Summary:
[Folly] Ensure global inits in Random are static-safe by using the machinery of `static` local variables.

The default mutex type used with `once_flag` is `SharedMutex`, but that mutex type is not trivially-destructible. An alternative mutex type might be trivially-destructible, such as `MicroLock`, so an alternative solution could be to use `once_flag` with `MicroLock`.

Reviewed By: ot, luciang

Differential Revision: D19207285

fbshipit-source-id: a7d817cf8521337543e3e27bc0e0e61241714ced

Summary:
Declare gflag to enable turning it on and off at run time.
Added folly_ prefix to gflag name to reduce name pollution.

Reviewed By: davidtgoldblatt

Differential Revision: D19144459

fbshipit-source-id: 1e82cb102e47fea856952d1e630c03baf38adf1d

Summary: [Folly] `FOLLY_SAFE_PCHECK`, for reporting errors from C standard library functions which may set errno from within code which must be reentrant.

Reviewed By: nbronson

Differential Revision: D19222785

fbshipit-source-id: 532ccee39c9d180b9b17edbd31517c09024539d7

Summary: [Folly] Avoid unnecessary virtualization in `MemoryMapping`.

Reviewed By: Orvid

Differential Revision: D19261862

fbshipit-source-id: 520f1182477bc723c8f42505b30fc09e00758d05

Summary: This change also introduces a dummy no-op stats implementation for non-fbcode builds.

Reviewed By: farnz

Differential Revision: D19193954

fbshipit-source-id: f513368a2387987bf129ddf935217c90b7c84f2a

Summary: Note: also making the `__GNUC__ || __clang__` nearby part in Pretty.h the same so that the header doesn't appear inconsistent, likewise for Hardware.h

Reviewed By: yfeldblum

Differential Revision: D19237732

fbshipit-source-id: bf3270ddf2102ea722be1eb1a557234d264f9af7

Summary:
Found with readability-const-return-typ
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1297

Reviewed By: Orvid

Differential Revision: D19258433

Pulled By: yfeldblum

fbshipit-source-id: f544c9168887c2ecb27d59ab799cda05c9a49a98

Summary:
Found with performance-for-range-copy
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1296

Reviewed By: markisaa

Differential Revision: D19258365

Pulled By: yfeldblum

fbshipit-source-id: aa516b18fc2e2405c57ab92fc000a57262dbf96a

Summary:
Contrary to other systems, jemalloc is part of the FreeBSD's libc.
As such APIs are directly exposed. Enabling super pages
as well, supporting same page size.
Pull Request resolved: https://github.com/facebook/folly/pull/1294

Reviewed By: interwq

Differential Revision: D19249354

Pulled By: yfeldblum

fbshipit-source-id: 845f594b28f30b0360d4573d8a7354f9790bc8b7

Summary: [Folly] Delegate `#include <jemalloc/jemalloc.h>` to the portability header.

Reviewed By: interwq

Differential Revision: D19254447

fbshipit-source-id: f07ac46e65313cbe5b9b7ba3df139978fad83b33

Summary:
Found with readability-container-size-empty
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1293

Reviewed By: markisaa

Differential Revision: D19241094

Pulled By: yfeldblum

fbshipit-source-id: 9362e8dc5dc1458fcd2b27c11ce81eafe674bcd2

Summary: [Folly] Apply `clang-format` in places.

Reviewed By: markisaa

Differential Revision: D19249033

fbshipit-source-id: 1a768eea013112ee5cd9d8b78114927fd6d6db04

Summary: [Folly] Apply python formatting to a few unsynced places in some `.pyi` files.

Reviewed By: markisaa

Differential Revision: D19249037

fbshipit-source-id: 6b14c5fedbee0f25a4b0fdfdd0f157e62a95a766

Reviewed By: yfeldblum

Differential Revision: D19235483

fbshipit-source-id: 5caea4efb3e811f78f1be3f05adc315bccb0fc0d

Summary:
Found with modernize-use-equals-default
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1291

Reviewed By: markisaa

Differential Revision: D19229643

Pulled By: yfeldblum

fbshipit-source-id: 69f0bd4f55ea5a3697ab176aab80683b99b4e11c

Summary: [Folly] Return spelled default ctor in `AlignedSysAllocator` - it is necessary until gcc7.

Reviewed By: LeeHowes

Differential Revision: D19235864

fbshipit-source-id: fd0e9c33283e8b646aff56efb10d8e2a8c05e293

Summary:
I'd like to enable UBSAN's unsigned-integer-overflow check. It's not an undefined behavior, but it is often unintentional.

This adds suppressions to folly/hash where unsigned-integer-overflow is expected due the the nature of operations.

Reviewed By: yfeldblum

Differential Revision: D19217819

fbshipit-source-id: 1a468018823e0ac1d01c3ae889e38e9d42bdd8b7

Summary: [Folly] Cut some outdated lines in docs and comments which refer to no-longer supported versions of gcc.

Reviewed By: Orvid

Differential Revision: D19229059

fbshipit-source-id: c8abb81c292091339622e9cc3d4e5fd3cd6459b4

Summary:
libcxx does not have strcmp or strlen as constexpr.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1288

Reviewed By: Orvid

Differential Revision: D19226813

Pulled By: yfeldblum

fbshipit-source-id: 4b5de4f923f4d7871c059af54eb7d375f344f48a

Summary: [Folly] Use `getCurrentThreadName()` in `EventBaseTestLib.h` v.s. using pthread directly.

Reviewed By: Orvid

Differential Revision: D19226856

fbshipit-source-id: 90f825ab0fc186a896cd63c425e55153daa0fefd

Summary: [Folly] Update a comment in `CacheLocalityBenchmark.cpp` to refer to the direct function used for fetching the thread-id.

Reviewed By: Orvid

Differential Revision: D19226855

fbshipit-source-id: 8cf203daa7b3fbc00e387dca881be7f9ce47af95

Summary:
[Folly] Use `std::thread` in `AtomicHashMapTest.cpp` instead of using the pthread API.

In general in normal C++ code, prefer C++ standard library facilities over C facilities.

Reviewed By: Orvid

Differential Revision: D19226850

fbshipit-source-id: 7edc5fca25266b48ab3e8c8829708594d7563409

Summary:
- EventBaseTestLib.cpp file was moved to EventBaseTestLibProvider.cpp
  in Differential Revision: D19001877.
- Change CMakeLists.txt to specify EventBaseTestLibProvider.cpp so that
  CMake does not fail at configure time due to EventBaseTestLib.cpp not
  existing in the source tree.
Pull Request resolved: https://github.com/facebook/folly/pull/1284

Reviewed By: markisaa

Differential Revision: D19226798

Pulled By: yfeldblum

fbshipit-source-id: 45a5e5ca03a52897a6f8f711761bf78ef386e6f6

Summary: [Folly] Fix typo in `clang-format` control.

Reviewed By: markisaa

Differential Revision: D19224383

fbshipit-source-id: 1d153fb4eaead422d93c235e40163b8ae855d13a

Summary: It's a read-only interface.

Reviewed By: philippv, luciang

Differential Revision: D19199710

fbshipit-source-id: 5e900edd35640cded45e6376eb0b81b9036bd792

Summary: It can be useful for diagnostics to print the stack traces of thrown exceptions that are not yet handled, for example if a fatal error happens during unwinding.

Reviewed By: philippv, luciang

Differential Revision: D19191089

fbshipit-source-id: dae8cc0a095570bfe7d6c4c7ef2b5dc08aa9f1f9

Summary: [Folly] Let `AtFork` accept ptr-to-`const` handle rather than requiring only ptr-to-non-`const`.

Reviewed By: luciang

Differential Revision: D19208255

fbshipit-source-id: c7a6ab42b7b652ebc4ea6465f80db37ef1d155df

Summary: [Folly] Rename `object` to `handle` in `AtFork` since it is used opaquely.

Reviewed By: luciang

Differential Revision: D19208246

fbshipit-source-id: 3fc23e85b64f911f8e172a22f7e7080952d4dfe9

Summary: [Folly] Register with `AtFork` without an object in `BufferedRandomDevice`, since no object is actually needed and since it is possible for the object passed in to be destroyed.

Reviewed By: luciang

Differential Revision: D19207385

fbshipit-source-id: 7c0e2793d9b438e576bf26bc3b3b459b372aec6d

Summary: [Folly] Let `AtFork` ignore unregistration of `nullptr` objects - i.e., let `nullptr` objects be the way of signaling immunity from unregistration.

Reviewed By: luciang

Differential Revision: D19207380

fbshipit-source-id: 6dfee52bf815fc257697fb89d67f69cede0f39f6

Summary:
[Folly] Define `to_narrow` in terms of implicit conversion to take advantage of ubsan's truncation checks.

Note that integer truncation is not undefined behavior per se but is commonly unintended, so ubsan checks for it. Uses of `to_narrow` should not be subject to truncation; wherever truncation is anticipated, better to use an explicit conversion with a named destination type.

Requires suppressing `-Wconversion` around the implicit conversion to avoid extraneous warnings and errors when the compiler detects narrowing implicit conversions. Since `to_narrow` models explicit conversion while merely being defined in terms of implicit conversion, it is already anticipated by all users that a narrowing conversion may occur; the warning is there to inform users only about unanticipated cases.

Reviewed By: luciang

Differential Revision: D19201430

fbshipit-source-id: b3739d3d9aa05e95f7b3e4dff54e719691c1f04c

Summary: [Folly] Mark `co_error` as `final`.

Reviewed By: lewissbaker

Differential Revision: D19173904

fbshipit-source-id: 3dedf9a27629dd495d7097d00566a92753b65778

Summary: [Folly] Forbid `co_error` with empty `exception_wrapper`.

Reviewed By: lewissbaker

Differential Revision: D19172250

fbshipit-source-id: 5ffa7502eb97b6a084be3112b17f29bad2d96a0f

Summary:
Found with modernize-loop-convert
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Pull Request resolved: https://github.com/facebook/folly/pull/1260

Reviewed By: markisaa

Differential Revision: D18747876

Pulled By: yfeldblum

fbshipit-source-id: f2b19654bb3fcbb75b1210edeef536c88f3052d1

Summary: [Folly] Let `PriorityUnboundedQueueSet` align its queues, which have extended alignment, even when `std::allocator` does not honor extended alignment.

Reviewed By: magedm

Differential Revision: D19174993

fbshipit-source-id: 0bc0d75c6d239c97b6e1c7380bffff0363008e16

Summary: [Folly] Let AlignedSysAllocator consider the natural or specified alignment of the type for which memory is being allocated, rather than only considering a constant or fixed alignment.

Reviewed By: nbronson

Differential Revision: D19174932

fbshipit-source-id: e8ed0fbaf30d285cadd061d0fe5d69fcbb16f9e2

Summary: The tagged/specific exception variant of thenError and deferError propagate interrupts where currently the untagged variant of thenError does not. As far as I can tell this was an oversight. This makes untagged thenError behave as expected.

Reviewed By: yfeldblum, mpark

Differential Revision: D18643814

fbshipit-source-id: 42688c4363de2f6b027766b19276b2e8ec6a48a8

Summary: [Folly] Remove spelled default ctor in `AlignedSysAllocator`, which was needed only for gcc49.

Reviewed By: aary

Differential Revision: D19171223

fbshipit-source-id: ee42f6ac188dc896519537870353116c0dfd3acd

Summary:
[Folly] Extended constructibility for `co_error` - constructible from everything `exception_wrapper` is constructible from, but not default-constructible. This permits:

```lang=c++
co_yield folly::coro::co_error(std::runtime_error("foo"));
```

Reviewed By: lewissbaker

Differential Revision: D19172210

fbshipit-source-id: df21c4f2a15ad2286e7cb34aae7c1be05e797a94

Summary: [Folly] Avoid a too-large pointer literal in StlVectorTest, keeping the literal value divisible by 8 to avoid ubsan violations.

Reviewed By: igorsugak

Differential Revision: D19170467

fbshipit-source-id: 2c93d2b3d9f7515761668d3364c4b8b5093a6145