Summary:
[Folly] Remove explicit `throw`, `try`, and `catch` uses from `IOBuf`, preferring scope guards and `throw_exception`.

(Note: this ignores all push blocking failures!)

Differential Revision: D15442544

fbshipit-source-id: 2f182256ff12f5aad6166a7b40ed018c1b32e923

Summary:
[Folly] Mark some more `IOBuf` members as `noexcept` and remove an explicit `abort` which is there to simulate `noexcept`.

(Note: this ignores all push blocking failures!)

Differential Revision: D15442545

fbshipit-source-id: adaef4d1f6c2e551e9c19b151e3f08c6ba11f717

Summary:
- GCC 9.1 warns when moving a local object in a return statement
  prevents copy elision.
- Remove the explicit `std::move` in those return statements.
Pull Request resolved: https://github.com/facebook/folly/pull/1131

Reviewed By: LeeHowes

Differential Revision: D15391413

Pulled By: yfeldblum

fbshipit-source-id: af29ea0f6f65957645fa2debce2f5303906a2507

Summary:
[Folly] Cut `folly/LockTraitsBoost.h`. Also, discourage specializing `LockTraits`.

For ODR sanity, the better mechanism is to wrap mutex types with non-standard interfaces in mutex wrapper types with standard interfaces.

Reviewed By: Orvid

Differential Revision: D15426235

fbshipit-source-id: f54aee8c47951f6214f5b23fce7f8f6c29c8b5c2

Summary: Add lizmq and sigar in prepare for fbzmq oss build migration.

Reviewed By: wez

Differential Revision: D15387435

fbshipit-source-id: 1ec2057972bf801a1598af0a1ae87f958394d812

Summary:
with the reshuffling around getdeps.py, and changes in folly/thrift/wangle,
our internal package stopped building.  This diff brings it back to life.  It is
a little bit interwined:

* Introduce a `--facebook-internal` switch to getdeps that causes `fb` in the
  build context to be set to `on` rather than `off`.  This allows specifying
  options in the manifest that are FB specific.
* Introduce cmake options to control the WATCHMAN_STATE_DIR and whether cmake
  should attempt to create WATCHMAN_STATE_DIR when it is installed.
* For FB specific builds on macOS, set the state dir to match our existing location
  and disable creation of WATCHMAN_STATE_DIR because that will ultimately be
  handling during package installation and we don't have privs for that on
  the machine generating the package.
* Adjust Facebook CI scripts to use getdeps for building and testing watchman

Reviewed By: simpkins

Differential Revision: D15337970

fbshipit-source-id: 8ab3c1f810decf5f0b16337dc1cbe6dc73d26426

Summary: Add a comment that the Dynamic version of MPMCQueue is deprecated and that UnboundedQueue should be used instead.

Reviewed By: yfeldblum

Differential Revision: D15392160

fbshipit-source-id: 5958b7602d3062878dd3e78907cb7091532ad4b8

Summary: Guard pages are cheap (VM only + mapping state). More than one is occasionally useful (e.g. imagine a 10k object on stack that's not used, subsequent function calls will create a frame that jumps over a single 4K guard page and we get hard to debug memory corruption instead of a clean segfault).

Reviewed By: andriigrynenko

Differential Revision: D15367640

fbshipit-source-id: 8f1d42a71926653929dd847eeba172c680f98589

Summary:
The schedule type is used to differentiate between a build run during
code review (and thus runs code that isn't yet in the master branch),
from continuous or other types of run that operate on landed code.

This doesn't change any behavior yet; this diff just adds
the plumbing to pass down an optional arbitrary schedule type string
from the CI system.

In the future, we'll use the schedule type to influence the behavior
of running tests.

Reviewed By: strager

Differential Revision: D15300120

fbshipit-source-id: 3b46afef2ff171b3fa095763dd5006a54ea328b8

Summary:
F14TestUtil's SwapTrackingAlloc's metadata accesses were
triggering a -Wclass-memaccess warning in gcc 9.1 despite being safe. This
diff changes the accesses to use explicit casting to avoid the warning.

Reviewed By: yfeldblum

Differential Revision: D15393963

fbshipit-source-id: 088811354f00d623a58678c3f41a0e8effdb5ce3

Reviewed By: udippant

Differential Revision: D15368849

fbshipit-source-id: 5e101fc38417b807a531fa456b38234d6dd08b7a

Summary: [Folly] Namespace the helper macro in `folly/ssl/OpenSSLPtrTypes.h`.

Reviewed By: Orvid

Differential Revision: D15370971

fbshipit-source-id: aa907f60fee5ef58c5484838a3bde0d68e65b249

Reviewed By: lewissbaker

Differential Revision: D15363378

fbshipit-source-id: 0343d6cc42624a6c3e9b01a338afb1646625928a

Summary: This is not a good default. We can build a dedicated SpinningBaton if needed

Reviewed By: andriigrynenko

Differential Revision: D15387138

fbshipit-source-id: 433ce2b1af7c661edd2c0e264dc618cc9c680aa3

Summary: This diff replaces all 2 occurrences of "NDEBUG" with "folly::kIsDebug" in folly/SharedMutex.h. Compared to NDEBUG, folly::kIsDebug has the equivalent functionally  but helps consolidate the conditional-compilation logic to a single place, namely folly/portability.

Reviewed By: shixiao

Differential Revision: D15379370

fbshipit-source-id: 58c563624e58a4fd18b0cfbe4c0b9ba879f8eea9

Summary:
When the udp socket is connected we shouldn't supply the peer's address. This changes it so that
we can use write() methods in AsyncUDPSocket when the socket is connected.

Differential Revision: D15384587

fbshipit-source-id: ad26e2d04a7303c7ff9e25f32422d7c56e4283e5

Reviewed By: andriigrynenko

Differential Revision: D15366589

fbshipit-source-id: e8d1c17e158437a3ca882f51f12071bf666c1a6e

Summary:
if we find ccache in the path, then we instruct cmake to use
it as the compiler launcher.

Reviewed By: pkaush

Differential Revision: D15375441

fbshipit-source-id: 602fe54742a5ec07b0533bd2cf63cb905b051e85

Summary:
Adds a benchmark to show the benefit of running work inline.

Dev:
| fourThensOnThread | 2.15% | 245.01us | 4.08K |
| fourThensOnThreadInline | 2.36% | 223.51us | 4.47K |
| hundredThensOnThread | 0.67% | 783.54us | 1.28K |
| hundredThensOnThreadInline | 0.88% |  595.24us | 1.68K |

Opt:
| fourThensOnThread                                |   0.18% |    70.37us  |  14.21K |
| fourThensOnThreadInline                          |   0.26%  |   50.11us   | 19.96K |
| hundredThensOnThread            | 0.18% | 72.54us | 13.79K |
| hundredThensOnThreadInline | 0.21%  | 64.79us | 15.43K |

Reviewed By: yfeldblum

Differential Revision: D15319231

fbshipit-source-id: 978c26538d89de15e76eb69bf20c1129b210a442

Summary:
We were computing `SCRATCH/install` and hashing based on that value,
but the build stuff was later computing `SCRATCH/installed` and passing that
to the builders.

Fixup the mismatch.

Reviewed By: simpkins

Differential Revision: D15337969

fbshipit-source-id: 70288f2d9286aaacf4c1f6e0dac4680a55edac6d

Summary:
Adds inline-supporting forms of thenValue and thenTry. Carries this information in the Core through a variant of the HasCallback state. Dispatches the continuation inline rather than on the executor if that state is set.

Propagate completing executor through intermediate promise and make inline execution conditional on matching executors.

Only continue inline if the executor of the completing task matches that of the task-to-be-enqueued, to ensure that calls to via always trigger reposting and that there are no surprises where work runs on the wrong executor.

Reviewed By: yfeldblum

Differential Revision: D15292236

fbshipit-source-id: 51ddd5cfd1196ea9e9f408e2b1337ff9d4fa5b9f

Summary: [Folly] Extract the unique-instance enforcer from `SingletonThreadLocal` into a standalone `UniqueInstance`.

Reviewed By: andriigrynenko

Differential Revision: D14927839

fbshipit-source-id: cfe98ef7a88746a557a64263f3165cd01caeec37

Summary: [Folly] Cut `presorted` and `unsorted` tags; the `presorted` tag has  ambiguous meaning and is replaced by either `sorted_unique` or `sorted_equivalent`.

Reviewed By: vitaut

Differential Revision: D15196179

fbshipit-source-id: a2cdaa06c7c5a55147db67a877b29c5944039e6d

Summary:
See the test for a concrete example.
We have a scenario where both a library, and a service using that library, are using fiber. Managing a separate similar-sized thread and event-base pool costs us complexity and performance. So this seems like a legit need.

The overall solution is to hash the Options struct.
But because that has some cost, I made it optional, by introducing a new FrozenOptions struct.

Reviewed By: andriigrynenko

Differential Revision: D15305274

fbshipit-source-id: 5c53d8c4ed321ae88089c64de41af8229f877d36

Summary:
This modifies the Baton data-structure to store a linked list of awaiters rather than storing a single awaiter.

When a coroutine awaits the Baton it now does a lock-free push onto a list of awaiters.

When the Baton is posted it atomically dequeues all awaiters from the list and then resumes each of them in turn.

Reviewed By: andriigrynenko

Differential Revision: D15310137

fbshipit-source-id: 895ebcf2b113fb270ad7abfedbaab68ea51de84c

Summary: [Folly] Switch from `NDEBUG` to `kIsDebug` in `MPMCQueue`, preferring C++ over the preprocessor.

Reviewed By: Orvid

Differential Revision: D15303518

fbshipit-source-id: d014cea3035527bd27854f58b1a74a48740efd20

Summary:
If the value for backoff_max is less than backoff_min, we will see inconsistent behavior where the backoff_min sleep duration is used for the first 57 attempts, then backoff_max will be used.

In our case, we hit an issue where 0 was passed for backoff_max. We would see backoff_min sleep times for the first 57 attempts, then immediately drop to 0 sleep time which led to no retry wait at all.

If backoff_max < backoff_min, there's no way to respect both, but this makes the behavior consistent across all tries.

Reviewed By: yfeldblum

Differential Revision: D15282681

fbshipit-source-id: d9dd7b2498d77e4a01c9454a639b612ed9e26519

Differential Revision:
D15177816

Original commit changeset: 7298fcdc3a73

fbshipit-source-id: fc62ec2ecf7fb06bc60cdd0c33b17b53fe35af19

Summary:
HHWheelTimer is a member of EventBase, and EventBase is passed to it in the ctor.
However during the destruction of the evb, wheel timer gets destroyed after the evb is destroyed and still retains a reference to the evb. This is not so good.

This diff fixes it by canceling the timer before the evb gets destroyed.

Reviewed By: lnicco

Differential Revision: D15177816

fbshipit-source-id: 7298fcdc3a73041484a315c420bcb866175c19cc

Summary:
This diff updates docker os_image to ubuntu18 and gcc7 in travis settings.
The dependencies that we work with internally (e.g. gcc7, boost and so on) are not present in the ubuntu-16.04. Further, the support for gcc5 is going away. So maintaining this compatibility for these platforms is painful, especially for new projects, such as mvfst.

Reviewed By: simpkins

Differential Revision: D15286181

fbshipit-source-id: eb0da9556cdb17d0aa132ac2aa4c35f9c6eccd97

Summary: [Folly] `XLOG_EVERY_N`, `XLOG_EVERY_N_EXACT`. Renaming `XLOG_EVERY_N` to `XLOG_EVERY_N_EXACT` and adding a new `XLOG_EVERY_N` which performs better under contention, at the cost of possibly missing increments and therefore over-logging.

Reviewed By: simpkins

Differential Revision: D15230953

fbshipit-source-id: cf7b86b96340f78156f7cd25f7e784b0f1b8d5fa

Summary:
[Folly] Let CPUThreadPoolExecutor use unbounded queues when no queue bounds are specified.

One constructor remains which specifies queue bounds; that constructor uses bounded queues.

Reviewed By: andriigrynenko

Differential Revision: D15113118

fbshipit-source-id: 9c8ace4f170fdfdeef152e55b371d434ad248866

Summary:
xplat folly sync got blocked due to an error with missing ::remainderl
function.
Not really sure why this function is missing, for now ignore it.

Reviewed By: Orvid

Differential Revision: D15298056

fbshipit-source-id: 34035da5de826b55e28c79b8540aacd827e1ffdf

Summary:
- On GCC 4.9 and below, the type of `vec` could not be deduced in the
  `for_each` call.
- Since folly requires GCC 5.1 or later now, which has no issues in the
  deduction, remove the workaround and associated comment.
Pull Request resolved: https://github.com/facebook/folly/pull/1126

Reviewed By: Orvid

Differential Revision: D15292875

Pulled By: yfeldblum

fbshipit-source-id: 50533536e9e73d2c3e5d466d1b6143d0d4085648

Summary: We need this for dependencies that require submodules

Reviewed By: saifhhasan, GirasoleY

Differential Revision: D15282792

fbshipit-source-id: b0cc8d645e73668252409934fd6741fb211e30ae

Summary:
On Windows, the writing operation would fail with:

```
Traceback (most recent call last):
  File ".\opensource\fbcode_builder\getdeps.py", line 400, in <module>
    sys.exit(main())
  File ".\opensource\fbcode_builder\getdeps.py", line 393, in main
    return args.func(args)
  File ".\opensource\fbcode_builder\getdeps.py", line 236, in run
    change_status = fetcher.update()
  File "C:\open\fbsource\fbcode\opensource\fbcode_builder\getdeps\fetcher.py", line 451, in update
    return mapping.mirror(self.build_options.fbsource_dir, self.repo_dir)
  File "C:\open\fbsource\fbcode\opensource\fbcode_builder\getdeps\fetcher.py", line 400, in mirror
    f.write(name + "\n")
  File "C:\Python36\lib\encodings\cp1252.py", line 19, in encode
    return codecs.charmap_encode(input,self.errors,encoding_table)[0]
UnicodeEncodeError: 'charmap' codec can't encode characters in position 104-105: character maps to <undefined>
```

and this is caused by a file in libgit2: https://github.com/libgit2/libgit2/blob/master/tests/resources/status/%E8%BF%99, which is intended to test handling non-ASCII path.

Python on Windows will write file in cp1252 encoding by default, which does not contain that Chines character. (Caveat: that file on my system doesn't have the correct file name as well, it is being encoded in IBM861 for some reason. However the characters in IBM861 does not exist in CP1252 either)

Reviewed By: wez

Differential Revision: D15281521

fbshipit-source-id: 8a75e32bc1042167c945d67e26b549fda83b6b41

Summary:
The goal is to return an error code > 127 in the case of a
transient, retryable, infrastructure error.  This diff generates
those in the case of failure in downloading a URL or from interacting
with LFS.

Reviewed By: strager

Differential Revision: D15266838

fbshipit-source-id: 4f52a791320123968869032c37912dded464a86e

Summary:
the cmake `add_test` and related functions allow specifying
the WORKING_DIRECTORY to use for tests.  We weren't respecting this
value, so this diff looks up the WORKING_DIRECTORY from the ctest
json info and adjusts the buck test info json blob that we pass
on the testpilot.

Since that interface only allows passing an argv array, we use
the `cmake -E chdir` command to run a command in a specified
directory in a portable manner.

Reviewed By: strager

Differential Revision: D15274012

fbshipit-source-id: 1f02d461d73178745794703d455494e31c2e09ed

Summary:
std::remainder isn't supplied by all platforms. Specifically, uclibc
doesn't have it. Do a similar case as with std::nextafter, i.e. implement a
folly version in case it doesn't exists.

Reviewed By: yfeldblum

Differential Revision: D15291533

fbshipit-source-id: 96a70b52af11135102fda3183626df69cdaf4261

Summary:
When building using uclibc, std::nextafter doesn't exist. This is
similar case as for android, where std::nextafter doesn't exist.
Add additional condition to choose folly supplied versions of this function.

Reviewed By: yfeldblum

Differential Revision: D15291534

fbshipit-source-id: f6c9e213248f4e2a6e88e20578aeade46dc85d6b