folly.git
8 years agoFix a race in Promise::setTry
Andrii Grynenko [Mon, 16 May 2016 21:06:54 +0000 (14:06 -0700)]
Fix a race in Promise::setTry

Summary:
This fixes a race which can only be exposed if Promise is owned by the same thread which was calling await and Promise is fulfilled from a different thread. What could happen then was:
1. Thread 2 fulfills the promise and thus saves the value and posts Baton.
2. Thread 1 wakes up, await returns and then Thread 1 destroys the Promise.
3. Promise destructor checks value_, which is still not null, so it tries to fulfil it with exception.

Reviewed By: spalamarchuk

Differential Revision: D3306969

fbshipit-source-id: adf799cfd7b75b0201fa675a9e44ac7c7c42ac95

8 years agoMake await exception safe
Andrii Grynenko [Mon, 16 May 2016 20:13:24 +0000 (13:13 -0700)]
Make await exception safe

Summary: This fixes fibers::await to correctly handle exception thrown by user-passed lambda. Await still always waits for the promise to be fulfilled (if the promise was not moved out, it will be destroyed and thus auto-fulfilled with "promise not fulfilled" exception). However if user-passed lambda throws, promise result is ignored (even if exception) and exception thrown by lambda is re-thrown.

Reviewed By: pavlo-fb

Differential Revision: D3303393

fbshipit-source-id: c4ba01fde0e156cc88e5d07aaf763e3abe121d11

8 years agoPrefer constexpr to preprocessor conditionals when checking endianness
Yedidya Feldblum [Sat, 14 May 2016 19:49:52 +0000 (12:49 -0700)]
Prefer constexpr to preprocessor conditionals when checking endianness

Summary:
[Folly] Prefer `constexpr` to preprocessor conditionals when checking endianness.

Using `if (folly::kIsLittleEndian) {` v.s. `#if __BYTE_ORDER__ == __LITTLE_ENDIAN__`.

Reviewed By: meyering

Differential Revision: D3296770

fbshipit-source-id: b26df83fdd42a50663746fc7c9d5fbe67e6671eb

8 years agoA thread-per-task executor
Yedidya Feldblum [Sat, 14 May 2016 11:29:37 +0000 (04:29 -0700)]
A thread-per-task executor

Summary:
[Wangle] A thread-per-task executor.

Moved from Folly into Wangle and fleshed out.
* Starts task threads from a control thread, rather than starting task threads from the submitter thread. Because starting task threads is likely to be more expensive than moving a functor.
* Joins task threads as they finish, rather than joining all task threads in the executor's destructor.

Suitable for running tasks which spend most of their time sleeping. Such as blocking IO, blocking fork-exec-wait, etc., when it is inconvenient to use the nonblocking variants with an IO executor.

Reviewed By: simpkins

Differential Revision: D3286988

fbshipit-source-id: 4b91133a7d55332ebbae020c1515c60e816906b3

8 years agofolly: fbstring: ubsan: memcpy/memmove are marked as nonnull - avoid calling them...
Lucian Grijincu [Sat, 14 May 2016 07:38:50 +0000 (00:38 -0700)]
folly: fbstring: ubsan: memcpy/memmove are marked as nonnull - avoid calling them when size == 0 and (likely) source is nullptr

Summary: Also see {D3295811}

Differential Revision: D3302564

fbshipit-source-id: 3f2dbf5a6cfa8199682cb4af90aac372d501919a

8 years agofolly: ubsan: elias-fano avoid memset(nullptr)
Lucian Grijincu [Sat, 14 May 2016 07:09:41 +0000 (00:09 -0700)]
folly: ubsan: elias-fano avoid memset(nullptr)

Reviewed By: ot

Differential Revision: D3300437

fbshipit-source-id: 7e0add5dab6dc1c891b742b4bcedea656c8284a9

8 years agofolly: ubsan: redo BitsTest (no macros) and fix DCHECK in test
Lucian Grijincu [Sat, 14 May 2016 00:23:32 +0000 (17:23 -0700)]
folly: ubsan: redo BitsTest (no macros) and fix DCHECK in test

Reviewed By: meyering

Differential Revision: D3283226

fbshipit-source-id: 7ae2ec7741ca53c494e8325e30f2995a241674c0

8 years agoTest creation of RequestContext inside a fiber task
Mirek Klimos [Fri, 13 May 2016 21:17:12 +0000 (14:17 -0700)]
Test creation of RequestContext inside a fiber task

Summary: ^, as discussed on D3279644

Reviewed By: andriigrynenko

Differential Revision: D3299473

fbshipit-source-id: dc302132db64399f768d69c641f848d0da1075c4

8 years agoReplace RequestContext::setContext with RequestContextScopeGuard in folly
Mirek Klimos [Fri, 13 May 2016 19:02:35 +0000 (12:02 -0700)]
Replace RequestContext::setContext with RequestContextScopeGuard in folly

Summary: To make sure RequestContext is properly unset when we stop processing request on a thread. This changes the API in Fibers, NotificationQueue, HHWheelTimer and AsyncTimeout, and fixes RequestContext handling in Futures (reset RC after the callback is done)

Reviewed By: andriigrynenko

Differential Revision: D3279644

fbshipit-source-id: a6a1c2840cdce179334aa1a3b28fa514cd5358c1

8 years agoConstruct all HHWheelTimer instances with the factory method
Chad Parry [Fri, 13 May 2016 18:56:03 +0000 (11:56 -0700)]
Construct all HHWheelTimer instances with the factory method

Summary:
This diff standardizes `HHWheelTimer` construction. It's not that safe to allow anyone to call `new HHWheelTimer` because the caller needs to remember to call `DelayedDestruction::destroy`.

Once callers are made to be safer, I'll be able to change the `HHWheelTimer` to use standard smart pointers instead of `DelayedDestruction`.

This picks up some work that I started in D2627941 but had to postpone.

This was mostly a mechanical change:

  ( fbgs -ls 'new HHWheelTimer' ; fbgs -ls 'new folly::HHWheelTimer' ) | xargs perl -pi -e 's/\bnew\s+((?:folly::)?HHWheelTimer)\b/$1::newTimer/g'

Then I manually inspected the code. I reverted `folly/io/async/HHWheelTimer.h`, since the `newTimer` factory method is the one place that we still want to call `new HHWheelTimer`. I manually edited `proxygen/lib/utils/SharedWheelTimer.cpp`, since that was using a `shared_ptr` with a custom destructor, which isn't needed anymore. I reverted `common/io/async/d` since the D shim is meant to pass around only raw pointers. I had to replace each instance of `foo.reset(…)` with `foo = …` . Then I made manual edits to `common/clientpool/ClientPool2-inl.h` because that code wants to store the timer in a `ThreadLocalPtr`.

Reviewed By: yfeldblum

Differential Revision: D3237530

fbshipit-source-id: 96bfb6987098618ad5430c21b1314f385f04a93d

8 years agoSuppress new GCC 5 warning -Wbool-compare
Mark Santaniello [Fri, 13 May 2016 15:41:37 +0000 (08:41 -0700)]
Suppress new GCC 5 warning -Wbool-compare

Summary: This -Wbool-compare is apparently new in GCC 5.  We should ignore it, similar to the existing suppression of -Wsign-compare.

Reviewed By: pixelb, yfeldblum

Differential Revision: D3293675

fbshipit-source-id: 5f90b0dbf049a06c0f6c3df2cdfcc8055358e367

8 years agoFix extra bytes in err message
Subodh Iyengar [Fri, 13 May 2016 07:24:05 +0000 (00:24 -0700)]
Fix extra bytes in err message

Summary:
Error message might be < 256 bytes
in which case buf.size() would be greater
than the size of the message written out
and we might have garbled bytes at the end.

ERR_error_string_n null terminates the string
so just use that.

Reviewed By: anirudhvr

Differential Revision: D3297320

fbshipit-source-id: 2ae626ce4c49ca478806c0bcc40a390e5a84f24b

8 years agoRegister custom stack for signal handlers
Andrii Grynenko [Thu, 12 May 2016 23:20:40 +0000 (16:20 -0700)]
Register custom stack for signal handlers

Summary: This updates FiberManager to automatically register custom stack for signal handlers. This ensures signal handlers will work correctly not only in case of stack overflow, but also if fiber stack size is just not enough for them.

Reviewed By: yfeldblum

Differential Revision: D3289912

fbshipit-source-id: 44959b727f70e1f1748d837b1a06c7414433e5ec

8 years agoTransfer ownership from a unique_ptr to a ThreadLocalPtr
Chad Parry [Thu, 12 May 2016 21:47:19 +0000 (14:47 -0700)]
Transfer ownership from a unique_ptr to a ThreadLocalPtr

Summary:
This `ThreadLocalPtr::reset` overload will accept a `unique_ptr`. It's actually not totally exception safe, simply because `ElementWrapper::set` is not exception safe. The best I can say is that my additional code is exactly as safe as the underlying implemenation.

liketolivedangerously

Reviewed By: ericniebler

Differential Revision: D3271563

fbshipit-source-id: 774bcf31924b1ed4e29a6cb1c0a36ad710ab6034

8 years agoPrevent leaks in ThreadLocalPtr initialization
Chad Parry [Thu, 12 May 2016 18:46:42 +0000 (11:46 -0700)]
Prevent leaks in ThreadLocalPtr initialization

Summary: While making an unrelated change, (D3271563, which was needed from an unrelated change, (D3237530)), I noticed a lack of exception safety here. If `std::bad_alloc` were thrown, then we don't want to leak memory.

Reviewed By: ericniebler

Differential Revision: D3271911

fbshipit-source-id: 0d316c0fa865a7d64622c1d62160bb0c2b061d78

8 years agoInclude limits.h in IOVec.h.
Michael Lee [Thu, 12 May 2016 16:44:29 +0000 (09:44 -0700)]
Include limits.h in IOVec.h.

Summary:
According to the POSIX definition, IOV_MAX is defined in
limits.h. Given that IOVec.h is used to provide IOV_MAX and similar,
it seems like we should add limits.h as well.

Differential Revision: D3291996

fbshipit-source-id: 238140ebef7b1b90a28f3a67a25c5f5863538fc7

8 years agoMake the sys/uio.h portability header work for Windows
Christopher Dykes [Wed, 11 May 2016 23:55:28 +0000 (16:55 -0700)]
Make the sys/uio.h portability header work for Windows

Summary: Windows doesn't have sys/uio.h, so implement it. This doesn't adjust any includes of it yet, I'll leave that for a separate diff.

Reviewed By: yfeldblum

Differential Revision: D2978042

fbshipit-source-id: 16d398a856de50b6c662f739eb11dfb851096592

8 years agowangle support for TFO
Subodh Iyengar [Wed, 11 May 2016 19:34:00 +0000 (12:34 -0700)]
wangle support for TFO

Summary: Add server support for TFO to wangle.

Reviewed By: Orvid

Differential Revision: D3275427

fbshipit-source-id: 466baa342e2e0c4751e9c96d1e660cffb90e0634

8 years agofolly: ubsan: format: avoid UB in abs(-INT_MIN)
Lucian Grijincu [Wed, 11 May 2016 17:45:33 +0000 (10:45 -0700)]
folly: ubsan: format: avoid UB in abs(-INT_MIN)

Summary:
abs(-INT_MIN) is UB. Avoid it by first casting the number to
its unsigned equivalent and then negating that.

Reviewed By: yfeldblum

Differential Revision: D3280835

fbshipit-source-id: 40922d686173e6467e15d5a6dc2b62ad718349b5

8 years agoAdd a compatibility shim for working with libevent on MSVC
Christopher Dykes [Wed, 11 May 2016 16:51:11 +0000 (09:51 -0700)]
Add a compatibility shim for working with libevent on MSVC

Summary:
MSVC builds of libevent expect `evutil_socket_t` to be `HANDLE` values, but Folly, HHVM, and Thrift all use them as file descriptors.
This adds a `folly_event_set` function that always expects a file descriptor rather than a socket. This also changes the places in Folly that use it.

Reviewed By: mzlee, yfeldblum

Differential Revision: D2874655

fbshipit-source-id: 66cfd86fd69a9fbac30d150445e4814cd5ca799e

8 years agoMake error size more explicit
Subodh Iyengar [Wed, 11 May 2016 07:14:01 +0000 (00:14 -0700)]
Make error size more explicit

Summary:
We shouldn't rely on the internal buffer size of openssl
for errors.

This makes it more explicit.

Reviewed By: knekritz

Differential Revision: D3285918

fbshipit-source-id: 452bf16a3de151d0aa79c774f2fdfc1e08ee6f9c

8 years agoAdd NETWORK_ERROR to switch statement
Neel Goyal [Tue, 10 May 2016 23:23:24 +0000 (16:23 -0700)]
Add NETWORK_ERROR to switch statement

Summary: Add NETWOR_ERROR to switch so it doesn't fall into Invalid Exception block.

Differential Revision: D3282448

fbshipit-source-id: 2cc5d21e436e6ac8f2f6581d99341ac47cf5b023

8 years agoAdd additional ALPN mismatch tests.
Kyle Nekritz [Tue, 10 May 2016 19:14:39 +0000 (12:14 -0700)]
Add additional ALPN mismatch tests.

Reviewed By: siyengar

Differential Revision: D3264727

fbshipit-source-id: fb10558096e73a3cbb6f2fb93296524c0206910d

8 years agoFix a few typos all over the code
Faidon Liambotis [Tue, 10 May 2016 18:05:26 +0000 (11:05 -0700)]
Fix a few typos all over the code

Summary:
These are pretty trivial. Debian's lint tool (lintian) warns about those
in its verbose mode, and since we saw them when packaging for Debian, we
thought of fixing them and forwarding them upstream.
Closes https://github.com/facebook/folly/pull/345

Reviewed By: Orvid

Differential Revision: D2769801

Pulled By: elliottneilclark

fbshipit-source-id: ad37cc53792f21aae6558ba256f20cb56c6caab5

8 years agodocs/Overview.md: Fixed 2 hyperlinks.
Jaak Ristioja [Tue, 10 May 2016 17:42:47 +0000 (10:42 -0700)]
docs/Overview.md: Fixed 2 hyperlinks.

Summary:
These didn't work properly in GitHub.
Closes https://github.com/facebook/folly/pull/375

Differential Revision: D3281872

Pulled By: elliottneilclark

fbshipit-source-id: b5398267ff5856bb6a7c4a3b963ae5cb47859bc3

8 years agofolly: ubsan: disable integer overflow tests in Histogram
Lucian Grijincu [Tue, 10 May 2016 06:00:19 +0000 (23:00 -0700)]
folly: ubsan: disable integer overflow tests in Histogram

Summary:
It would be nice to fix, but it's going to be easier to do when folly
support is for GCC>=5 which adds `__builtin_add_overflow`.

For now disable the check as the Histgram code handles overflow
correctly and has tests for these cases.

Fixes need to be done for both float and integer types as Histogram is used with both.

Reviewed By: meyering

Differential Revision: D3280260

fbshipit-source-id: 56e524c517566a4547346859be7770eda2acee96

8 years agofolly: ubsan: cast to unsigned to avoid negative-left-shift
Lucian Grijincu [Tue, 10 May 2016 04:16:26 +0000 (21:16 -0700)]
folly: ubsan: cast to unsigned to avoid negative-left-shift

Summary:
The result of E1 << E2 is E1 left-shifted E2 bit positions; vacated
  bits are filled with zeros.

  If E1 has a signed type and non-negative value, and E1 × 2E2 is
  representable in the result type, then that is the resulting value;
  otherwise, the behavior is undefined.

The code assumed the signed left shift worked like an unsigned left
shift, so make that explicit.

Reviewed By: meyering

Differential Revision: D3280325

fbshipit-source-id: 178b95ee36b7a1126a79bb825c2ad2ffa6fa9464

8 years agofolly: ubsan: fix "reference binding to null pointer of type 'char'"
Lucian Grijincu [Tue, 10 May 2016 00:50:08 +0000 (17:50 -0700)]
folly: ubsan: fix "reference binding to null pointer of type 'char'"

Summary:
third-party-buck/build/libgcc/include/c++/4.9.x/bits/stl_vector.h:866:9: runtime error: reference binding to null pointer of type 'char'

```
Breakpoint 2, 0x0000000000494894 in __ubsan_handle_type_mismatch_abort ()
(gdb) bt
```

Reviewed By: meyering, yfeldblum

Differential Revision: D3279234

fbshipit-source-id: 63a761587e5b3f79ece09fc99b9a593da0e44b75

8 years agoDeal with some oddities of MSVC's preprocessor
Christopher Dykes [Tue, 10 May 2016 00:23:41 +0000 (17:23 -0700)]
Deal with some oddities of MSVC's preprocessor

Summary: MSVC's preprocessor is different in how it deals with passing arguments expanded from `__VA_ARGS__`, so we have to add a stub to force it to expand the arguments.

Reviewed By: yfeldblum

Differential Revision: D3256328

fbshipit-source-id: 551434833d40e55498a9ab352eb74acdfd094835

8 years agoFix detection of std::make_unique under MSVC
Christopher Dykes [Tue, 10 May 2016 00:15:22 +0000 (17:15 -0700)]
Fix detection of std::make_unique under MSVC

Summary: MSVC doesn't define `__cplusplus` as a high enough value to trigger this, so add an explicit check for MSVC instead.

Reviewed By: yfeldblum

Differential Revision: D3271647

fbshipit-source-id: a1e5a5a7eb75dce066dfc7fae8b2086880dc4c3d

8 years agofolly: ubsan: &v[0] is undefined if v.empty()
Lucian Grijincu [Tue, 10 May 2016 00:03:34 +0000 (17:03 -0700)]
folly: ubsan: &v[0] is undefined if v.empty()

Summary:
before:

  [ RUN      ] RangeFunc.Vector
  bits/stl_vector.h:866:9: runtime error: reference binding to null pointer of type 'int'

Reviewed By: yfeldblum

Differential Revision: D3279253

fbshipit-source-id: 957a8ec050f5c3b27febf7e6cd16ad1d307da169

8 years agofolly test: fix typo in RandomDataHolder::RandomDataHolder
Jim Meyering [Mon, 9 May 2016 21:05:43 +0000 (14:05 -0700)]
folly test: fix typo in RandomDataHolder::RandomDataHolder

Summary:
Running the folly/io/test:compression_test with a UBSAN-enabled binary would fail with this:

  folly/io/test/CompressionTest.cpp:96:36: runtime error: shift exponent 16777216 is too large for 64-bit type 'size_t' (aka 'unsigned long')

That exposed the unwanted left-shift.
That made it so the actual "start" offsets were in the range [0..numThreads-1], rather than being
e.g., with numThreads == 8, [0, 1677721633554432, ...].  Oops.

Reviewed By: luciang

Differential Revision: D3277606

fbshipit-source-id: 4bc48f84b0065de070b62a810f08b96e36ea7136

8 years agoInclude <nmmintrin.h> in GroupVarint.h
Christopher Dykes [Mon, 9 May 2016 19:57:08 +0000 (12:57 -0700)]
Include <nmmintrin.h> in GroupVarint.h

Summary: Because it is the standard (well, Intel's standard) location for __m128i, and GCC, Clang, MSVC and ICC all implement it in that location.

Reviewed By: yfeldblum

Differential Revision: D3270872

fbshipit-source-id: fa0f0410670cdb310e656ca55ee0f568ca5ee06f

8 years agofolly: fbvector: ubsan: avoid memcpy(dest, nullptr, 0)
Lucian Grijincu [Mon, 9 May 2016 19:16:40 +0000 (12:16 -0700)]
folly: fbvector: ubsan: avoid memcpy(dest, nullptr, 0)

Reviewed By: meyering

Differential Revision: D3277138

fbshipit-source-id: 7050d18c256ac15c8122d4af59aa31208b1466fd

8 years agoExplicitly reference folly::detail
Christopher Dykes [Mon, 9 May 2016 18:36:40 +0000 (11:36 -0700)]
Explicitly reference folly::detail

Summary: Due to some fun with the order MSVC does name lookup in templates, it ends up complaining that these are ambigious between folly::detail and folly::recordio_helpers::detail. The solution is to simply reference folly::detail explicitly.

Reviewed By: yfeldblum

Differential Revision: D3271606

fbshipit-source-id: e599ad8d8fe33b1e4ec0b838ddb5dfacfdf60159

8 years agoFix a few issues in the portability headers
Christopher Dykes [Mon, 9 May 2016 18:36:21 +0000 (11:36 -0700)]
Fix a few issues in the portability headers

Summary:
This fixes a few issues in the portability headers for MSVC.
Not entirely sure how I managed to cause these in the first place, but this fixes them anyways.

Reviewed By: yfeldblum

Differential Revision: D3271859

fbshipit-source-id: 11a5d35246f29112563ee9079525aa37ced206ff

8 years agoReverted commit D3265572
Philip Pronin [Mon, 9 May 2016 07:06:57 +0000 (00:06 -0700)]
Reverted commit D3265572

Summary:
I would switch these to just use the intrinsic functions, but GCC 4.8 doesn't support them.
MSVC supports the intrinsics, which is the primary reason for the switch.

Reviewed By: yfeldblum

Differential Revision: D3265572

fb-gh-sync-id: 3268f90d5234bdf77b3579504efd3dcd43f38aff
fbshipit-source-id: 3268f90d5234bdf77b3579504efd3dcd43f38aff

8 years agoDon't re-throw exceptions from default FiberManager exception handler
Andrii Grynenko [Sat, 7 May 2016 05:02:15 +0000 (22:02 -0700)]
Don't re-throw exceptions from default FiberManager exception handler

Summary: It's very unlikely that caller will handle such exceptions and FiberManager can proceed even if one task failed. DFATAL will ensure we still fail on such exceptions in debug builds.

Reviewed By: yfeldblum

Differential Revision: D3273573

fb-gh-sync-id: 55eba19c16429a3c89a2fdf5363fd4b55b3998a2
fbshipit-source-id: 55eba19c16429a3c89a2fdf5363fd4b55b3998a2

8 years agoDon't throw on late singleton registration
Andrii Grynenko [Fri, 6 May 2016 17:26:04 +0000 (10:26 -0700)]
Don't throw on late singleton registration

Summary: Having singleton registered after registrationComplete() is called is not a hard failure. Replacing exception with ERROR, since we probably still want to avoid such situation.

Reviewed By: yfeldblum

Differential Revision: D3266376

fb-gh-sync-id: 8130d279775348c835265987c95189bd3d8106f9
fbshipit-source-id: 8130d279775348c835265987c95189bd3d8106f9

8 years agoUse intrinsics rather than inline assembly where possible
Christopher Dykes [Fri, 6 May 2016 16:32:30 +0000 (09:32 -0700)]
Use intrinsics rather than inline assembly where possible

Summary:
I would switch these to just use the intrinsic functions, but GCC 4.8 doesn't support them.
MSVC supports the intrinsics, which is the primary reason for the switch.

Reviewed By: yfeldblum

Differential Revision: D3265572

fb-gh-sync-id: f12d16b8f64cf48adcb97958d7ae88d56689a96f
fbshipit-source-id: f12d16b8f64cf48adcb97958d7ae88d56689a96f

8 years agoCreate the stdio.h portability header
Christopher Dykes [Fri, 6 May 2016 16:31:20 +0000 (09:31 -0700)]
Create the stdio.h portability header

Summary: Windows doesn't define a couple of functions.

Reviewed By: yfeldblum

Differential Revision: D2990676

fb-gh-sync-id: 1a1354bd7976cb8b4747436d6907d65210ba506c
fbshipit-source-id: 1a1354bd7976cb8b4747436d6907d65210ba506c

8 years agoBind to a random file in unix domain socket test
Haijun Zhu [Fri, 6 May 2016 09:20:30 +0000 (02:20 -0700)]
Bind to a random file in unix domain socket test

Summary:
test console reports "address already in use" failure so add some
randomness to it.

Reviewed By: yfeldblum

Differential Revision: D3265329

fb-gh-sync-id: a825cce787394783bb8d18d92385d2497b163385
fbshipit-source-id: a825cce787394783bb8d18d92385d2497b163385

8 years agoCreate the unistd.h portability header
Christopher Dykes [Fri, 6 May 2016 00:24:55 +0000 (17:24 -0700)]
Create the unistd.h portability header

Summary: Windows doesn't have it, and it's used in a lot of places.

Reviewed By: yfeldblum

Differential Revision: D2997499

fb-gh-sync-id: fbf8f5f087b8f39197d5bf2177e4d34efd74631f
fbshipit-source-id: fbf8f5f087b8f39197d5bf2177e4d34efd74631f

8 years agoAllow optional atfork hook for singleton destruction
Andrii Grynenko [Thu, 5 May 2016 23:41:33 +0000 (16:41 -0700)]
Allow optional atfork hook for singleton destruction

Summary: Grouping all folly::ThreadLocal atfork hooks in one. This allows proper ordering between folly::ThreadLocal hooks and hooks added by other libraries (we always want folly::ThreadLocal hooks to be registered first).

Reviewed By: ericniebler

Differential Revision: D3262666

fb-gh-sync-id: 1aa928b9ddd484580f4a4a7b320e9e64164bb04a
fbshipit-source-id: 1aa928b9ddd484580f4a4a7b320e9e64164bb04a

8 years agoImplement LoopKeepAlive for EventBase
Andrii Grynenko [Thu, 5 May 2016 20:22:42 +0000 (13:22 -0700)]
Implement LoopKeepAlive for EventBase

Summary: LoopKeepAlive can be useful to tell EventBase, that loop() shouldn't return even if there are no events registered, because some tasks will later be added via runInEventBaseThread. While at least one LoopKeepAlive is alive - EventBase::loop() behaves like EventBase::loopForever().

Reviewed By: yfeldblum

Differential Revision: D3261706

fb-gh-sync-id: d91424d3d12cae11abd72cffdbd57f136f628dae
fbshipit-source-id: d91424d3d12cae11abd72cffdbd57f136f628dae

8 years agoAllow adding tasks to TaskIterator dynamically
Andrii Grynenko [Thu, 5 May 2016 01:15:53 +0000 (18:15 -0700)]
Allow adding tasks to TaskIterator dynamically

Reviewed By: yfeldblum

Differential Revision: D3244669

fb-gh-sync-id: 73fa4ecb0432a802e67ef922255a896d96f32374
fbshipit-source-id: 73fa4ecb0432a802e67ef922255a896d96f32374

8 years agoSwitch gflags to portability/GFlags.h
Michael Lee [Thu, 5 May 2016 00:08:47 +0000 (17:08 -0700)]
Switch gflags to portability/GFlags.h

Summary:
There are two places that still includes gflags/gflags.h:
portability/GFlags.h and futures/exercises/Koan.h. Also removing
gflags entirely from a couple of places which never reference it at
all.

Reviewed By: yfeldblum

Differential Revision: D3203187

fb-gh-sync-id: e7725fbf2c4b3c475c0035e1b0e8c6e1227f277f
fbshipit-source-id: e7725fbf2c4b3c475c0035e1b0e8c6e1227f277f

8 years agoCreate the stdlib.h portability header
Christopher Dykes [Wed, 4 May 2016 23:45:56 +0000 (16:45 -0700)]
Create the stdlib.h portability header

Summary: Windows doesn't have these functions under these names, so be nice to the code using it.

Reviewed By: yfeldblum

Differential Revision: D3003641

fb-gh-sync-id: 1657545af6975c96aea4a48c359250828b5921e0
fbshipit-source-id: 1657545af6975c96aea4a48c359250828b5921e0

8 years agoCreate the fcntl.h portability header
Christopher Dykes [Wed, 4 May 2016 23:45:53 +0000 (16:45 -0700)]
Create the fcntl.h portability header

Summary: Windows has it, but all it does is define a few constants.

Reviewed By: yfeldblum

Differential Revision: D3002161

fb-gh-sync-id: a719faf391125522b2cb21471a56e8a9fe894225
fbshipit-source-id: a719faf391125522b2cb21471a56e8a9fe894225

8 years agorearrange folly::Function so that its template arguments are deducable.
Eric Niebler [Wed, 4 May 2016 21:45:20 +0000 (14:45 -0700)]
rearrange folly::Function so that its template arguments are deducable.

Summary:
`folly::Function` was an alias to a more complex type with template arguments that could not be deduced. For example, the call to `foo` below was failing to compile.

```
template <class R, class... As>
void foo(folly::Function<R(As...)> f) {
}

int main() {
  foo( folly::Function<void(int)>{ [](int i){} } );
}
```

Rearrange the code so that folly::Function is no longer an alias, thus making its template arguments deducable.

Reviewed By: luciang, spacedentist

Differential Revision: D3256130

fb-gh-sync-id: fb403e48d161635b3b7f36e53b1679eb46cbfe7f
fbshipit-source-id: fb403e48d161635b3b7f36e53b1679eb46cbfe7f

8 years agofolly::Future integration with Python
Andrii Grynenko [Wed, 4 May 2016 21:32:54 +0000 (14:32 -0700)]
folly::Future integration with Python

Summary:
This diff implements two functions which can be used to bridge C++ asynchronous APIs with Python.
1. getExecutor(event_loop) allows extracting a folly::Executor-compatible object from Python event_loop.
2. wrapFuture wraps any folly::Future object in a Python Future object.

Reviewed By: ambv

Differential Revision: D2878731

fb-gh-sync-id: 83628b6b24d6039999f88c5de441e41064b2417e
fbshipit-source-id: 83628b6b24d6039999f88c5de441e41064b2417e

8 years agoFix the weak references to JEMalloc under MSVC
Christopher Dykes [Wed, 4 May 2016 17:29:15 +0000 (10:29 -0700)]
Fix the weak references to JEMalloc under MSVC

Summary: While the previous version did compile, it would generate warnings about symbols being defined in multiple places.

Reviewed By: yfeldblum

Differential Revision: D3255815

fb-gh-sync-id: 72f5079a3973a04bdc0b469b97ba46181e40353b
fbshipit-source-id: 72f5079a3973a04bdc0b469b97ba46181e40353b

8 years agoImplement __builtin_ctzll for MSVC
Christopher Dykes [Wed, 4 May 2016 17:28:49 +0000 (10:28 -0700)]
Implement __builtin_ctzll for MSVC

Summary: MSVC doesn't have it, but Folly uses it, so implement it in the Builtins portability header.

Reviewed By: yfeldblum

Differential Revision: D3256123

fb-gh-sync-id: fd9ea1b6098d97cf1fde4732905bae9bde8cd8ad
fbshipit-source-id: fd9ea1b6098d97cf1fde4732905bae9bde8cd8ad

8 years agoDon't initialize kDestroyWaitTime in-class
Christopher Dykes [Wed, 4 May 2016 17:28:12 +0000 (10:28 -0700)]
Don't initialize kDestroyWaitTime in-class

Summary: MSVC doesn't like it when you try to initialize a static in-class non-primitive constexpr member. It was only used in one place, so stick the definition in that function instead.

Reviewed By: yfeldblum

Differential Revision: D3255178

fb-gh-sync-id: e4a862a2662c4f5376c19fc52827cb2953f01e52
fbshipit-source-id: e4a862a2662c4f5376c19fc52827cb2953f01e52

8 years agoUse std::thread::id in Fiber rather than a syscall
Christopher Dykes [Wed, 4 May 2016 17:27:25 +0000 (10:27 -0700)]
Use std::thread::id in Fiber rather than a syscall

Summary: syscall() doesn't exist on Windows, but std::thread::id is standardized, and can do what we need it for, so use it instead.

Reviewed By: yfeldblum

Differential Revision: D3256064

fb-gh-sync-id: efddac82c117ccd8a53c8c93248529000b4326a5
fbshipit-source-id: efddac82c117ccd8a53c8c93248529000b4326a5

8 years agoDon't attempt to implicitly convert an atomic<T> to bool
Christopher Dykes [Wed, 4 May 2016 17:26:43 +0000 (10:26 -0700)]
Don't attempt to implicitly convert an atomic<T> to bool

Summary: MSVC doesn't like it, so the other uses were changed to .load() explicitly a while back, but this use has been added since then.

Reviewed By: yfeldblum

Differential Revision: D3255261

fb-gh-sync-id: 11d49902eb3ef7113a7c57cc760353554a487496
fbshipit-source-id: 11d49902eb3ef7113a7c57cc760353554a487496

8 years agoUnpoison stack memory before deallocation
Marcus Holland-Moritz [Wed, 4 May 2016 10:28:44 +0000 (03:28 -0700)]
Unpoison stack memory before deallocation

Summary:
This is a workaround (maybe even the correct fix if it turns out that ASan
can't detect this FP case) for the problem described in

  https://llvm.org/bugs/show_bug.cgi?id=27627

where a memory region previously allocated by a fiber stack can overlap with
the region of an mmap'd file. Accessing parts of the mmap'd file close to
the stack region will trigger a false positive ASan error.

This change makes sure each fiber explicitly unpoisons its stack memory by
calling __asan_unpoison_memory_region in an ASan-enabled build.

Reviewed By: yhfung

Differential Revision: D3257924

fb-gh-sync-id: 484062e80af67dfd39d2eaf3cbb52fa3483924eb
fbshipit-source-id: 484062e80af67dfd39d2eaf3cbb52fa3483924eb

8 years agoOptimize copy/move
Pavlo Kushnir [Wed, 4 May 2016 07:27:53 +0000 (00:27 -0700)]
Optimize copy/move

Summary: this diff provides methods that allow to avoid temporaries while cloning IOBufs. Also optimizes move constructor a bit - there is no need to call `decrementRefCount`, check for self assignment, etc.

Reviewed By: yfeldblum

Differential Revision: D3250456

fb-gh-sync-id: 32b0180c28f36151c6786dba6f511b491b224200
fbshipit-source-id: 32b0180c28f36151c6786dba6f511b491b224200

8 years agoLet via take a functor by universal reference
Yedidya Feldblum [Wed, 4 May 2016 04:32:23 +0000 (21:32 -0700)]
Let via take a functor by universal reference

Summary:
[Folly] Let `via` take a functor by universal reference.

And perfect-forward the functor to the call to `then`.

Reviewed By: fugalh

Differential Revision: D3255684

fb-gh-sync-id: 1b46a70fb9de88d21d785c0ce4d9dd3078829f13
fbshipit-source-id: 1b46a70fb9de88d21d785c0ce4d9dd3078829f13

8 years agoCreate the sockets portability header
Christopher Dykes [Wed, 4 May 2016 00:38:24 +0000 (17:38 -0700)]
Create the sockets portability header

Summary: The one header that doesn't split well. This is the last piece of the portability headers for Windows and is also the piece that has had the most thought put into it's design so-as to be as unintrusive as possible.

Reviewed By: yfeldblum

Differential Revision: D3046525

fb-gh-sync-id: 761a5e359bc130b95585608ee18fda94a30dae2c
fbshipit-source-id: 761a5e359bc130b95585608ee18fda94a30dae2c

8 years agoMore useful logging upon failed AsyncServerSocket::bind()
Marcelo Juchem [Tue, 3 May 2016 22:35:30 +0000 (15:35 -0700)]
More useful logging upon failed AsyncServerSocket::bind()

Summary: `AsyncServerSocket::bind()` would not give information like port or family name when failing to bind a socket. This diff addresses that by including this information in the exception. Two additional helper methods were added to `SocketAddress` to retrieve both the port and the family name from a `sockaddr` structure.

Reviewed By: ckwalsh, yfeldblum

Differential Revision: D3249778

fb-gh-sync-id: 4edb28af5c211b7bf8d525b40844a5b0b6261e07
fbshipit-source-id: 4edb28af5c211b7bf8d525b40844a5b0b6261e07

8 years agofix passing move-only types to via(Executor*, Func)
Sven Over [Tue, 3 May 2016 21:27:37 +0000 (14:27 -0700)]
fix passing move-only types to via(Executor*, Func)

Summary: This diff fixes a problem with passing move-only types to folly::via.

Reviewed By: ericniebler

Differential Revision: D3254906

fb-gh-sync-id: 8a9c703a8db0ccf20b9fb4fe9b80ad6cdcb3e388
fbshipit-source-id: 8a9c703a8db0ccf20b9fb4fe9b80ad6cdcb3e388

8 years agoDeprecate boost::mpl
Yedidya Feldblum [Tue, 3 May 2016 09:15:22 +0000 (02:15 -0700)]
Deprecate boost::mpl

Summary:
[Folly] Deprecate `boost::mpl`.

The `std` has equivalents; prefer them.

Reviewed By: Gownta

Differential Revision: D3247294

fb-gh-sync-id: 873697b7ea3ad069ff14662bc458a2e1cc840339
fbshipit-source-id: 873697b7ea3ad069ff14662bc458a2e1cc840339

8 years agoUse portability for SYS_gettid
Michael Lee [Mon, 2 May 2016 23:14:23 +0000 (16:14 -0700)]
Use portability for SYS_gettid

Summary: It's there for a reason, so use it.

Reviewed By: yfeldblum

Differential Revision: D3249849

fb-gh-sync-id: e9fe330693a173acd987e0c8fd1f03a982c8e082
fbshipit-source-id: e9fe330693a173acd987e0c8fd1f03a982c8e082

8 years agoDon't try to use the malloc_usable_size portability implementation if we have JEMalloc
Christopher Dykes [Mon, 2 May 2016 19:22:28 +0000 (12:22 -0700)]
Don't try to use the malloc_usable_size portability implementation if we have JEMalloc

Summary:
This is only an issue for the OSS HHVM build on OSX, where we have JEMalloc headers included directly.
In the OSS HHVM build, we define `USE_JEMALLOC` globally if we have JEMalloc, so gate the portability implementation behind that.

Reviewed By: mzlee

Differential Revision: D3237449

fb-gh-sync-id: 07cf9b31114723ffd9f2b29cf449bee0ed8f5703
fbshipit-source-id: 07cf9b31114723ffd9f2b29cf449bee0ed8f5703

8 years agoSet the appropriate AsyncSocketExceptionType from SSLException
Neel Goyal [Mon, 2 May 2016 16:30:18 +0000 (09:30 -0700)]
Set the appropriate AsyncSocketExceptionType from SSLException

Summary:
We should use the appropriate AsyncSocketExceptionType when firing SSLExceptions.  This derives the type from the constructor args, and introduces a `NETWORK_ERROR` exception type.  Open to suggestions on reusing something there.

Additionally, rename getType() to getSSLError() to prevent hiding the AsyncSocketException::getType.

Reviewed By: siyengar

Differential Revision: D3241204

fb-gh-sync-id: de631dfb3554177d5bd561f4b91e767c882767d3
fbshipit-source-id: de631dfb3554177d5bd561f4b91e767c882767d3

8 years agoSimplify CryptoSSLRequestManager using unified futures API
Anirudh Ramachandran [Mon, 2 May 2016 16:12:01 +0000 (09:12 -0700)]
Simplify CryptoSSLRequestManager using unified futures API

Summary:
This diff unifies the nearly-replicated call paths for ECDSA and RSA
sign request/response/fallback/etc. into one generic futures-based sign API.
Once this is tested, we should be able to remove ~500 lines of code from
CryptoSSLRequestManager

Depends on D3094660

Reviewed By: siyengar

Differential Revision: D3116535

fb-gh-sync-id: 11c50d81dab39e5426679b5a172c1ece2039fc10
fbshipit-source-id: 11c50d81dab39e5426679b5a172c1ece2039fc10

8 years agoAvoid glog when warning about Singleton double-registration
Yedidya Feldblum [Mon, 2 May 2016 09:26:07 +0000 (02:26 -0700)]
Avoid glog when warning about Singleton double-registration

Summary:
[Folly] Avoid glog when warning about `Singleton` double-registration.

Because registration happens at static initialization time, long before main, and possibly long before glog is initialized.

This makes a difference because, in some cases of double-registration, we can get a SIGSEGV with no message, because we are attempting to `LOG(FATAL)` when glog is uninitialized. What we would much rather have is a SIGABRT with a message.

Reviewed By: meyering

Differential Revision: D3245047

fb-gh-sync-id: 4c5dd9d25025f197d7c490ffbb429af5ccb82182
fbshipit-source-id: 4c5dd9d25025f197d7c490ffbb429af5ccb82182

8 years agoDon't require folly::Unit in addTaskFinally
Andrii Grynenko [Sat, 30 Apr 2016 00:50:50 +0000 (17:50 -0700)]
Don't require folly::Unit in addTaskFinally

Summary:
In D3241498 addTaskFuture was fixed to work with function returning void. This however changed addTaskFinally API to not allow finally functor which accepts Try<void>.

folly::fibers generally supports Try<void> so there's no reason to force users to use folly::Unit instead of void for addTaskFinally too.

Reviewed By: yfeldblum

Differential Revision: D3243893

fb-gh-sync-id: d1df54738157d2019120103956f59b3971ba25ff
fbshipit-source-id: d1df54738157d2019120103956f59b3971ba25ff

8 years agoMake addTask[Remote]Future() work for functions returning void.
Martin Martin [Fri, 29 Apr 2016 18:54:34 +0000 (11:54 -0700)]
Make addTask[Remote]Future() work for functions returning void.

Summary:
There's no Future<void>, you're supposed to use Future<Unit>
instead.  Unit has a "Lift" template to do the conversion.

Reviewed By: andriigrynenko

Differential Revision: D3241498

fb-gh-sync-id: db12d4f343685bc613b701e023c860c2c903ed4c
fbshipit-source-id: db12d4f343685bc613b701e023c860c2c903ed4c

8 years agoClang-format in preparation for other change
Martin Martin [Fri, 29 Apr 2016 17:50:23 +0000 (10:50 -0700)]
Clang-format in preparation for other change

Summary: Clang-format in preparation for other change

Reviewed By: andriigrynenko

Differential Revision: D3241297

fb-gh-sync-id: b7b26812e9e61c291d5c7bdb523df8f28f2d9b4f
fbshipit-source-id: b7b26812e9e61c291d5c7bdb523df8f28f2d9b4f

8 years agodon't destory defaultElfCache in exit handlers
Philip Pronin [Fri, 29 Apr 2016 00:21:06 +0000 (17:21 -0700)]
don't destory defaultElfCache in exit handlers

Summary:
As it might be used during global destruction (for example, to
symbolize unhandled exception happened at that time).

Reviewed By: luciang, ot

Differential Revision: D3238303

fb-gh-sync-id: 39d8a6f99d95e700586349436e961f3624c95718
fbshipit-source-id: 39d8a6f99d95e700586349436e961f3624c95718

8 years agouse std::forward instead of std::move on objects whose types have been deduced; don...
Eric Niebler [Thu, 28 Apr 2016 23:39:30 +0000 (16:39 -0700)]
use std::forward instead of std::move on objects whose types have been deduced; don't take the sizeof incomplete types

Summary: Pretty sure std::forward is needed here instead of std::move. If you trace the call chain to see where the types of the objects come from, you'll see they can be deduced to be lvalues, so std::forward is the right choice. Also, moved some dicey looking code that appeared to be taking the size of some incomkplete types.

Reviewed By: spacedentist

Differential Revision: D3214199

fb-gh-sync-id: 778190ffb25a648b839760a3dddfad8dc6d41c88
fbshipit-source-id: 778190ffb25a648b839760a3dddfad8dc6d41c88

8 years agoCreate ReadHolder::unlock
Andrew Birchall [Thu, 28 Apr 2016 23:39:10 +0000 (16:39 -0700)]
Create ReadHolder::unlock

Summary:
Currently you need to depend on the destructor of `ReadHolder` (using closures as in code block #1 below or empty assignment as in code block #2 below) to ensure that a `ReadHolder` goes out of scope (and unlocks) in order to subsequently acquire a write lock via `WriteHolder` without deadlocking.
This diff introduces a way of unlocking a `ReadHolder` while it's still in scope such that a `WriteHolder` can be acquired. This makes the code more straight forward (reducing the risk of deadlock due to a programmer's misunderstanding of how `SharedMutex` / the holder framework works) => see code block # 3 below
Also add some documentation about why `WriteHolder::WriteHolder(ReadHolder&&)` doesn't exist

Code Block #1 : Use of closures
```
class foo {
 public:
  std::string getMemoizedData() {
    {
      folly::SharedMutex::ReadHolder readHolder(lock_);
      if (!data_.empty()) {
        // important to return by value, otherwise caller might access
        // data_ after we release the read lock
        return data_;
      }
    }
    {
      // try again with a write lock
      folly::SharedMutex::WriteHolder writeHolder(lock_);
      if (data_.empty()) {
        data_ = "my awesome string";
      }
      return data_;
    }
  }

 private:
  folly::SharedMutex lock_;
  std::string data_;
};
```

Code Block #2 : Use of empty assignment
```
class foo {
 public:
  std::string getMemoizedData() {
    folly::SharedMutex::ReadHolder readHolder(lock_);
    if (!data_.empty()) {
      // important to return by value, otherwise caller might access
      // data_ after we release the read lock
      return data_;
    }
    readHolder = {};

    // try again with a write lock
    folly::SharedMutex::WriteHolder writeHolder(lock_);
    if (data_.empty()) {
      data_ = "my awesome string";
    }
    return data_;
  }

 private:
  folly::SharedMutex lock_;
  std::string data_;
};
```

Code Block #3 : Use of unlock()
```
class foo {
 public:
  std::string getMemoizedData() {
    folly::SharedMutex::ReadHolder readHolder(lock_);
    if (!data_.empty()) {
      // important to return by value, otherwise caller might access
      // data_ after we release the read lock
      return data_;
    }
    readHolder->unlock();

    // try again with a write lock
    folly::SharedMutex::WriteHolder writeHolder(lock_);
    if (data_.empty()) {
      data_ = "my awesome string";
    }
    return data_;
  }

 private:
  folly::SharedMutex lock_;
  std::string data_;
};
```

Reviewed By: yfeldblum

Differential Revision: D3176025

fb-gh-sync-id: c7d47ca71df08673c7c1f1fd5ed9e01a663c1797
fbshipit-source-id: c7d47ca71df08673c7c1f1fd5ed9e01a663c1797

8 years agofolly: Function: in-class init, drop (void*) casts
Lucian Grijincu [Thu, 28 Apr 2016 21:34:27 +0000 (14:34 -0700)]
folly: Function: in-class init, drop (void*) casts

Summary: A few code cleanups, no functionality changes.

Reviewed By: ericniebler

Differential Revision: D3231369

fb-gh-sync-id: 1a1c8508aab6dce7279e9dfc3f3da6add5496c67
fbshipit-source-id: 1a1c8508aab6dce7279e9dfc3f3da6add5496c67

8 years agoBetter exception types to SSLexception
Subodh Iyengar [Thu, 28 Apr 2016 19:10:12 +0000 (12:10 -0700)]
Better exception types to SSLexception

Summary:
Better enum exception types for SSLException.
These enum types are fairly high level, and more
info can be gleaned from the error.what()

Reviewed By: knekritz

Differential Revision: D3234501

fb-gh-sync-id: 7cd4fbccd7f4367354dc3bd1fe4cd480d58d6173
fbshipit-source-id: 7cd4fbccd7f4367354dc3bd1fe4cd480d58d6173

8 years agoStop abusing errno
Subodh Iyengar [Thu, 28 Apr 2016 19:10:01 +0000 (12:10 -0700)]
Stop abusing errno

Summary:
We abuse errno to propagate exceptions from AsyncSSLSocket.
Stop doing this and propagate exceptions correctly.

This also formats the exception messages better.

Reviewed By: anirudhvr

Differential Revision: D3226808

fb-gh-sync-id: 15a5e67b0332136857e5fb85b1765757e548e040
fbshipit-source-id: 15a5e67b0332136857e5fb85b1765757e548e040

8 years agoStrict validation for certs
Subodh Iyengar [Thu, 28 Apr 2016 18:12:54 +0000 (11:12 -0700)]
Strict validation for certs

Summary: Add X509 param types

Reviewed By: knekritz

Differential Revision: D3230008

fb-gh-sync-id: 660a2f5bdc9355f5b42c2ccfb1d7cab83df34fc2
fbshipit-source-id: 660a2f5bdc9355f5b42c2ccfb1d7cab83df34fc2

8 years agoFix asan build
Nicholas Ormrod [Thu, 28 Apr 2016 17:41:19 +0000 (10:41 -0700)]
Fix asan build

Summary:
The array-bounds checker in gcc complains that r could be -1 after leaving CHECK_ERR in ASAN mode. Switch to DCHECK to avoid this possibility. The code was not initially compiling; now it does.

Error from gcc:
  array subscript is below array bounds [-Werror=array-bounds]

Reviewed By: yfeldblum

Differential Revision: D3232694

fb-gh-sync-id: 17f579eb152b661a8e5e4ed29ef91eec405fb90a
fbshipit-source-id: 17f579eb152b661a8e5e4ed29ef91eec405fb90a

8 years agoUse the portability header for malloc_usable_size in small_vector.h
Christopher Dykes [Thu, 28 Apr 2016 17:17:19 +0000 (10:17 -0700)]
Use the portability header for malloc_usable_size in small_vector.h

Summary: `small_vector.h` had it's own thing being defined inline, and it was breaking the OSX build of HHVM, so have it use the portability header instead.

Reviewed By: yfeldblum

Differential Revision: D3232885

fb-gh-sync-id: 2078a54d1ec9700e81e0a9f4baed8e2d623fa5f3
fbshipit-source-id: 2078a54d1ec9700e81e0a9f4baed8e2d623fa5f3

8 years agoFix races in TLRefCount
Andrii Grynenko [Thu, 28 Apr 2016 04:00:57 +0000 (21:00 -0700)]
Fix races in TLRefCount

Summary:
This fixes 2 races in TLRefCount:
1. Thread-local constructor race, exposed by the stress test. It was possible for LocalRefCount to be created (grabbing collectGuard), but not be added to the thread-local list, so that accessAllThreads wasn't collecting it. collectAll() was then blocking waiting on baton to be posted, causing a dead-lock.
2. LocalRefCount::count_ has to be made atomic, because otherwise += operation may be not flushed (nbronson explained the race in D3133443).

Reviewed By: djwatson

Differential Revision: D3166956

fb-gh-sync-id: 17d58a215ebfc572f8316ed46bafaa5e6a9e2368
fbshipit-source-id: 17d58a215ebfc572f8316ed46bafaa5e6a9e2368

8 years agoFurther Function simplification
Giuseppe Ottaviano [Thu, 28 Apr 2016 00:47:43 +0000 (17:47 -0700)]
Further Function simplification

Summary:
Use tag dispatching instead of `enable_if`: it is clearer, it
sidesteps the GCC mangling bug, and more importantly the conditional
doesn't leak into the symbol, making stack traces and profiles more
readable.

Testing on a compilation unit with 1000 `Function`s from simple lambdas.
Before:
```
folly::impl::Function<int (), false>::Function<main::{lambda()#1}, {lambda()#1}>(main::{lambda()#1}&&, std::enable_if<std::integral_constant<bool, ((sizeof (std::decay<main::{lambda()#1}>::type))<=(sizeof folly::detail::function::Data::small))&&std::is_nothrow_move_constructible<std::decay<main::{lambda()#1}> >::value>::value, folly::detail::Tag>::type)::Ops::call(folly::detail::function&)
```
After:
```
folly::impl::Function<int (), false>::OpsSmall<main::{lambda()#1}>::call(folly::detail::function::Data&)
```
Note that the function type is repeated 5 times before, and only once after. This makes a large difference with long namespaces.

Binary size is almost unaffected, compile times slightly better:
Before:
GCC opt: 22.3 seconds, 4435232 bytes
Clang dev: 7.7 seconds, 5257344 bytes

After:
GCC opt: 18.6 seconds, 4493920 bytes
Clang dev: 7.2 seconds, 5469136 bytes

Reviewed By: ericniebler

Differential Revision: D3231530

fb-gh-sync-id: 6aa76e7f780a8afdbfed8a378f257ceb86dce704
fbshipit-source-id: 6aa76e7f780a8afdbfed8a378f257ceb86dce704

8 years agoFaster onDestroy
Pavlo Kushnir [Thu, 28 Apr 2016 00:45:07 +0000 (17:45 -0700)]
Faster onDestroy

Summary: there is no need to call `std::function::invoke` for every DestructorGuard.

Reviewed By: yfeldblum

Differential Revision: D3229345

fb-gh-sync-id: c42f8cd05576d56b6a9b2f9d06878d9b01a36e94
fbshipit-source-id: c42f8cd05576d56b6a9b2f9d06878d9b01a36e94

8 years agoHave the GFlags portability header stick things in namespaces
Christopher Dykes [Wed, 27 Apr 2016 21:34:33 +0000 (14:34 -0700)]
Have the GFlags portability header stick things in namespaces

Summary:
Just like the real thing does.
This was causing some issues when trying to link against glog, which expects the flags to be namespaced like this.

Reviewed By: mzlee

Differential Revision: D3230630

fb-gh-sync-id: a73ab3044560d561a39eb91ceee1588c147a46c5
fbshipit-source-id: a73ab3044560d561a39eb91ceee1588c147a46c5

8 years agoReimplement folly::Function to improve compile times.
Eric Niebler [Wed, 27 Apr 2016 17:02:00 +0000 (10:02 -0700)]
Reimplement folly::Function to improve compile times.

Summary:
folly::Function is causing significant compile time regressions. Reimplement it in a simpler way.

These are the times for a file containing 1000 instantiations of folly::Fuction (old), folly::Function (new), and std::function with **g++ 4.8 -O3** on my CentOS7 server.

|        | Old `folly::Function` | `std::function` | New `folly::Function` |
|--------|-----------------------|-----------------|-----------------------|
| Time   | 10m37s                | 0m16.81s        | 0m14.75s              |

And for the executable size:

|        | Old `folly::Function` | `std::function` | New `folly::Function` |
|--------|-----------------------|-----------------|-----------------------|
| Size   | 10,409,504            | 732,150         | 562,781               |

That's a **43X** improvement in compile times and an **18X** reduction in executable bloat over the old implementation.

The times for **clang (trunk)** are very different:

|       | Old `folly::Function` | `std::function` | New `folly::Function` |
|-------|-----------------------|-----------------|-----------------------|
| Time  | 4m6s                  | 0m45.27s        | 0m11.78s              |

That's a **20X** improvement over the old implementation and almost a **4X** improvement over `std::function`.

For **gcc-5.3.0**, compile times are again different:

|       | Old `folly::Function` | `std::function` | New `folly::Function` |
|-------|-----------------------|-----------------|-----------------------|
| Time  | 2m49s                 | 0m18.99s        | 0m20.70s              |

With gcc-5.3, the new implementation "only" compiles 8x faster than the old one, and is roughly the same as `std::function`.

Reviewed By: spacedentist, ot, luciang

Differential Revision: D3199985

fb-gh-sync-id: b97982a9dc3a63140510babea34988932e89f2d9
fbshipit-source-id: b97982a9dc3a63140510babea34988932e89f2d9

8 years agoAdd recursion limit to folly::parseJson.
Kyle Nekritz [Wed, 27 Apr 2016 16:42:56 +0000 (09:42 -0700)]
Add recursion limit to folly::parseJson.

Summary: Without this, malicious inputs can crash anything using folly::parseJson.

Reviewed By: yfeldblum

Differential Revision: D3219036

fb-gh-sync-id: 3604a060170c0201473c420035b21b018383789c
fbshipit-source-id: 3604a060170c0201473c420035b21b018383789c

8 years agoRemove dead code in AsyncSSLSocket
Subodh Iyengar [Wed, 27 Apr 2016 16:13:48 +0000 (09:13 -0700)]
Remove dead code in AsyncSSLSocket

Summary: Remove dead code in AsyncSSLSocket.

Reviewed By: knekritz

Differential Revision: D3226948

fb-gh-sync-id: e85823f311de2539c6aa2d6bcc3ff3aee07045bf
fbshipit-source-id: e85823f311de2539c6aa2d6bcc3ff3aee07045bf

8 years agofix access-after-free bug in ProducerConsumerQueueTest
Sven Over [Wed, 27 Apr 2016 13:46:08 +0000 (06:46 -0700)]
fix access-after-free bug in ProducerConsumerQueueTest

Summary: This bug causes failure when the test is run under ASAN.

Reviewed By: meyering

Differential Revision: D3229494

fb-gh-sync-id: a43c8332cc45f7892ac86cd0abb799616bca7779
fbshipit-source-id: a43c8332cc45f7892ac86cd0abb799616bca7779

8 years agoCheaper bumpHandlingTime
Pavlo Kushnir [Wed, 27 Apr 2016 07:20:04 +0000 (00:20 -0700)]
Cheaper bumpHandlingTime

Summary: no need in virtual call from EventHandler.

Reviewed By: yfeldblum

Differential Revision: D3226960

fb-gh-sync-id: eb9c191630e1a1ac022666201100e3778eb7b611
fbshipit-source-id: eb9c191630e1a1ac022666201100e3778eb7b611

8 years agoOptimize away IOBuf clones on hot path
Jon Maltiel Swenson [Wed, 27 Apr 2016 01:12:20 +0000 (18:12 -0700)]
Optimize away IOBuf clones on hot path

Summary:
'get' is the most frequent memcached operation. On a get reply, we should go to heap
as little as possible.  This diff optimizes for this scenario, where replies
have only one IOBuf field.

Reviewed By: pavlo-fb

Differential Revision: D3226592

fb-gh-sync-id: 92e1a1fac5735bd268691cf11990a96ae6fa8309
fbshipit-source-id: 92e1a1fac5735bd268691cf11990a96ae6fa8309

8 years agoAdding a constructor to UTF8Range that uses std::string
Aditya Muttur [Tue, 26 Apr 2016 22:53:40 +0000 (15:53 -0700)]
Adding a constructor to UTF8Range that uses std::string

Summary:
Currently UTF8Range has a constructor that allows you construct an object of type UTF8Range using only an object of type folly::Range. Adding a constructor so that we can construct an UTF8Range object using a std::string.

Currently,
  void sampleMethod(UTF8StringPiece sp) {...}
  /*
  ...
  */
  std::string str = "example";
  folly::UTF8Range utf8Range(str);
  folly::StringPiece sp(str);
  sampleMethod(utf8Range); // works
  sampleMethod(sp); // works
  sampleMethod(str); // doesn't work

This diff hopes to fix this issue.

Reviewed By: ddrcoder

Differential Revision: D3221144

fb-gh-sync-id: dd6ec4d7790d4602dccb3b63a4861d358aed3608
fbshipit-source-id: dd6ec4d7790d4602dccb3b63a4861d358aed3608

8 years agoAPI to set folly::RequestContext for current scope, try 2
Mirek Klimos [Tue, 26 Apr 2016 16:58:33 +0000 (09:58 -0700)]
API to set folly::RequestContext for current scope, try 2

Summary: same as D3156698, without changes in Cpp2Connection (which was the only real change in the diff)

Reviewed By: haijunz

Differential Revision: D3222792

fb-gh-sync-id: 245c7add837c0fc6d0bc84aa7d80b929ba2ce386
fbshipit-source-id: 245c7add837c0fc6d0bc84aa7d80b929ba2ce386

8 years agoDefine DECLARE_VARIABLE in portability/GFlags.h
Christopher Dykes [Tue, 26 Apr 2016 16:38:18 +0000 (09:38 -0700)]
Define DECLARE_VARIABLE in portability/GFlags.h

Summary: If we don't, then anywhere we include `glog/logging.h` after the portability header will get the `DECLARE_*` and `DEFINE_*` macros undefined -_-....

Reviewed By: mzlee

Differential Revision: D3220797

fb-gh-sync-id: 52907ddcd6b222fb1c6423034ea999eac5ce09ab
fbshipit-source-id: 52907ddcd6b222fb1c6423034ea999eac5ce09ab

8 years agoRemove extra dependency on Portability.h
Michael Lee [Tue, 26 Apr 2016 15:43:27 +0000 (08:43 -0700)]
Remove extra dependency on Portability.h

Summary: This dependency is in the wrong order.

Reviewed By: knekritz

Differential Revision: D3224312

fb-gh-sync-id: 98d2c58b75cf5f16892d462ae1b877d06bce3faa
fbshipit-source-id: 98d2c58b75cf5f16892d462ae1b877d06bce3faa

8 years agoUse std::string in folly::dynamic
Pavlo Kushnir [Tue, 26 Apr 2016 09:27:37 +0000 (02:27 -0700)]
Use std::string in folly::dynamic

Summary: Why not kill fbstring entirely? There are some legitimate use cases. For example, `folly::IOBuf` supports `moveToFbstring()` method which is impossible to implement with `std::string`.

Reviewed By: joeg, snarkmaster

Differential Revision: D3189410

fb-gh-sync-id: 9bb9090ca6012ac32ba9fb79041b26ec4888781f
fbshipit-source-id: 9bb9090ca6012ac32ba9fb79041b26ec4888781f

8 years agoRetire BOOST_STATIC_ASSERT in favor of static_assert
Yedidya Feldblum [Tue, 26 Apr 2016 07:56:27 +0000 (00:56 -0700)]
Retire BOOST_STATIC_ASSERT in favor of static_assert

Summary:
Retire `BOOST_STATIC_ASSERT` in favor of `static_assert`.

`static_assert` is part of C++ now, so we don't need workarounds like `BOOST_STATIC_ASSERT` anymore.

Partially automated:

    hg grep -lw BOOST_STATIC_ASSERT | xargs perl -pi -e 's~\bBOOST_STATIC_ASSERT\(([^;]*)\);~static_assert(\1, "");~g'
    hg grep -lw 'boost/static_assert.hpp' | xargs perl -pi -e 's,^#include <boost/static_assert\.hpp>\n,,gm'

Caught most instances. Formatting and remaining instances addressed manually.

Reviewed By: meyering

Differential Revision: D3215944

fb-gh-sync-id: f4552d5d9bfc416ce283923abe880437a4d0cba5
fbshipit-source-id: f4552d5d9bfc416ce283923abe880437a4d0cba5

8 years agoRemove unnecessary construction of IOBuf in Cursor
Stepan Palamarchuk [Mon, 25 Apr 2016 21:12:21 +0000 (14:12 -0700)]
Remove unnecessary construction of IOBuf in Cursor

Summary: This assignment is unnecessary, because in `for` loop we **always** call `cloneOneInto` that would assign another IOBuf to that IOBuf.

Reviewed By: alikhtarov

Differential Revision: D3215861

fb-gh-sync-id: ad87b99848aaae79f0870d49e0474f1abf0f28e5
fbshipit-source-id: ad87b99848aaae79f0870d49e0474f1abf0f28e5

8 years agoUndelayedDestruction virtual=>override
Andrew Cox [Mon, 25 Apr 2016 17:51:47 +0000 (10:51 -0700)]
UndelayedDestruction virtual=>override

Summary:
Had this in another diff, but it caused too much test noise. So I've separated
it out on it's own.

Reviewed By: yfeldblum

Differential Revision: D3213605

fb-gh-sync-id: 9ebfcf8430da7c66a31868032a0cef1e616ffc58
fbshipit-source-id: 9ebfcf8430da7c66a31868032a0cef1e616ffc58

8 years agouse forward instead of move for universal references
Sven Over [Sat, 23 Apr 2016 15:07:25 +0000 (08:07 -0700)]
use forward instead of move for universal references

Summary:When dealing with universal references, std::move will move from
objects that are passed as lvalue references. Instead std::forward
should be used, which only moves from an object if it's passed
as rvalue reference.

Reviewed By: yfeldblum

Differential Revision: D3200402

fb-gh-sync-id: 14be071e8498dd64cb8b2583c0cc2dd383bfebb8
fbshipit-source-id: 14be071e8498dd64cb8b2583c0cc2dd383bfebb8

8 years agoupdate futures README
Hans Fugal [Fri, 22 Apr 2016 21:11:29 +0000 (14:11 -0700)]
update futures README

Summary:It's been awhile.

Still has intern URLs (#10943549) but I'll fix that separately.

Reviewed By: jsedgwick

Differential Revision: D3213245

fb-gh-sync-id: ff60193ff368deaac8ca233d4289f30d8f6bb223
fbshipit-source-id: ff60193ff368deaac8ca233d4289f30d8f6bb223