Summary:
Add support for onClear() callback function to RequestData. The callback is to be executed on the last clearing of RequestData. This typically happens upon the destruction of the last RequestContext that contains the RequestData or due to explicit calls to RequestContext::clearContextData that remove the RequestData from the RequestContext.

The onClear() callback is executed exactly once. If there are no calls to clearContextData, then onClear is executed upon the destruction of the last RequestContext that refers to the RequestData. If there are calls to clearContextData, then onClear is executed upon the removal of the last reference to the RequestData from a RequestContext.

RequestData::hasCallback() applies only to onSet() and onUnset() which may be called multiple times, whereas onClear() is called exactly once.

If a class derived from RequestData overrides onClear and either or both of onSet and onUnset, then care must be taken that the callbacks are mutually safe, as it is possible for the execution of onClear to be concurrent with executions of onSet and onUnset.

Motivation: Some uses of RequestContext::clearContextData expect the immediate destruction of associated RequestData. The hazard pointer-based implementation of RequestContext guarantees only that all associated RequestData is destroyed before the completion of the destructor of the RequestContext. This callback provides users of RequestContext more control of the timing of executing actions related to clearing the data. The addition of this capability is intended to provide a tool to eliminate differences in timing of executing actions between the two implementations of RequestContext (hazard pointer-based vs. read lock-based).

Also, fixes a minor bug in request_context_test, where the key "test2" was used in two tests (setIfAbsentTest and deadlockTest) without being cleared, which generated a warning when the tests were run in the same process.

Reviewed By: davidtgoldblatt

Differential Revision: D20332008

fbshipit-source-id: c9cefffb28c8da098e97f851f31d29a1bf704061

Summary:
Clean the sigar build dependency completely under:
- fbcode/fbzmq/public_tld/build/...
- fbcode/openr/public_tld/build/...

Delete files:
- fbcode/opensource/fbcode_builder/manifests/sigar
- fbcode/opensource/fbcode_builder/specs/sigar.py

Reviewed By: steven1327

Differential Revision: D20376067

fbshipit-source-id: 3321a14df2551525acee605028ef06a04cda7c6a

Summary: This is mainly used for logging, and we can elide an unnecessary copy in some cases.

Reviewed By: mingtaoy

Differential Revision: D20402430

fbshipit-source-id: 79b73792a69cf2817ffb88b07d9973139c3c67bb

Summary: Split WTCallback cod

Reviewed By: LeeHowes

Differential Revision: D20355082

fbshipit-source-id: 9ead4905cafeef27a9e2956fc50ebd59098cb80e

Summary:
Use decltype(event::ev_fd) instead of #ifdef

(Note: this ignores all push blocking failures!)

Reviewed By: yfeldblum, kevin-vigor

Differential Revision: D18525100

fbshipit-source-id: 7cef32def7b40c71b2298f9dc354fbe63607fb8e

Summary:
As written in the documentation, {fmt} should be built alongside folly. This is not possible in my case. My usecase is to provide some kind of small distribution of shared libraries that can be installed on any systems. Hence I really need to have libfmt treated as a normal shared lib.

Moreover, libfmt can be consumed on our side as a normal library outside on cmake. We do not provide the cmake config file which are strongly dependent on the system on which libfmt is installed.

My proposal is to first look for the fmt-config.cmake as it is now but in case of cmake support is not provided by system, we fall back to a normal library lookup.
Pull Request resolved: https://github.com/facebook/folly/pull/1328

Reviewed By: simpkins

Differential Revision: D20312046

Pulled By: yfeldblum

fbshipit-source-id: 2d6b4917b37a8ed1d553600599702259db22c212

Reviewed By: Orvid

Differential Revision: D20353603

fbshipit-source-id: 1910261be637237923b1f1d6a56fc954554176d2

Summary:
The result of `errnoStr` is often converted to `std::string` so returning `fbstring` adds an extra copy. Make it return `std::string` instead. This will also allow removing dependency between `String.h` and `FBString.h`.

(Note: this ignores all push blocking failures!)

Reviewed By: yfeldblum

Differential Revision: D20195395

fbshipit-source-id: 0dc65f1566911156be3fcb715dd105c58f2a8822

Summary:
Ideally we'd be using mkscratch, but this still isn't shipped
to our Windows systems.

Pick a path that is more friendly to our corp windows environment by default.

Reviewed By: pkaush

Differential Revision: D20342277

fbshipit-source-id: c85bccee6701adc03b26c92ba217b18bd684257a

Summary: [Folly] Cut `CachelinePadded` after migrating all uses to `cacheline_aligned`.

Reviewed By: aary

Differential Revision: D19657696

fbshipit-source-id: d2e5b4695bdc3734340723d5d3681e724734691e

Summary:
When Gtest cannot be discovered by CMake config file, the tests fails to compile:
```
In file included from /home/docker/opensource/folly/folly/test/common/TestMain.cpp:23:
/home/docker/opensource/folly/folly/portability/GTest.h:32:10: fatal error: gtest/gtest.h: No such file or directory
   32 | #include <gtest/gtest.h>
      |          ^~~~~~~~~~~~~~~
```
This PR fixes this problem with a classic `find_package`.
Pull Request resolved: https://github.com/facebook/folly/pull/1331

Differential Revision: D20323404

Pulled By: yfeldblum

fbshipit-source-id: 2e41f99172c504063c00c45a9fa5c08a26bcc0c8

Summary:
Migration from Future-returning executor-erasing collectX forms to SemiFuture-returning forms, that are less risky in particular with coroutines.

An earlier change added collectXSemiFuture. This diff adds collectXUnsafe as a migration path.

Reviewed By: yfeldblum

Differential Revision: D20255061

fbshipit-source-id: 379b5ed95aa7782aa121dba8e84cb48f802a57cd

Summary: SingletonThreadLocal is not needed by Future.cpp.

Reviewed By: Orvid

Differential Revision: D20142400

fbshipit-source-id: d0f3b7c1bbef03226c8d6264abeb96822587a465

Summary: As title. Expose the useful portable mlock2() wrapper.

Reviewed By: alikhtarov

Differential Revision: D20239849

fbshipit-source-id: 0fa776faf29ad69e551dfbe6eb69369560800a0e

Summary: Cast from unsigned to signed *before* negating. The resulting assembly code is the same, but there's no warning.

Reviewed By: yfeldblum, jdonald

Differential Revision: D19588115

fbshipit-source-id: d0d61ab0543211e2399f42a86a7a7a2366679d37

Summary:
The absence of GFlags is already partially optional -
portability/GFlags.h defines a number of stub macros / clases if
FOLLY_HAVE_LIBGFLAGS is false.

However the top-level CMake rules don't correctly handle when GFlags
isn't present - they unconditionally add GFlags-related variables to
CMAKE_REQUIRED_xxx variables.

Additionally there are a couple of source files which should not be
compiled if GFlags isn't available.
Pull Request resolved: https://github.com/facebook/folly/pull/1325

Differential Revision: D20252030

Pulled By: yfeldblum

fbshipit-source-id: 7c3441e8b2850d68df125fb9c14b8417f4d08b55

Summary: Support the brand-new TCMalloc version that was just released.

Reviewed By: luciang

Differential Revision: D19943852

fbshipit-source-id: 10ae07051d2bc56a502ee0d5e5767035cf53794f

Summary:
Replace the methods to get CPU and memory usage statistics:
- For the memory: use `VmRSS` of `/proc/[pid]/status`: http://man7.org/linux/man-pages/man5/proc.5.html
- For the CPU%: calculate the process is occupied how much percentage of the CPU time, use `getrusage()`: http://man7.org/linux/man-pages/man2/getrusage.2.html
   - Implemented like the sigar: https://our.intern.facebook.com/intern/diffusion/FBS/browse/master/third-party/sigar/src/sigar.c?commit=4f945812675131ea64cb3d143350b1414f34a351&lines=111-169
  - Formula:
    - CPU% = `process used time` during the period / `time period` * 100
    -  `time period` = current query timestamp - last query timestamp
    - `process used time` = current `process total time` - last query `process total time`
    - `process total time` = CPU time used in user mode + CPU time used in system mode // get from the API `ru_utime` and `ru_stime`

Remove the `fbzmq::ResourceMonitor` and `sigar`:
- Change and rename the UT
  - `ResourceMonitorTest.cpp` -> `SystemMetricsTest.cpp`
  - `ResourceMonitor` -> `SystemMetricsTest` in `openr/tests/OpenrSystemTest.cpp`
- Remove `ResourceMonitor` code and dependency for `Watchdog` and `ZmqMonitor`
- Remove `sigar` dependency used in building

Reviewed By: saifhhasan

Differential Revision: D20049944

fbshipit-source-id: 00b90c8558dc5f0fb18cc31a09b9666a47b096fe

Summary: This seems to not happen reliably and causes rss to grow. The `__dealloc__` cannot fix it completely, although I verified it is called. So calling it explicitly seems to guarantee it to happen.

Reviewed By: yfeldblum

Differential Revision: D20206887

fbshipit-source-id: 6285a8e8943a53d20714e54efd3220a545010102

Summary: as title, just to add a mock function

Reviewed By: yfeldblum, lnicco

Differential Revision: D20130909

fbshipit-source-id: 469d904cce93dc8e88c5dc4728b1f1fbd0baf798

Summary:
We have noticed some inappropriate uses of `assume_unreachable()`, and we believe that the documentation does not convey strongly enough that it is not an assertion.

This diff moves the implementation to a separate header (as the assertions in there might be misleading) and fleshes out more the intended use cases.

Reviewed By: yfeldblum, mkatsevVR, luciang

Differential Revision: D20182339

fbshipit-source-id: 8a3be1ada06dead7ea936971c42c4f30389fc23a

Summary:
[Folly] Fix `small_vector::insert` with non-random-access iterator arguments.

Closes https://github.com/facebook/folly/issues/1322.

Reviewed By: ot, terrelln

Differential Revision: D20150453

fbshipit-source-id: 184e748dcbb7cb60318b5f0a3104d9aeb8352d7b

Summary:
Codel is statically initialized with some default target delay and interval period. It would be useful to have the ability to update these parameters at runtime to fine tune load shedding to a given use case.

Adding methods here to update the target delay and interval parameters.

Reviewed By: yfeldblum

Differential Revision: D19987838

fbshipit-source-id: 4d67277f38037840883f6d9d5fd2d5f9be760f44

Summary: `type_traits` needed for `std::enable_if_t`

Reviewed By: yfeldblum, Orvid

Differential Revision: D20196749

fbshipit-source-id: 4d4e44c523959dda1f04b452d65880ca3401e036

Summary: Might as well store the count of newlines in the Log Message - as the field is already padded to <pointersize> bytes by the alignment requirements of the field either side it will not use any more space than the existing `bool`

Reviewed By: simpkins

Differential Revision: D20200298

fbshipit-source-id: 84cc52898069eae1af5bd0f103d4ab3177c45206

Summary: Add support for futures folly::HighResDuration

Reviewed By: kevin-vigor

Differential Revision: D19981652

fbshipit-source-id: 4a321fda487d5acdee04d138b0e4f0f10a4119f1

Summary: [Folly] Avoid uses of gflags integer aliases, preferring std integer aliases.

Reviewed By: WillerZ, vitaut

Differential Revision: D20204609

fbshipit-source-id: e00b183ab2574af7ea064b9602f1ab8efced50b0

Summary: g++ aggressively inlines F14VectorMap::operator[] calls from some testing loops with the loop index as a key. On platforms that have SSE2 but not SSE4.2 we use a multiplication and xor based bit mixer, which undergoes a strength reduction in optimized builds. Somehow the resulting code miscalculates the result (no minimal repro yet), which causes the tests to persistently fail. This diff blocks the compiler optimization, allowing the tests to pass.

Reviewed By: yfeldblum

Differential Revision: D19910180

fbshipit-source-id: 6cccba67645481061f923bd7b3e681df573c4039

Summary: [Folly] Let `Init` work without `gflags` support.

Reviewed By: simpkins

Differential Revision: D20150102

fbshipit-source-id: ceaae7ec3373f46457ba0d2a524b1223dec9961c

Reviewed By: vitaut

Differential Revision: D20098574

fbshipit-source-id: 409c3c1e7716a38f253fa3d0d8afe615139426cc

Summary:
This diff adds fb303 dependency to Open/R. fb303 is FB's open-sourced project
which provides basic infrastructure for thrift service e.g. base functions,
counters supports etc.

Using this in Open/R will allow both internal & external users leverage counters
via thrift APIs

Reviewed By: xiangxu1121

Differential Revision: D20134243

fbshipit-source-id: 23782ec99dc749dd27dc19f2b2b023f1c229608b

Summary: Adds a few simple benchmarks to get an idea for the overhead of `folly::coro::Task` coroutines.

Reviewed By: yfeldblum

Differential Revision: D19729527

fbshipit-source-id: cdfbd5c408c24dd258c56bc159af0c03cb1cd55e

Summary: As a step to eventually supporting TLS 1.3 by default for OpenSSL, we want to add the ability to selectively enable TLS 1.3 in folly::SSLContext.

Reviewed By: mingtaoy

Differential Revision: D20102778

fbshipit-source-id: 0865821ad2730482e797b888d16561b6989dca62

Summary: We should probably just free this.

Reviewed By: yfeldblum

Differential Revision: D20119718

fbshipit-source-id: 2a6c79730243b7af5ba90dac631ce0c8e1e0e6a4

Summary: Add API to Set Socket Option for AsyncUDPServerSocket.h, which is the same as AsyncUDPSocket.

Reviewed By: yfeldblum, lnicco

Differential Revision: D20064430

fbshipit-source-id: d448fbc217d0098e0ab94e2f113d4eba97958b02

Differential Revision: D19882830

fbshipit-source-id: 031217f9890351022bc8d171f0ccd7e045dd6972

Summary: Workaround for a TSAN fork deadlock

Reviewed By: kevin-vigor

Differential Revision: D20107616

fbshipit-source-id: df4d6222f63131e49fc1e831ec81d6441269b1f4

Summary: Sanitizers always intercept the allocator, so there's no reason to do the elaborate `usingJEMalloc()` logic -- it will always fail.

Reviewed By: yfeldblum

Differential Revision: D20056781

fbshipit-source-id: c0d4a3b103e26bacc0a6589f8d54b4a2ec85f771

Summary: This makes the function name more accurate. The function does not start the thread pool executor. It only enables the construction and use of a thread pool executor by the default hazptr domain when needed.

Reviewed By: davidtgoldblatt

Differential Revision: D19852860

fbshipit-source-id: 5ae71d0fd2d2d20c9ca66f45d57e90085703b3fb

Summary:
clang fails to optimize the current version even in the opt mode

code:
```
void clear(folly::small_vector<int, 5>& vec) {
    vec.clear();
}
```

before:

cat_huh

```
0000000000000000 <clear(folly::small_vector<int, 5ul, void, void, void>&)>:
   0:	55                   	push   %rbp
   1:	48 89 e5             	mov    %rsp,%rbp
   4:	41 57                	push   %r15
   6:	41 56                	push   %r14
   8:	53                   	push   %rbx
   9:	50                   	push   %rax
   a:	49 89 fe             	mov    %rdi,%r14
   d:	48 8b 07             	mov    (%rdi),%rax
  10:	48 8d 4f 08          	lea    0x8(%rdi),%rcx
  14:	48 85 c0             	test   %rax,%rax
  17:	78 5a                	js     73 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x73>
  19:	48 89 cf             	mov    %rcx,%rdi
  1c:	49 bf ff ff ff ff ff 	movabs $0x7fffffffffffffff,%r15
  23:	ff ff 7f
  26:	48 89 c3             	mov    %rax,%rbx
  29:	4c 21 fb             	and    %r15,%rbx
  2c:	74 3a                	je     68 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x68>
  2e:	48 8d 34 9f          	lea    (%rdi,%rbx,4),%rsi
  32:	48 85 c0             	test   %rax,%rax
  35:	78 53                	js     8a <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x8a>
  37:	48 8d 14 99          	lea    (%rcx,%rbx,4),%rdx
  3b:	48 89 d9             	mov    %rbx,%rcx
  3e:	48 29 f2             	sub    %rsi,%rdx
  41:	74 0e                	je     51 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x51>
  43:	e8 00 00 00 00       	callq  48 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x48>
  48:	49 8b 06             	mov    (%r14),%rax
  4b:	48 89 c1             	mov    %rax,%rcx
  4e:	4c 21 f9             	and    %r15,%rcx
  51:	48 c1 e3 02          	shl    $0x2,%rbx
  55:	48 c1 fb 02          	sar    $0x2,%rbx
  59:	48 29 d9             	sub    %rbx,%rcx
  5c:	49 ff c7             	inc    %r15
  5f:	49 21 c7             	and    %rax,%r15
  62:	49 09 cf             	or     %rcx,%r15
  65:	4d 89 3e             	mov    %r15,(%r14)
  68:	48 83 c4 08          	add    $0x8,%rsp
  6c:	5b                   	pop    %rbx
  6d:	41 5e                	pop    %r14
  6f:	41 5f                	pop    %r15
  71:	5d                   	pop    %rbp
  72:	c3                   	retq
  73:	48 8b 39             	mov    (%rcx),%rdi
  76:	49 bf ff ff ff ff ff 	movabs $0x7fffffffffffffff,%r15
  7d:	ff ff 7f
  80:	48 89 c3             	mov    %rax,%rbx
  83:	4c 21 fb             	and    %r15,%rbx
  86:	75 a6                	jne    2e <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x2e>
  88:	eb de                	jmp    68 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x68>
  8a:	48 8b 09             	mov    (%rcx),%rcx
  8d:	48 8d 14 99          	lea    (%rcx,%rbx,4),%rdx
  91:	48 89 d9             	mov    %rbx,%rcx
  94:	48 29 f2             	sub    %rsi,%rdx
  97:	75 aa                	jne    43 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x43>
  99:	eb b6                	jmp    51 <clear(folly::small_vector<int, 5ul, void, void, void>&)+0x51>
```

after (much better but still 3 instructions too many):
```
0000000000000000 <clear(folly::small_vector<int, 5ul, void, void, void>&)>:
   0:	55                   	push   %rbp
   1:	48 89 e5             	mov    %rsp,%rbp
   4:	48 b8 00 00 00 00 00 	movabs $0x8000000000000000,%rax
   b:	00 00 80
   e:	48 21 07             	and    %rax,(%rdi)
  11:	5d                   	pop    %rbp
  12:	c3                   	retq
```

Reviewed By: ot, philippv, luciang

Differential Revision: D19890921

fbshipit-source-id: 62f1b5325b4deda95976894b476849b5098fad4f