posts say it makes no sense to use virtualenv within docker
container, this method simplifies the process considerably as we can rely on the
name pip being valid and we don't need to either ensure we are root or pass the
--user flag to pip and setuptools.

Pull Request resolved: https://github.com/facebookincubator/LogDevice/pull/76

Reviewed By: wez

Differential Revision: D14875633

Pulled By: calebmarchent

fbshipit-source-id: aabbcdd509d2a59fa36f8004032a052f014ce1ba