Summary:
If a pthread_rwlock_t shared mutex is currently write-locked, we can
find its owner. However, if it is not write-locked or read-locked
(possibly by multiple threads), we will be unable to find the owner.

We can extend our deadlock detector to find deadlocks involving write-locked
pthread_rwlock_t mutexes as well.

Example output:

  Type "deadlock" to detect deadlocks.
  (gdb) deadlock
  Found deadlock!
  Thread 2 (LWP ...) is waiting on pthread_mutex_t (0x...) held by Thread 3 (LWP ...)
  Thread 3 (LWP ...) is waiting on pthread_rwlock_t (0x...) held by Thread 2 (LWP ...)

Reviewed By: yfeldblum

Differential Revision: D6974509

fbshipit-source-id: 8e64cb35f93958a9785070184df4c818b086ff0f

Summary:
The install step for folly CMake-based builds installs a folly-config.cmake
file that downstream projects can use to depend on folly in their own
CMakeLists.txt files.

CMakeLists.txt was previously generating the folly-config.cmake file by copying
CMake/folly-deps.cmake.  This seems wrong, as the installed folly-config.cmake
file should not contain `find_package()` statements that depend on CMake
modules only available in the folly source.

This adds a new folly-config.cmake.in template instead.

Reviewed By: yfeldblum

Differential Revision: D7035992

fbshipit-source-id: e6023094a807cf481ac49998c6f21b213be6c288

Summary:
When compiling with optimization enabled, gcc 7.2 complains about the
`childDir` variable potentially being clobbered by the child process after
`vfork()`.  This shouldn't matter since `childDir` is only used in the child
process, and not in the parent after the `vfork()`.

Reviewed By: yfeldblum

Differential Revision: D7035993

fbshipit-source-id: c80a4e18ad4f629a3d08e310cfd4500ebaf70994

Summary:
gcc does not complain about unknown `-Wno-*` flags by default, and older
versions of gcc do not support `-Werror=unknown-warning-option`.  Therefore
these checks for `-Wno-noexcept-type` and `-Wno-nullability-completeness` could
incorrectly indicate that the compiler supports these flags.

This changes the checks to check against the positive version of the warning
flag.

Reviewed By: yfeldblum

Differential Revision: D7035995

fbshipit-source-id: b1511d31721aeb779859848e20820c8a7fea4faa

Summary:
This brings us closer to being able to compile the open
source version of Folly with Clang. There are a few other
open pull requests that need to be addressed to allow us to
compile on Clang without any warnings that aren't already
explicitly disabled. See pull requests #745 and #756.

Closes https://github.com/facebook/folly/pull/772

Reviewed By: simpkins

Differential Revision: D7021595

Pulled By: yfeldblum

fbshipit-source-id: 350b972c74502ca2bc147f21428c622b19805b89

Summary:
[Folly] Avoid assuming that `FOLLY_HAVE_VLA` is defined.

Treat undefined as meaning `0` rather than requiring it be defined to either `0` or `1`.

Closes #773.

Reviewed By: andrewjcg

Differential Revision: D7028174

fbshipit-source-id: 90c52a3c5b640776f728c70b69f0500749d7ab7f

Summary: [Folly] Split definition of `kMscVer` from `kIsWindows` and use in `hazptr`.

Reviewed By: mzlee

Differential Revision: D7027159

fbshipit-source-id: 1dd2bad821919a2d579aa615ecb2921e5a3ce4d8

Summary:
During move construction and assignment `size_` and `max_size_` aren't set.
Added a test case that attempted to allocate 4096 GB before the patch, and passes after.

Reviewed By: yfeldblum

Differential Revision: D7005041

fbshipit-source-id: b42a0a408ce49ddded4ba805fc064cd11df2457e

Summary: Log bytes before/after (de)compression, and timing to the logging backend with probability 1/50.

Reviewed By: yfeldblum

Differential Revision: D6817204

fbshipit-source-id: 7fde5284b8c97355ec445376a7c481e9ffd2cfce

Summary: s/lock_unshared/unlock_shared/

Reviewed By: yfeldblum

Differential Revision: D7017355

fbshipit-source-id: 7f9d12b4c80cc796086a74890c58aceea931d93c

Summary:
[Folly] Less `goto` in `ltrimWhitespace` and `rtrimWhitespace`.

The generated machine code remains unchanged for `g++ -O3` builds.

Generally, prefer to use `goto` in cases which cannot, or which can only arduously, be accomplished without it. In other words, reserve `goto` to cases where `goto` is necessary, rather than to cases where it saves one line.

Closes #732.

Reviewed By: andrewjcg

Differential Revision: D7027269

fbshipit-source-id: 91a4c5dba0e4c13dc727393799edc9b596cdb8b9

Summary: It's not available in VS2015u3.

Reviewed By: yfeldblum

Differential Revision: D7005874

fbshipit-source-id: 5477018e444d9bf118089bbbcfabbbf4e16a3198

Summary: Closes https://github.com/facebook/folly/pull/770

Reviewed By: nbronson

Differential Revision: D6991705

Pulled By: yfeldblum

fbshipit-source-id: 9f6ae0e09b436993ee998fe6c5d347121c229489

Summary:
std::atomic is no longer trivial, and VS2015 and VS2017 both
fail this static assert, as will future versions of conforming compilers.
Updates to folly mean that the hazptr header is transitively included in
more places which makes it easier to trip this even when not using hazptr.
Disabling this assert to visual studio at least saves users a headache.

Reviewed By: yfeldblum, magedm

Differential Revision: D7006587

fbshipit-source-id: 8481e1d5668aee8cab703a04b2a390550c210a2c

Summary:
UndefinedBehaviorSanitizer: null-pointer-use buck-out/dev-asan-ubsan/gen/folly/concurrency/concurrent_hash_map#header-mode-symlink-tree-with-header-map,headers/folly/concurrency/ConcurrentHashMap.h:440:30 in
Finished test run: https://our.intern.facebook.com/intern/testinfra/testrun/882372015
Summary (total time 1.49s):
  PASS: 0
  FAIL: 0
  SKIP: 0
  FATAL: 1
    folly/concurrency/test:concurrent_hash_map_test - ConcurrentHashMap.CopyIterator
  TIMEOUT: 0
  OMIT: 0
```

Reviewed By: djwatson

Differential Revision: D7003359

fbshipit-source-id: d6764caf2dfb1f19cf5b38a9e89d7c0583c75412

Summary: [Folly] Cut outdated comment referencing `TFramedAsyncChannel`.

Reviewed By: knekritz

Differential Revision: D7002980

fbshipit-source-id: 206ac77abf8882e26a5b98ad4229c7a645912947

Summary: This makes it easier to integrate folly::Observer with callback-based code.

Reviewed By: spalamarchuk

Differential Revision: D6888234

fbshipit-source-id: ccf9b57c07ee323ee11496d14f8c90e3331cc7bb

Reviewed By: yfeldblum

Differential Revision: D6957008

fbshipit-source-id: 848932827efaf8529513026b45446da2c1697253

Summary: The template that this template wraps requires only forward iterators, so it doesn't make sense for the wrapper to be stricter than that.

Reviewed By: yfeldblum

Differential Revision: D6988987

fbshipit-source-id: 575dcacf7b1873ea0d7112acf84670842ee82fa7

Summary: Add hazptr_cleanup to ConcurrentHashMap destructor & add a test to ensure deletion is working.

Reviewed By: djwatson

Differential Revision: D6951359

fbshipit-source-id: 277c304a0e5073605010b14764d05a27e6e10b2d

Summary:
Added missing steps for registering and deregistering per-thread hazptr_priv structures with the default domain.
Updated tests to detect the error

Reviewed By: djwatson, davidtgoldblatt

Differential Revision: D6965459

fbshipit-source-id: 6eb6a5981c1d22518f318ace7387a3941a013352

Summary:
It's common to use UBSAN with ASAN at the same time, and we'll merge them eventually too. As a result we can check for ASAN defined macros to conditionally compile code for UBSAN.

We still have a few build modes that use standalone UBSAN, and unlike other sanitizers, UBSAN doesn't define a magic macro that we could use. In the past we added UNDEFINED_SANITIZER to our compiler flags, but I'd like to go away from it towards a more standard solution (eg check for ASAN).

Reviewed By: yfeldblum, luciang

Differential Revision: D6931054

fbshipit-source-id: 4b298d1cd72cecd693bf99eda309568ef019c544

Summary: Starting with C++11, implicit generation of copy/move constructors/operators in presence of user-defined destructors/constructors/operators is deprecated

Reviewed By: yfeldblum

Differential Revision: D6875337

fbshipit-source-id: a623a3ce2e7f5559d2de2324565c8995682085c7

Summary:
The `bulk_insert()` method left duplicates from the input range when the
container was empty.

Also, `bulk_insert()` relies on `std::inplace_merge()` to be stable, but
`libc++`s version of `std::inplace_merge()` was buggy until [August 2017](https://github.com/llvm-mirror/libcxx/commit/25a78dcd773ce509469a2b892adbeac0c230cdb9).
Since it is fixed upstream, and it is a library bug, I've not mitigated the bug.

Reviewed By: yfeldblum

Differential Revision: D6943195

fbshipit-source-id: 136caa52c21a9e0dbc09b0a47ad57e99d2fde380

Summary:
Clang warns about formatting when we were using `stringPrintf` to print
the result of overflowing an unsigned 64-bit value.

Closes issue #757.
Closes https://github.com/facebook/folly/pull/765

Reviewed By: simpkins

Differential Revision: D6934612

Pulled By: yfeldblum

fbshipit-source-id: df8c29c0ca8a47aca99007b5182529df17a27149

Summary:
Add a domain argument to frestanding RCU functions.
Perhaps slightly less awkward than using the domain, especially
for retire() and rcu_reader, but nothing that can't already
be done with domain calls directly.

Reviewed By: yfeldblum

Differential Revision: D6878941

fbshipit-source-id: 2397bc8c18a4175793932687e7c3c59f2825caa2

Reviewed By: yfeldblum

Differential Revision: D6893567

fbshipit-source-id: 054e109e3c3c3b2b20ac6ffe8f2ee05335e3dd89

Summary:
[Folly] A `transparent` adapter to facilitate the `is_transparent` protocol.

Compare types with member type or member alias `is_transparent` may be used to enable heterogeneous lookup in ordered containers which support it, such as `std::set` and `std::map` as of C++14, and `folly::sorted_vector_set` and `folly::sorted_vector_map`.

In some cases, nothing more is required than to add that `is_transparent` member to an existing compare type - `folly::transparent` may facilitate such cases.

Example:

```lang=c++
using TransparentStringPieceLess = folly::transparent<std::less<folly::StringPiece>>>;
std::map<std::string, float, /* Compare = */ TransparentStringPieceLess> m;
m["somekey"] = 12.;
m.find(folly::StringPiece("somekey"))->second = 13.; // heterogeneous lookup
```

To demonstrate the example, rewrite the `folly::sorted_vector_set` and `folly::sorted_vector_map` transparent-lookup tests.

Reviewed By: ot

Differential Revision: D6921862

fbshipit-source-id: 306ff1f6f178ae29ac80ee7cd28595b80fc393d1

Summary: Implement some OpenSSL 1.1.0 specific methods, so that code written with OpenSSL 1.1.0 can be compiled with OpenSSL 1.0.2

Reviewed By: yfeldblum

Differential Revision: D6930234

fbshipit-source-id: 257893787cecdfeaf9df13dce0ada46c15990699

Summary: Fix a case of null pointer dereference when ~hazptr_array() calls ~hazptr_holder(). For speed hazptr_array doesn't fill the domain_ field of hazptr_holder. The common case is that elements of the array are taken from and returned to the thread cache. If ~hazptr_array() doesn't find enough space in the thread cache, then it calls ~hazptr_holder() with a null domain_. The fix is to set domain_ to &default_hazptr_domain() before calling ~hazptr_holder() in such a case.

Reviewed By: yfeldblum

Differential Revision: D6928705

fbshipit-source-id: 9c756dce2854a2d1869a0712c4e6ed30f5df6b30

Summary: [Folly] Work around UB `signed-integer-overflow` in `Histogram`.

Reviewed By: igorsugak

Differential Revision: D6922381

fbshipit-source-id: 96ab6ffae9509ce8138f1741ea976226dec597c8

Summary: [Folly] Fix UB `signed-integer-overflow` in tracing test.

Reviewed By: igorsugak

Differential Revision: D6922450

fbshipit-source-id: 0d9a338fdcf205987d88875aac4112666ca32b65

Summary: Needed to update folly in the open source build of HHVM.

Reviewed By: yfeldblum

Differential Revision: D6913772

fbshipit-source-id: 50e540fe30651c991dc198ede66d4816a85f68db

Summary:
During program destruction we destroy registered log handlers (to allow them to
flush state if desired).  However we intentionally leak the LoggerDB object
itself and the LogCategory objects it contains, so it will still be safe if
static destruction code that runs later still tries to log messages.

However, when run with `ASAN_OPTIONS` set to `detect_leaks=1` ASAN complains
about the fact that we leak the LoggerDB singleton.

This changes the code to store the LoggerDB address in a static variable, to
prevent ASAN from complaining that we leak it.

Reviewed By: yfeldblum

Differential Revision: D6920772

fbshipit-source-id: 62f6837053ba4a538a89b4b61f9b1c722f253d33

Summary: The free function hazptr_cleanup reclaims all reclaimable objects retired to a domain before this call.

Reviewed By: djwatson

Differential Revision: D6876253

fbshipit-source-id: 54a6b98291712286597561e971fc9b6ac9c4d6f0

Summary:
[Folly] Do nothing in `spin_pause_until` if `spin_max` is zero.

Rather than go through the loop at all, if the instruction is not to spin - encoded as `spin_max <= 0` - then return early.

Reviewed By: nbronson

Differential Revision: D6910431

fbshipit-source-id: fa94510e7ec1ebdd7bc3f1a407dcf70e1e153b40

Summary: This dict holds strong reference to all the loops and leads to leak if new loop is created and used, change it to weakref to avoid it.

Reviewed By: yfeldblum

Differential Revision: D6892775

fbshipit-source-id: a628a3743876ef0580acce5fb9f283da573cf459

Summary:
This updates the code to create a default LogHandler that logs to stderr when
the main LoggerDB singleton is first created.  With this change users will get
reasonable log behavior even if they never call `folly::initLogging()` (log
messages will go to stderr instead of being ignored).

This new initialization code can potentially run very early during program
initialization (before `main()` starts and before folly `Singleton`
initialization) if some code in the program logs messages before main.  However
this current initialization code should be safe to run before main.

For compilers that support weak symbols (clang and gcc), I have defined
`initializeLoggerDB()` as a weak symbol, allowing users to override this
behavior at link time on a per-program basis should they choose to.

Reviewed By: yfeldblum

Differential Revision: D6893207

fbshipit-source-id: 13b72a01aa383c4611204df88ff8dad93abcc7b2

Summary: With the mobile thread_local changes, these assertions are no longer correct.

Reviewed By: magedm

Differential Revision: D6915065

fbshipit-source-id: 42072ed638b3dca56cba65ca0cff5e4d7aef2cb3

Summary:
Change `LoggerDB::get()` to a reference instead of a pointer since this
function can never return null.

Reviewed By: yfeldblum

Differential Revision: D6893206

fbshipit-source-id: af47063918a79c851fd39b838d6c63755166e033