folly.git
8 years agoto_string is not supported everywhere
Michael Lee [Tue, 1 Mar 2016 15:51:30 +0000 (07:51 -0800)]
to_string is not supported everywhere

Summary: Also remove an extra `#include <strstream>` from JsonTest.cpp as it is no longer necessary.

Reviewed By: yfeldblum

Differential Revision: D2991381

fb-gh-sync-id: 386b8281726d034933ebdc26733dd35dc5ace949
shipit-source-id: 386b8281726d034933ebdc26733dd35dc5ace949

8 years agoRemove unecessary int main from MemoryTest
Michael Lee [Tue, 1 Mar 2016 15:48:18 +0000 (07:48 -0800)]
Remove unecessary int main from MemoryTest

Summary: We do not need this main function, so remove it.

Reviewed By: markisaa

Differential Revision: D2992000

fb-gh-sync-id: 30bc0734af0f61e047276dab9639ae29cc999bbb
shipit-source-id: 30bc0734af0f61e047276dab9639ae29cc999bbb

8 years agoAdd conversion constructors for Future
Subodh Iyengar [Tue, 1 Mar 2016 03:38:32 +0000 (19:38 -0800)]
Add conversion constructors for Future

Summary:previously we were not able to do

Future<Base> f = makeFuture<Derived>();

This is because Future did not declare a conversion
constructor.

Adding a proper conversion ctor for Future is very
tricky because of the way Core is managed under the hood.
Core is not movable, and cannot be moved otherwise we
would have to retain pointers to the Future and Promises which
pointed to a particular core. This would be inefficient.

Instead we compromise and allow a very small subset of conversions
from objects whose templated types are convertible and also the
sizes of their Cores are also the same. As a result, we can
convert between types like unique_ptrs however not always between
full objects.

Reviewed By: jsedgwick

Differential Revision: D2943775

fb-gh-sync-id: 7c2388f2fb49d789c80ae2477814e960a099771b
shipit-source-id: 7c2388f2fb49d789c80ae2477814e960a099771b

8 years agoFix SingletonThreadLocal
Pavlo Kushnir [Mon, 29 Feb 2016 20:02:27 +0000 (12:02 -0800)]
Fix SingletonThreadLocal

Summary: it's really thread local at this point :)

Reviewed By: jmswen

Differential Revision: D2989668

fb-gh-sync-id: 423b9214922c92318181a62c583f58cd204b52e3
shipit-source-id: 423b9214922c92318181a62c583f58cd204b52e3

8 years agoUse yield, not wfe, for asm_volatile_pause on ARM and ARM64
Scott Wolchok [Sat, 27 Feb 2016 20:39:23 +0000 (12:39 -0800)]
Use yield, not wfe, for asm_volatile_pause on ARM and ARM64

Summary:Found this hunting down LockFreeRingBuffer iOS perf problems. `wfe` and `yield` are similar, but `yield` is the instruction meant for "hey, I'm doing a spinlock", whereas `wfe` tries to wait for certain processor events. See http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0204j/Cjafcggi.html

Conveniently, it's the same instruction on 32-bit and 64-bit ARM.

Reviewed By: yfeldblum

Differential Revision: D2986160

fb-gh-sync-id: 34671256112e605bf857f9db54a56cf6bb6f1ee2
shipit-source-id: 34671256112e605bf857f9db54a56cf6bb6f1ee2

8 years agoFix folly::ThreadLocal to have unique singleton in dev builds
Andrii Grynenko [Fri, 26 Feb 2016 18:28:31 +0000 (10:28 -0800)]
Fix folly::ThreadLocal to have unique singleton in dev builds

Summary:This re-uses StaticSingletonManager which was previously used to fix the same issue in folly::Singleton.

Because of the same issue we can no longer use static thread_local for the ThreadEntry. We now rely on pthread_getspecific/pthread_setspecific instead and use static thread_local ThreadEntry* only as a cache (to improve perf).

Reviewed By: yfeldblum

Differential Revision: D2978526

fb-gh-sync-id: cf1d9044afc27b62bd50a1ed931c0c420ae7107e
shipit-source-id: cf1d9044afc27b62bd50a1ed931c0c420ae7107e

8 years agoCreate the sys/time.h portability header
Christopher Dykes [Fri, 26 Feb 2016 17:28:51 +0000 (09:28 -0800)]
Create the sys/time.h portability header

Summary: Windows doesn't have it.

Reviewed By: yfeldblum

Differential Revision: D2979215

fb-gh-sync-id: 030bdcaa44deb18834903705a26369ce0b13abc6
shipit-source-id: 030bdcaa44deb18834903705a26369ce0b13abc6

8 years agoSplit tests into test and benchmarks.
Michael Lee [Thu, 25 Feb 2016 22:58:16 +0000 (14:58 -0800)]
Split tests into test and benchmarks.

Summary:Split up

  io/test/IOBufCursorTest
  test/BitsTest
  test/CacheLocalityTest
  test/CallOnceTest
  test/ConvTest
  test/LoggingTest
  test/MemoryIdlerTest
  test/RandomTest
  test/ThreadLocalTest

Reviewed By: yfeldblum

Differential Revision: D2973132

fb-gh-sync-id: 9dadbdf49a31e82c3a2e34c2fdb6a2b47aa0928d
shipit-source-id: 9dadbdf49a31e82c3a2e34c2fdb6a2b47aa0928d

8 years agoSplit FileTest to a smaller test and an extended test
Michael Lee [Thu, 25 Feb 2016 21:51:28 +0000 (13:51 -0800)]
Split FileTest to a smaller test and an extended test

Summary: FileTest pulls in subprocess and does a lot more than just directly test File.h

Reviewed By: yfeldblum

Differential Revision: D2971641

fb-gh-sync-id: 16e1096ab3b0f6434a26f84c889ebb082ee3f210
shipit-source-id: 16e1096ab3b0f6434a26f84c889ebb082ee3f210

8 years agoCreate a portability header for working with environment variables
Christopher Dykes [Thu, 25 Feb 2016 17:41:35 +0000 (09:41 -0800)]
Create a portability header for working with environment variables

Summary:Because `extern char** environ` is not the correct way to access environ on Windows.
This also implements setenv and unsetenv for Windows, which means that TestUtils no longer needs to be disabled for the Windows build.

Reviewed By: mzlee

Differential Revision: D2973704

fb-gh-sync-id: 84db7db3494cf183fcbcc25063cb0482ef84ebf4
shipit-source-id: 84db7db3494cf183fcbcc25063cb0482ef84ebf4

8 years agoshorten SharedMutex tests that timeout under ASAN
Nathan Bronson [Thu, 25 Feb 2016 04:28:37 +0000 (20:28 -0800)]
shorten SharedMutex tests that timeout under ASAN

Summary:New ASAN options are more thorough but slower, resulting
in timeouts for some tests.  Decreasing their internal loop count to
compensate.

Reviewed By: yfeldblum

Differential Revision: D2973982

fb-gh-sync-id: 67391936d0b4930f99d00358ebb93a1b3ee87140
shipit-source-id: 67391936d0b4930f99d00358ebb93a1b3ee87140

8 years agoFixing AtomicHashArrayTest ASan Test Failure
Jonathan Kron [Thu, 25 Feb 2016 02:36:30 +0000 (18:36 -0800)]
Fixing AtomicHashArrayTest ASan Test Failure

Summary: Both of these tests will unavoidably fail on ASan because of shared library loads, and indestructable objects.  Disabling ASAN testing for these test files.

Reviewed By: yfeldblum

Differential Revision: D2968169

fb-gh-sync-id: d2acc71b40541af5e10f15d5087a7c07da05453d
shipit-source-id: d2acc71b40541af5e10f15d5087a7c07da05453d

8 years agoMinor MSVC issues
Christopher Dykes [Thu, 25 Feb 2016 00:29:40 +0000 (16:29 -0800)]
Minor MSVC issues

Summary:MSVC doesn't let you have unicode characters in a narrow string without using a utf-8 literal string.
AsyncTransport needs the definition of IOBuf because of how the standard library implements a few things.
And longs are only 32-bit under Windows.

Reviewed By: yfeldblum

Differential Revision: D2973973

fb-gh-sync-id: c2f3e848e740a65c575598991d43dd79360ec1e3
shipit-source-id: c2f3e848e740a65c575598991d43dd79360ec1e3

8 years agoExplicit vtordisp for MSVC
Orvid King [Thu, 25 Feb 2016 00:28:14 +0000 (16:28 -0800)]
Explicit vtordisp for MSVC

Summary:Because sometimes things are just different.

MSVC warns about the `dynamic_cast` done against this class because the class's declaration doesn't show that it would need a vtordisp. While I don't believe the `dynamic_cast` in question is actually called in either the constructor or destructor paths for the class, better to be safe and force the disp to be created.

Reviewed By: yfeldblum

Differential Revision: D2887490

Pulled By: Orvid

fb-gh-sync-id: 821a6fc820087ac7120fffd9f8b95a26c2b4e71b
shipit-source-id: 821a6fc820087ac7120fffd9f8b95a26c2b4e71b

8 years agofolly/futures: keep Core alive until callback has returned
Sven Over [Wed, 24 Feb 2016 07:46:14 +0000 (23:46 -0800)]
folly/futures: keep Core alive until callback has returned

Summary:Callbacks passed to e.g. folly::Future::then (or
folly::Future::ensure) may delete the promise that keeps the
Core of the same future alive. The Core must protect itself from
destruction during callback execution.

This commit also adds a unit test to check for correct behaviour
in the "self destruction" scenario. The test should usually pass
even without the fix in Core.h. However, when you run the test
in Valgrind or ASAN, it will report problems unless the fix in
Core.h is applied.

Reviewed By: mhx

Differential Revision: D2938094

fb-gh-sync-id: 22796e168e1876ef2e3c7d7619da020be6a22073
shipit-source-id: 22796e168e1876ef2e3c7d7619da020be6a22073

8 years agoFix double definition for templated folly::Singletons
Andrii Grynenko [Wed, 24 Feb 2016 03:49:07 +0000 (19:49 -0800)]
Fix double definition for templated folly::Singletons

Summary: Allow createGlobal() to create dependent objects.

Reviewed By: yfeldblum

Differential Revision: D2963111

fb-gh-sync-id: 8e4da48a7a1000934963396b423e8eff98a8aade
shipit-source-id: 8e4da48a7a1000934963396b423e8eff98a8aade

8 years agoFix Build: folly/test/IndestructibleTest.cpp under LSAN
Yedidya Feldblum [Wed, 24 Feb 2016 00:35:17 +0000 (16:35 -0800)]
Fix Build: folly/test/IndestructibleTest.cpp under LSAN

Summary:[Folly] Fix Build: `folly/test/IndestructibleTest.cpp` under LSAN.

Leaks of Meyers singletons are ignored automatically, while normal locals are not. So we make our instances be Meyers singletons.

Differential Revision: D2968368

fb-gh-sync-id: 385ac4491d9a105885af82d85354af929d69cc80
shipit-source-id: 385ac4491d9a105885af82d85354af929d69cc80

8 years agoFix SIOF in folly/io/async/Request.cpp
Andrii Grynenko [Tue, 23 Feb 2016 22:56:29 +0000 (14:56 -0800)]
Fix SIOF in folly/io/async/Request.cpp

Summary: Use Meyers singleton instead.

Reviewed By: afrind

Differential Revision: D2963282

fb-gh-sync-id: 7aec36e7cc3b0ab09b2b093202f40d5b0d399f4d
shipit-source-id: 7aec36e7cc3b0ab09b2b093202f40d5b0d399f4d

8 years agoAsyncSocketException: add detailed information in error message
Beny Luo [Tue, 23 Feb 2016 17:44:14 +0000 (09:44 -0800)]
AsyncSocketException: add detailed information in error message

Summary: AsyncSocketException: add detailed information in error message

Reviewed By: yfeldblum

Differential Revision: D2953030

fb-gh-sync-id: a5e26036657ecc6c74ad0369650730a2af13bead
shipit-source-id: a5e26036657ecc6c74ad0369650730a2af13bead

8 years agoSplit benchmarks and tests
Michael Lee [Tue, 23 Feb 2016 16:00:49 +0000 (08:00 -0800)]
Split benchmarks and tests

Summary:Benchmarks need their own main function, which means they
aren't really useful in cases we can't use a custom main function.

Reviewed By: yfeldblum

Differential Revision: D2962104

fb-gh-sync-id: 25bdc6e5a8bdf8c3aa94d393207a74797b2e1234
shipit-source-id: 25bdc6e5a8bdf8c3aa94d393207a74797b2e1234

8 years agofolly: symbolizer: dwarf: don't fallback to linear scan if address missing from ...
Lucian Grijincu [Tue, 23 Feb 2016 09:50:18 +0000 (01:50 -0800)]
folly: symbolizer: dwarf: don't fallback to linear scan if address missing from .debug_aranges

Summary:Presence of .debug_aranges implies user expects fast address lookup.
Some addresses might not be avaialble in .debug_aranges.

Don't do slow lookup in .debug_info, as it can lead to unexpected slowdowns.

override-unit-failures

Reviewed By: philippv

Differential Revision: D2965323

fb-gh-sync-id: 405daefd57cdff4344fd231c5f5b7ff4dcd9df8c
shipit-source-id: 405daefd57cdff4344fd231c5f5b7ff4dcd9df8c

8 years agoFix OSX build
Neel Goyal [Mon, 22 Feb 2016 22:42:49 +0000 (14:42 -0800)]
Fix OSX build

Summary: Fixed the build using the homebrew instructions

Reviewed By: yfeldblum

Differential Revision: D2960409

fb-gh-sync-id: 218b879f26b3baab09676cac51787bf36d0b4bed
shipit-source-id: 218b879f26b3baab09676cac51787bf36d0b4bed

8 years agofix SIOF in CacheLocality.h's AccessSpreader
Nathan Bronson [Mon, 22 Feb 2016 21:08:30 +0000 (13:08 -0800)]
fix SIOF in CacheLocality.h's AccessSpreader

Summary:This diff moves all data accessed during
AccessSpreader<>::current(x) into the .data segment, avoiding SIOF
without adding indirection or dynamic gating as would be the case for
normal singleton-like constructs.  The diff also trims the AccessSpreader
API to include only those methods that people actually seem to use.

Reviewed By: djwatson

Differential Revision: D2945205

fb-gh-sync-id: 847e31adc4450217f4ed0575686be261fb504d7c
shipit-source-id: 847e31adc4450217f4ed0575686be261fb504d7c

8 years agoRemove extra `int main`s from unit tests.
Michael Lee [Mon, 22 Feb 2016 16:42:18 +0000 (08:42 -0800)]
Remove extra `int main`s from unit tests.

Summary: test/common/TestMain.cpp should provide the necessary initialization.

Reviewed By: yfeldblum

Differential Revision: D2947245

fb-gh-sync-id: 75b9d241673667badf4dc8fb81285c60cbb8166a
shipit-source-id: 75b9d241673667badf4dc8fb81285c60cbb8166a

8 years agoMove RequestContext::getStaticContext to .cpp
Andrii Grynenko [Sat, 20 Feb 2016 09:08:28 +0000 (01:08 -0800)]
Move RequestContext::getStaticContext to .cpp

Summary: All singleton registration has to happen in .cpp. Otherwise singleton may be double registered if inlined.

Reviewed By: yfeldblum

Differential Revision: D2956951

fb-gh-sync-id: bea425c31270d614f4b8a780204694168602fe86
shipit-source-id: bea425c31270d614f4b8a780204694168602fe86

8 years agoSupport nested FiberManagers
Andrii Grynenko [Sat, 20 Feb 2016 00:13:42 +0000 (16:13 -0800)]
Support nested FiberManagers

Summary:It's possible to have nested loopUntilNoReady (with different FiberManagers). See unit test for a simple example.
This diff fixes currentFiberManager_ to be a stack.

Reviewed By: meyering

Differential Revision: D2953468

fb-gh-sync-id: 0abdcb7f43c94e7bb0adef8440699dc8e138d21a
shipit-source-id: 0abdcb7f43c94e7bb0adef8440699dc8e138d21a

8 years agoFix EventBaseOnDestructionCallback to loop FiberManager
Andrii Grynenko [Fri, 19 Feb 2016 19:19:18 +0000 (11:19 -0800)]
Fix EventBaseOnDestructionCallback to loop FiberManager

Summary: EventBase loop should do it as well, but it's possible for LoopCallback to be removed before EventBaseOnDestructionCallback.

Reviewed By: pavlo-fb

Differential Revision: D2951303

fb-gh-sync-id: 43f5a91e7ecffa7ab7feec32fa45e01b28db66a3
shipit-source-id: 43f5a91e7ecffa7ab7feec32fa45e01b28db66a3

8 years agoFix folly::Singleton to work in dynamically linked binaries
Andrii Grynenko [Fri, 19 Feb 2016 19:02:35 +0000 (11:02 -0800)]
Fix folly::Singleton to work in dynamically linked binaries

Summary:This implements StaticSingletonManager which is then used to create all leaked Meyers singletons.

StaticSingletonManager is a singleton itself, which is created in a separate compilation unit (Singleton.cpp) and so we can be sure that other compilation units will always see a single instance of StaticSingletonManager, even if linked dynamically.

StaticSingletonManager then keeps a dictionary of typeid -> object pointer, which is used to de-duplicate same singleton being re-created from different compilation units (linked dynamically), usually because of code inlining.

override-unit-failures

Reviewed By: yfeldblum

Differential Revision: D2913027

fb-gh-sync-id: 1f5015a79a7a8297ebf5f0fe3fd0cc7eb44f706b
shipit-source-id: 1f5015a79a7a8297ebf5f0fe3fd0cc7eb44f706b

8 years agofix dangling pointer bug in IPAddressV6
Tianjiao Yin [Fri, 19 Feb 2016 18:40:46 +0000 (10:40 -0800)]
fix dangling pointer bug in IPAddressV6

Summary: `mask(numBits)` returns a temporary variable. After its lifetime ends, subbytes becomes a dangling pointer.

Reviewed By: yfeldblum

Differential Revision: D2953699

fb-gh-sync-id: 5d5a35716ecaa1b3d96edb5a459615756848b92f
shipit-source-id: 5d5a35716ecaa1b3d96edb5a459615756848b92f

8 years agoIndestructible
Yedidya Feldblum [Fri, 19 Feb 2016 07:10:20 +0000 (23:10 -0800)]
Indestructible

Summary:[Folly] `Indestructible`.

For when you need a Meyers singleton that will never be destructed, even at program exit.

Good for large data structures with many heap allocations, as long as they have no behavior. No point in ever destructing them.

HT: Proxygen, and Orvid.

Reviewed By: andriigrynenko

Differential Revision: D2947959

fb-gh-sync-id: 7dd1f725edf137ba81fbf4032a0819fd7c627261
shipit-source-id: 7dd1f725edf137ba81fbf4032a0819fd7c627261

8 years agofolly: symbolizer: slow address->{file+line-number} lookup if `.debug_aranges` is...
Lucian Grijincu [Fri, 19 Feb 2016 02:46:44 +0000 (18:46 -0800)]
folly: symbolizer: slow address->{file+line-number} lookup if `.debug_aranges` is missing (e.g. --strip-debug-non-line)

Summary:Binaries linked with `gold` and `--strip-debug-non-line` don't have an `.debug_aranges` section

We still want to map `address->{file+line-number}` to get nice stack
traces even though this might be slower (linear search all compilation unit entries in `.debug_info`).

Before:
```
$ # link with gold + --strip-debug-non-line
$ folly/experimental/exception_tracer/exception_tracer_test
E0217 15:02:13.694947 1321814 ExceptionTracer.cpp:179] Exception type: std::runtime_error (9 frames)
    @ 000000000040ad2d __cxa_throw
    @ 0000000000409df3 bar()
    @ 0000000000409eab baz()
    @ 0000000000407c77 main
    @ 00007f00dd9860f5 __libc_start_main
    @ 000000000040991b (unknown)
```

After (similar to the output without `--strip-debug-non-line`):
```
E0217 18:37:37.579596 1583124 ExceptionTracer.cpp:179] Exception type: std::runtime_error (9 frames)
    @ 000000000040ad6d __cxa_throw
                       ./folly/experimental/exception_tracer/ExceptionTracerLib.cpp:57
    @ 0000000000409e33 bar()
                       ./folly/experimental/exception_tracer/ExceptionTracerTest.cpp:24
    @ 0000000000409eeb baz()
                       ./folly/experimental/exception_tracer/ExceptionTracerTest.cpp:51
    @ 0000000000407c87 main
                       ./folly/experimental/exception_tracer/ExceptionTracerTest.cpp:96
    @ 00007f1d16ff80f5 __libc_start_main
    @ 000000000040995b (unknown)
```

Differential Revision: D2947965

fb-gh-sync-id: e517bab324b1dcb70cadc9a5211ce794e35c83a5
shipit-source-id: e517bab324b1dcb70cadc9a5211ce794e35c83a5

8 years agoDon't use a VLA for the double->string buffer.
Christopher Dykes [Thu, 18 Feb 2016 23:24:48 +0000 (15:24 -0800)]
Don't use a VLA for the double->string buffer.

Summary: MSVC is actually smarter about this than GCC, as MSVC doesn't support VLAs, it tries to eval the length at compile time. GCC on the other hand doesn't try to eval it at compile time, resulting in compiles via CMake telling us that this is a VLA.

Reviewed By: yfeldblum

Differential Revision: D2911929

fb-gh-sync-id: ffaa133bcf4129a3e02f7e875966d3ae6a97be6a
shipit-source-id: ffaa133bcf4129a3e02f7e875966d3ae6a97be6a

8 years agofolly/ApplyTuple.h: fix const-correctness & other issues, simplify
Sven Over [Thu, 18 Feb 2016 13:31:22 +0000 (05:31 -0800)]
folly/ApplyTuple.h: fix const-correctness & other issues, simplify

Summary:The existing implementation of folly::applyTuple does not support
mutable callables (such as mutable lambdas and other functor objects
that only implement non-const operator()).

This commit adds a few more unit tests and changes the implementation
so that new and existing tests pass.

Reviewed By: yfeldblum

Differential Revision: D2942622

fb-gh-sync-id: 82478f290e9fd2020358ff79ef0a6bcf8a43738c
shipit-source-id: 82478f290e9fd2020358ff79ef0a6bcf8a43738c

8 years agoSort the Makefile headers a bit
Yedidya Feldblum [Wed, 17 Feb 2016 23:30:46 +0000 (15:30 -0800)]
Sort the Makefile headers a bit

Summary:[Folly] Sort the `Makefile` headers a bit.

Just manually moving a few obviously out-of-order entries.

Reviewed By: Orvid

Differential Revision: D2945668

fb-gh-sync-id: 604976d25e5913ef01d591e23737d61dacfbfc86
shipit-source-id: 604976d25e5913ef01d591e23737d61dacfbfc86

8 years agoclang-format some code in preparation for real changes
Nathan Bronson [Wed, 17 Feb 2016 22:20:19 +0000 (14:20 -0800)]
clang-format some code in preparation for real changes

Reviewed By: djwatson

Differential Revision: D2945770

fb-gh-sync-id: 9e4ee3b052b0bbb33ef796b8070edd24c6942589
shipit-source-id: 9e4ee3b052b0bbb33ef796b8070edd24c6942589

8 years agoAdd call_once, wrapper around std::call_once with a fast path
Lovro Puzar [Wed, 17 Feb 2016 13:22:40 +0000 (05:22 -0800)]
Add call_once, wrapper around std::call_once with a fast path

Summary: std::call_once is a nice API but the current GCC implementation is slower than it needs to be.  Adding a header-only wrapper with an additional atomic bool.

Reviewed By: luciang

Differential Revision: D2938884

fb-gh-sync-id: 5939c94fe62a1523053dcff26c880ecaec9e1150
shipit-source-id: 5939c94fe62a1523053dcff26c880ecaec9e1150

8 years agoHandle wrapvFull when IOV_MAX is not defined.
Michael Lee [Wed, 17 Feb 2016 04:38:20 +0000 (20:38 -0800)]
Handle wrapvFull when IOV_MAX is not defined.

Summary:--Perhaps this should move to Portability.h instead.  There is another instance of this in io/async/AsyncSocket.cpp.--

Added `portability/SysUio.h` to handle the c-macro conversion into a usable value.

Reviewed By: yfeldblum

Differential Revision: D2940778

fb-gh-sync-id: 897e44b430c02e5a7d826f3e8da9e4979b7b898c
shipit-source-id: 897e44b430c02e5a7d826f3e8da9e4979b7b898c

8 years agoECDSA async/offloading support for proxygen
Anirudh Ramachandran [Wed, 17 Feb 2016 01:28:22 +0000 (17:28 -0800)]
ECDSA async/offloading support for proxygen

Summary: Minor changes to AsyncSSLSocket for async crypto

Reviewed By: siyengar

Differential Revision: D2516765

fb-gh-sync-id: 354baeb94e6f63e8d5cdf8455ff5ca49a6aa479c
shipit-source-id: 354baeb94e6f63e8d5cdf8455ff5ca49a6aa479c

8 years agoFix wrapvFull for the case when iovec* has size more than 1024
Iaroslav Tverdokhlib [Tue, 16 Feb 2016 18:21:35 +0000 (10:21 -0800)]
Fix wrapvFull for the case when iovec* has size more than 1024

Summary: `folly::wrapvFull` fail if the passed in `iovec*` has more than 1024 (`IOV_MAX`) elements. In particular, it returns -1 with errno 22 [Invalid argument]. The fix is to limit maximum size of iovec* to IOV_MAX that is passed in to `readv/writev/...` in a single iteration of outer loop.

Reviewed By: yfeldblum

Differential Revision: D2935540

fb-gh-sync-id: 6c0a073ac0b59db3d53fb4269b13ddfcc479efb1
shipit-source-id: 6c0a073ac0b59db3d53fb4269b13ddfcc479efb1

8 years agofolly::Optional<T> should be trivially destructible when T is
Brett Simmers [Tue, 16 Feb 2016 05:51:35 +0000 (21:51 -0800)]
folly::Optional<T> should be trivially destructible when T is

Summary:There doesn't appear to be a way to use std::enable_if on a
destructor, so conditionally select from one of two storage types.

Reviewed By: yfeldblum

Differential Revision: D2923902

fb-gh-sync-id: 2def8d1031d70379fd84e8eb555dad9d2b4996f2
shipit-source-id: 2def8d1031d70379fd84e8eb555dad9d2b4996f2

8 years agoModification to futures to remove deadlock in certain use cases for getVia(executor).
Lee Howes [Mon, 15 Feb 2016 21:22:30 +0000 (13:22 -0800)]
Modification to futures to remove deadlock in certain use cases for getVia(executor).

Summary: If getVia was called on a future modified using via, getVia could deadlock if the original future was updated to a new executor and there was no callback chained after the call to via. In effect: f.via(executor).getVia(executor); deadlocks. This can be a problem if the code is hidden in a library and the precise semantics are unclear. This diff adds a test that exposes the problem and a fix by forcing waitVia to add a callback that will satisfy the new exector, ensuring that drive() has a callback to trigger once the future is satisfied.

Reviewed By: andriigrynenko

Differential Revision: D2906858

fb-gh-sync-id: a3105079530f15d7a7d39a9381c4078665b721a7
shipit-source-id: a3105079530f15d7a7d39a9381c4078665b721a7

8 years agoUnify runInMainContext for void and non-void
Andrii Grynenko [Mon, 15 Feb 2016 19:32:24 +0000 (11:32 -0800)]
Unify runInMainContext for void and non-void

Summary: This also fixes a bug where exception was not re-thrown for functions returning void.

Reviewed By: spalamarchuk

Differential Revision: D2936887

fb-gh-sync-id: 9828dec131203528c27eae874aba147168f40d0d
shipit-source-id: 9828dec131203528c27eae874aba147168f40d0d

8 years agoMake Gold more like Core.
Michael Lee [Mon, 15 Feb 2016 14:00:09 +0000 (06:00 -0800)]
Make Gold more like Core.

Summary:Adding padding for Android was wrong, instead, we should have
used `std::aligned_storage`.

Reviewed By: fugalh

Differential Revision: D2934566

fb-gh-sync-id: f541b89309be70791fced48f63b6b5aecc49bfbb
shipit-source-id: f541b89309be70791fced48f63b6b5aecc49bfbb

8 years agofolly/test/json_test: fix heap-buffer-overflow in Json.PrintTo test
Igor Sugak [Fri, 12 Feb 2016 23:54:28 +0000 (15:54 -0800)]
folly/test/json_test: fix heap-buffer-overflow in Json.PrintTo test

Summary:Json.PrintTo test if failing with heap-buffer-overflow asan abort. The problem here comes up
when values of `std::string` and `std::ostrstring::str()` are compared.  `std::ostrstring::str()` returns
*non null* terminated array of `char`s. When compared with `std::string` a read from memory after
the end that array is made. Fixing the test by replacing `std::ostrstream` with `std::ostringstream`,
that does append `\0`.

Reviewed By: yfeldblum

Differential Revision: D2934352

fb-gh-sync-id: c6a5d765c9951716b8a14715702cf3d940c6d723
shipit-source-id: c6a5d765c9951716b8a14715702cf3d940c6d723

8 years agoConvert Thrift1(2)RequestDispatcher::sendMessage()
Andrii Grynenko [Fri, 12 Feb 2016 22:34:00 +0000 (14:34 -0800)]
Convert Thrift1(2)RequestDispatcher::sendMessage()

Summary:This depends on D2897095.

The main goal here is to start moving onRequestSuccess and onRequestFailure calls higher in the stack.

Reviewed By: mmcduff

Differential Revision: D2899959

fb-gh-sync-id: 4e074c2d734f88f5be56000095b892c6d47c0dcc
shipit-source-id: 4e074c2d734f88f5be56000095b892c6d47c0dcc

8 years agofolly copyright 2015 -> copyright 2016
Mark Isaacson [Fri, 12 Feb 2016 20:56:41 +0000 (12:56 -0800)]
folly copyright 2015 -> copyright 2016

Summary: Update copyright for 2016

Reviewed By: igorsugak

Differential Revision: D2828047

fb-gh-sync-id: 8638188aeb694c3ca2a615cc2bc3bc9d251e388d
shipit-source-id: 8638188aeb694c3ca2a615cc2bc3bc9d251e388d

8 years agoFix Singleton/ThreadLocal destruction order crashes in buck dev builds
Andrii Grynenko [Fri, 12 Feb 2016 18:22:25 +0000 (10:22 -0800)]
Fix Singleton/ThreadLocal destruction order crashes in buck dev builds

Summary: This adds a folly::Singleton->folly::ThreadLocal dependency to make sure folly::ThreadLocal is always loaded first. Otherwise PthreadKeyUnregister singleton is created after folly::Singleton storage, even though it has higher priority.

Reviewed By: andrewjcg

Differential Revision: D2931170

fb-gh-sync-id: 36392d76e98201f2b4416b4bbef451d6c0e8c69d
shipit-source-id: 36392d76e98201f2b4416b4bbef451d6c0e8c69d

8 years agoFix MultiLevelTimeseries::getRate()
Drew Hoskins [Thu, 11 Feb 2016 23:53:21 +0000 (15:53 -0800)]
Fix MultiLevelTimeseries::getRate()

Summary:I hit a landmine where rate() returned 0 for my tests where fewer than 60 items were added per minute.  This was because it was truncating it.  And yet, countRate() was working.
It doesn't make sense for a rate (value accumulated per time period) to be integral.
Folly: I changed rate() to return a double by default for folly, as was done by avg() and countRate().  Looked like a typo to me.
Common wrapper: I changed getRate() to allow you to override and make it double, as was done by getAvg().  Defaulting to int (which is usually the value type) is a bad call IMO but it's a riskier change to change it to double, and I want to be consistent with getAvg().

Reviewed By: tracelog

Differential Revision: D2921061

fb-gh-sync-id: 00875f2ab7963ef3ba2db475aedaf6ebd413b38f
shipit-source-id: 00875f2ab7963ef3ba2db475aedaf6ebd413b38f

8 years agofix -Wshadowing in String.cpp
Michael Lee [Thu, 11 Feb 2016 04:22:20 +0000 (20:22 -0800)]
fix -Wshadowing in String.cpp

Summary: Didn't check this with -Wshadow when I first wrote it.

Reviewed By: ranjeeth

Differential Revision: D2926047

fb-gh-sync-id: 8391986dee2d6b5698491bd2995f039468ec684d
shipit-source-id: 8391986dee2d6b5698491bd2995f039468ec684d

8 years agoFix typo in StringTest
Michael Lee [Thu, 11 Feb 2016 04:04:28 +0000 (20:04 -0800)]
Fix typo in StringTest

Summary: It should be UL instead of ULL.

Reviewed By: siyengar

Differential Revision: D2926005

fb-gh-sync-id: d2387eb601c50d9820907c831277aa6f8e938cbb
shipit-source-id: d2387eb601c50d9820907c831277aa6f8e938cbb

8 years agoSwitch stripLeftMargin to not use boost::regex
Michael Lee [Thu, 11 Feb 2016 02:51:43 +0000 (18:51 -0800)]
Switch stripLeftMargin to not use boost::regex

Summary: Remove boost regex from `stripLeftMargin`. We can shrink some binaries by not including it in the core folly library.

Reviewed By: yfeldblum

Differential Revision: D2922415

fb-gh-sync-id: cee89164c650706f0e5c07eed3d40500831918cd
shipit-source-id: cee89164c650706f0e5c07eed3d40500831918cd

8 years agoAdd MAX_STATIC_CONSTRUCTOR_PRIORITY to PthreadKeyUnregister
Andrii Grynenko [Thu, 11 Feb 2016 02:02:27 +0000 (18:02 -0800)]
Add MAX_STATIC_CONSTRUCTOR_PRIORITY to PthreadKeyUnregister

Summary: We want onThreadExit to be canceled as late as possible. Otherwise accessAllThreads() may crash on shutdown (because ThreadEntry is destroyed, but not removed from the global list).

Reviewed By: igorsugak

Differential Revision: D2923435

fb-gh-sync-id: 3a8eef05844bb547c06db33cb96734df1169852a
shipit-source-id: 3a8eef05844bb547c06db33cb96734df1169852a

8 years agoCompensate for -m32 platforms.
Michael Lee [Thu, 11 Feb 2016 00:05:12 +0000 (16:05 -0800)]
Compensate for -m32 platforms.

Summary:
When compiled for -m32, this test fails because of the
difference between longs.

Reviewed By: yfeldblum

Differential Revision: D2924626

fb-gh-sync-id: 8ee863fa2b6df8519bd52d0d3144697a297780bb
shipit-source-id: 8ee863fa2b6df8519bd52d0d3144697a297780bb

8 years agoMinor optimizations in ExceptionCounterLib
Giuseppe Ottaviano [Wed, 10 Feb 2016 20:17:35 +0000 (12:17 -0800)]
Minor optimizations in ExceptionCounterLib

Summary: Avoid allocation and logging inside the throw handler, and some other minor tweaks.

Reviewed By: bort, luciang

Differential Revision: D2921073

fb-gh-sync-id: 4491ab2f85a4251e59ba9394bba5abef0bdf7a10
shipit-source-id: 4491ab2f85a4251e59ba9394bba5abef0bdf7a10

8 years agoFix EventBase destruction race in FiberManagerMap
Andrii Grynenko [Wed, 10 Feb 2016 19:14:26 +0000 (11:14 -0800)]
Fix EventBase destruction race in FiberManagerMap

Summary:
Previously we could be reading from thread-local FiberManagerMap while it was modified.
This is now fixed by keeping a per-thread list of EventBases which need to be removed from local maps. On the fast-path no action is taken, since list will be empty.

This is second try, since D2853921 got reverted.
The new implementation is simpler and does not rely on AtomicLinkedList.

Reviewed By: yfeldblum

Differential Revision: D2908018

fb-gh-sync-id: 4d7aed974c19761f7e2732ddbf8694af57c69bd6
shipit-source-id: 4d7aed974c19761f7e2732ddbf8694af57c69bd6

8 years agostripLeftMargin
Yedidya Feldblum [Wed, 10 Feb 2016 00:28:21 +0000 (16:28 -0800)]
stripLeftMargin

Summary:
[Folly] `stripLeftMargin`.

So you can do:

    TEST(MyClass, doSomethingWithString) {
      //  Multiline string literal is indented properly inside the test case
      //  instead of being aligned all the way to the left, which would make
      //  the test case read poorly.
      auto input = folly::stripLeftMargin(R"TEXT(
        first line
        second line
        third line
      )TEXT");
      auto expected = //...
      auto actual = MyClass::doSomethingWithString(input);
      EXPECT_EQ(expected, actual);
    }

Reviewed By: markisaa

Differential Revision: D2909736

fb-gh-sync-id: ebb07da05e1a788535064cfcd9e07f617a007800
shipit-source-id: ebb07da05e1a788535064cfcd9e07f617a007800

8 years agoRelax HHWheelTimer::destroy assertion to accommodate SharedPtr
Alan Frindell [Tue, 9 Feb 2016 17:44:21 +0000 (09:44 -0800)]
Relax HHWheelTimer::destroy assertion to accommodate SharedPtr

Summary:
HHWheelTimer's auto-pointers are kind of funny.  You can do something like this:

```
HHWheelTimer::UniquePtr p = ...;
// create a SharedPtr from UniquePtr
HHWheelTimer::SharedPtr s(p);
// create another SharedPtr from raw ptr
HHWheelTimer::SharedPtr s(p.get());
// No problem.

If you do this:

HHWheelTimer::SharedPtr p = ....;
// this leaks
```

Why?  Because SharedPtr is only have of std::shared_ptr.  It's the refcounting half.  But when the last SharedPtr goes out of scope, it **does not** invoke HHWheelTimer::destroy().

So code like this is possible/expected:

```
MySweetObj::MySweetObj(HHWheelTimer::SharedPtr s) {
  s_ = s;
  s_.scheduleTimeout(a, b);
}

{
  HHWheelTimer::UniquePtr p = ...;

  obj = MySweetObj(p)

  // But what if I decide to kill p:
  p.reset();
}
```

Since MySweetObj takes a SharedPtr and holds it, it can reasonbly expect that it can schedule timeouts on it, and the HHWheelTimer will not be deleted until it releases the SharedPtr.  This is true, but the above code would hit the assert that count_ == 0.

Instead, relase the check that count_ == 0 only if there are no extra outstanding SharedPtrs.

Reviewed By: viswanathgs, chadparry

Differential Revision: D2908729

fb-gh-sync-id: 9abd4a7d692fe952c5514dbb8d85dfbad95a3cac
shipit-source-id: 9abd4a7d692fe952c5514dbb8d85dfbad95a3cac

8 years agoFix bad merge.
Michael Lee [Mon, 8 Feb 2016 20:53:55 +0000 (12:53 -0800)]
Fix bad merge.

Summary:
FBStringTest.cpp was refactored, but the TestUtil and
TemporaryFile was not properly removed.

Reviewed By: yfeldblum

Differential Revision: D2913045

fb-gh-sync-id: 710c3f5b808acb634dfcd65219484ddc257ed52c

8 years agoFix Build: conditionally_existent_test in folly/test/Makefile.am
Yedidya Feldblum [Mon, 8 Feb 2016 02:14:17 +0000 (18:14 -0800)]
Fix Build: conditionally_existent_test in folly/test/Makefile.am

Summary:
[Folly] Fix Build: `conditionally_existent_test` in `folly/test/Makefile.am`.

HT: https://github.com/shindo.

Reviewed By: meyering

Differential Revision: D2910782

fb-gh-sync-id: 9593f026e7c58b8644abc5996da65005bcc2d095

8 years agofolly/futures: fix early release of non-embedded callbacks
Sven Over [Sun, 7 Feb 2016 13:24:01 +0000 (05:24 -0800)]
folly/futures: fix early release of non-embedded callbacks

Summary:
folly::Future (more precisely folly::detail::Core) can store
callback functors (typically lambdas) up to a certain size
(8*sizeof(void*)) inside the main object. Only bigger functors
are stored in the std::function object in folly::detail::Core,
which will put them on the heap.

The behaviour of folly::Future is slightly different depending
on whether the callback can be embedded in the main object
or not. Small functors will be destructed after the callback
is executed. Functors too big to fit in the lambda space in
the Core object will be deleted when Core is deleted.

Some code assumes that functor objects are deleted as soon
as the callback has been executed. The implementations of
folly::Future::collect and variants do so. If you switch
off this optimisation temporarily (which can be done easily
by setting lambdaBufSize in folly/futures/detail/Core.h to
0), a number of unit tests fail.

Given that the lambda buffer is reasonably large, most
functors can be stored in the Core object. The different
behaviour for very large lambdas may not have been observed.

This diff fixes the inconsitent behaviour.

Firstly, it changes the unit test Future:finish to explicitly
check that the callback functor is deleted after the callback
has been executed. This should be tested, because this
behaviour is assumed in other parts of the futures
implementation (e.g. Future::collect et al.).

Secondly, it adds another unit test Future:finishBigLambda,
similar to Future:finish. The lambda captures much more data
to make sure the lambda won't be stored in the Core object,
but in the std::function object inside Core. The test verifies
that the behaviour is the same, i.e. the callback functor
is destructed after the callback has been executed.

Thirdly, it fixes Core and makes sure that functors of any
size are destructued after the callback has been called. The
new unit test fails without this.

Reviewed By: fugalh

Differential Revision: D2883772

fb-gh-sync-id: 21a410f6592b3e090772a7b55bef6729d8739922

8 years agoCall destructor for non-trivial destructors if arena allocator is used in ConcurrentS...
Den Raskovalov [Sat, 6 Feb 2016 05:31:32 +0000 (21:31 -0800)]
Call destructor for non-trivial destructors if arena allocator is used in ConcurrentSkipList.

Reviewed By: philippv

Differential Revision: D2900534

fb-gh-sync-id: c711a160d541d937ada24f2b524016dbceb40ee2

8 years agoLoosen restriction to get folly::future::CoreTest pass on android
Francis Ma [Fri, 5 Feb 2016 23:00:39 +0000 (15:00 -0800)]
Loosen restriction to get folly::future::CoreTest pass on android

Summary: Due to different way of padding, CoreTest failed on android. Loosen the check to make it pass.

Reviewed By: mzlee

Differential Revision: D2907987

fb-gh-sync-id: 81d715725d6908ec7b259b0a67789a91ed63df71

8 years agoMake folly/future portable on android
Francis Ma [Fri, 5 Feb 2016 22:42:28 +0000 (14:42 -0800)]
Make folly/future portable on android

Summary: Some small changes to port folly/futures onto android

Reviewed By: mzlee

Differential Revision: D2787564

fb-gh-sync-id: 8c161942c24c2b7b0484339eaa51c5a356f17de5

8 years agoConditionallyExistent<typename>
Yedidya Feldblum [Fri, 5 Feb 2016 20:27:56 +0000 (12:27 -0800)]
ConditionallyExistent<typename>

Summary:
[Folly] `ConditionallyExistent<typename>`.

For when we need extra member fields in dbg builds, but want to save the space in opt builds. Or other situations along similar lines, but with perhaps another the statically-known condition.

Conditional compilation can have some nasty side-effects. Best to avoid it. Let the compiler see everything, rather than having the preprocessor ruin our day. Let the optimizer remove code that will never be called, after the compiler has seen it. Let us have a single AST in our source file, not one AST for dbg and one for opt, or other statically-known conditions.

Reviewed By: andriigrynenko

Differential Revision: D2879397

fb-gh-sync-id: d631141a984eebd46674f27a40a97f670eb33f54

8 years agoRevert D2853921
Andrii Grynenko [Fri, 5 Feb 2016 20:01:26 +0000 (12:01 -0800)]
Revert D2853921

Summary: This was causing memory leaks. I assume that happens if FiberManager was once requested from a thread (so that thread local was initialized), but the thread was never used since then, accumulating EventBase* in the queue.

Reviewed By: simonmar

Differential Revision: D2906752

fb-gh-sync-id: 71ab14cb051a9cee3684a30eaf6729ef65888a52

8 years agoTurn off a FBStringTest for Android.
Michael Lee [Fri, 5 Feb 2016 15:27:02 +0000 (07:27 -0800)]
Turn off a FBStringTest for Android.

Summary:
The wstring support on KitKat is not exactly working.  Works
on Lollipop though.

Reviewed By: markisaa

Differential Revision: D2901732

fb-gh-sync-id: 97b57fb4d8c645192be62dca25d8fb1b2397ad7d

8 years agoGreatly expand the components overview in Overview.md
Louis Brandy [Fri, 5 Feb 2016 01:23:56 +0000 (17:23 -0800)]
Greatly expand the components overview in Overview.md

Summary: See title. This is create a big flat mess, which I may clean up in a follow on to highlight the biggest and most generally useful libraries.

Reviewed By: fugalh

Differential Revision: D2902152

fb-gh-sync-id: 028633448ab97f47a3419f31fa58c79b77a89e52

8 years agopromote a bunch of high level txt to the main Readme.md
Louis Brandy [Fri, 5 Feb 2016 00:55:29 +0000 (16:55 -0800)]
promote a bunch of high level txt to the main Readme.md

Summary: See title

Reviewed By: snarkmaster

Differential Revision: D2901817

fb-gh-sync-id: 5b9ede7b941cf18e7be352a93ec5686e6f1f50a7

8 years agoAdding addTaskFuture and addTaskRemoteFuture to FiberManager.
Lee Howes [Fri, 5 Feb 2016 00:19:09 +0000 (16:19 -0800)]
Adding addTaskFuture and addTaskRemoteFuture to FiberManager.

Summary: Adds functionality to fibermanager to add tasks locally and remotely that will return futures. As a side effect, also wraps the function in addTaskRemote in a move wrapper so that it behaves correctly with move-only types such as promises.

Reviewed By: andriigrynenko, yfeldblum

Differential Revision: D2892898

fb-gh-sync-id: 1666c103db35d9c149c36f8b8c3058cd6e0465fc

8 years agoImprove RequestContext::getStaticContext() perf
Andrii Grynenko [Thu, 4 Feb 2016 21:58:25 +0000 (13:58 -0800)]
Improve RequestContext::getStaticContext() perf

Summary: Avoid using folly::ThreadLocal on the fast-path. This uses the fact that it is a static object.

Reviewed By: yfeldblum

Differential Revision: D2873374

fb-gh-sync-id: 978108da0c9e8576fda6849058b0262478c2841e

8 years agoAllow SSLSessionCallbacks to be used on SSL* that isn't attached to AsyncSSLSocket
Neel Goyal [Thu, 4 Feb 2016 18:43:06 +0000 (10:43 -0800)]
Allow SSLSessionCallbacks to be used on SSL* that isn't attached to AsyncSSLSocket

Summary: The session callbacks assumed that the SSL* was associated with a folly::AsyncSSLSocket when it didn't need to.  This enables apps that manage their own SSL* to use these callbacks.

Reviewed By: yfeldblum

Differential Revision: D2896426

fb-gh-sync-id: c51df6b4cb3f4cc188a6411c1f3e7e89e96e8a67

8 years agoBuild workaround for nvcc in Malloc.h
Giuseppe Ottaviano [Thu, 4 Feb 2016 01:32:42 +0000 (17:32 -0800)]
Build workaround for nvcc in Malloc.h

Summary: See comment.

Reviewed By: juancarabina, luciang

Differential Revision: D2896262

fb-gh-sync-id: cb884651d47b86c0a5b3c2c22f421b2b1f39dc35

8 years agoMake HHWheelTimer take a TimeoutManager rather than EventBase
Alan Frindell [Tue, 2 Feb 2016 23:31:32 +0000 (15:31 -0800)]
Make HHWheelTimer take a TimeoutManager rather than EventBase

Summary: TimeoutManager is a Mockable parent class

Reviewed By: maxgeorg

Differential Revision: D2892671

fb-gh-sync-id: d20635b18a7c748009ce4880c80a13ac23b9222a

8 years agoAdd MockTimeoutManager
Peter Griess [Tue, 2 Feb 2016 23:27:21 +0000 (15:27 -0800)]
Add MockTimeoutManager

Summary: - Add new MockTimeoutManager mock. For testing.

Reviewed By: yfeldblum

Differential Revision: D2891814

fb-gh-sync-id: 6d15feb533e5a8984484d1efdf58d2c77b7b3f49

8 years agofolly: add bser encode/decode for dynamic
Wez Furlong [Tue, 2 Feb 2016 22:54:56 +0000 (14:54 -0800)]
folly: add bser encode/decode for dynamic

Summary:
To support consuming Watchman from within fbcode and hhvm
in particular, these functions add a BSER serialization for the
folly::dynamic data type.

Reviewed By: bhamiltoncx

Differential Revision: D2876539

fb-gh-sync-id: bc49d6bc453cc66cebda7185a5907a6f70970b24

8 years agofixed Ubuntu 12 build script dependencies
Denis Samoylov [Tue, 2 Feb 2016 22:38:25 +0000 (14:38 -0800)]
fixed Ubuntu 12 build script dependencies

Summary: added missed boost dependency and changed CMake build order

Reviewed By: snarkmaster, yfeldblum

Differential Revision: D2878880

fb-gh-sync-id: 2a5cded97afde4fd0d741168c3296f2c71426374

8 years agoAdd some quick comments about folly::fiber implementation classes.
Martin Martin [Tue, 2 Feb 2016 22:12:48 +0000 (14:12 -0800)]
Add some quick comments about folly::fiber implementation classes.

Summary: Add some quick comments about folly::fiber implementation classes.

Reviewed By: yfeldblum

Differential Revision: D2891940

fb-gh-sync-id: b9899bb63df03f32763f93a5dae8dff94c8e354e

8 years agoKill writeTerminator in fbstring
Giuseppe Ottaviano [Tue, 2 Feb 2016 18:27:10 +0000 (10:27 -0800)]
Kill writeTerminator in fbstring

Summary:
`writeTerminator` needs to determine the string category, and in the small case, also its size. This information is often known, but the compiler cannot optimize it away especially if `capacity_` was just overwritten. Also, some occurrences are just redundant. We can remove the latter and specialize the former.

Small operations such as `push_back` are up to 40% faster.

Before/after:
```
$ ./fbstring_benchmark_using_jemalloc --bm_regex '_fbstring' --bm_min_usec 200000 --bm_max_secs 2
============================================================================ ===================
./folly/test/FBStringTestBenchmarks.cpp.h       relative  time/iter  iters/s  time/iter  iters/s
============================================================================ ===================
BM_initRNG_fbstring(0)                                       0.00fs  Infinity    0.00fs  Infinity
BM_defaultCtor_fbstring(0)                                   6.82us  146.60K     6.26us  159.74K
BM_copyCtor_fbstring(32768)                                 16.86ns   59.32M    15.50ns   64.52M
BM_ctorFromArray_fbstring(32768)                             2.25us  444.78K     2.12us  471.58K
BM_ctorFromTwoPointers_fbstring(0)                           9.76ns  102.44M     7.85ns  127.35M
BM_ctorFromTwoPointers_fbstring(7)                           9.68ns  103.32M     8.10ns  123.47M
BM_ctorFromTwoPointers_fbstring(15)                          9.77ns  102.34M     8.12ns  123.17M
BM_ctorFromTwoPointers_fbstring(23)                          9.46ns  105.68M     8.78ns  113.88M
BM_ctorFromTwoPointers_fbstring(24)                         31.23ns   32.02M    30.71ns   32.56M
BM_ctorFromChar_fbstring(1048576)                           40.04ns   24.97M    35.68ns   28.03M
BM_assignmentOp_fbstring(256)                               66.76ns   14.98M    63.93ns   15.64M
BM_assignmentFill_fbstring(0)                                8.23ns  121.47M     7.02ns  142.49M
BM_resize_fbstring(524288)                                   4.09us  244.63K     3.94us  253.84K
BM_equality_fbstring(65536)                                  2.47ms   404.63     2.19ms   455.92
BM_replace_fbstring(256)                                   172.36ns    5.80M   159.10ns    6.29M
BM_push_back_fbstring(1)                                     9.32ns  107.27M     7.79ns  128.40M
BM_push_back_fbstring(23)                                  252.44ns    3.96M   158.31ns    6.32M
BM_push_back_fbstring(127)                                 721.50ns    1.39M   515.08ns    1.94M
BM_push_back_fbstring(1024)                                  5.21us  191.87K     3.14us  318.03K
BM_short_append_fbstring(23)                               431.71ns    2.32M   335.74ns    2.98M
BM_short_append_fbstring(1024)                              11.96us   83.64K     9.19us  108.78K
BM_getline_fbstring(23)                                     57.06ns   17.53M    39.78ns   25.14M
BM_getline_fbstring(1000)                                  232.26ns    4.31M   203.24ns    4.92M
============================================================================ ===================

```

(`find` benchmarks were removed due to high variance caused by randomness)

Reviewed By: luciang

Differential Revision: D2879646

fb-gh-sync-id: 10fd8573495d285220ae98858d11de9ec6d97e54

8 years agoAllow building without JEMalloc under MSVC
Orvid King [Tue, 2 Feb 2016 17:49:08 +0000 (09:49 -0800)]
Allow building without JEMalloc under MSVC

Summary: This was falling back to a hard depency on JEMalloc, and MSVC doesn't have proper weak linkage, so emulate it with some linker magic to get things building.

Reviewed By: yfeldblum

Differential Revision: D2887695

Pulled By: Orvid

fb-gh-sync-id: 50c812a07c4e4e5bbe71263b86386783bf76ba82

8 years agofolly: #define UNDEFINED_SANITIZER in ubsan mode
Lucian Grijincu [Tue, 2 Feb 2016 01:00:24 +0000 (17:00 -0800)]
folly: #define UNDEFINED_SANITIZER in ubsan mode

Summary:
Undefined Sanitizer doesn't define any macro to detect that it's active.

The build system should provide `-DUNDEFINED_SANITIZER` when building
with -fsanitize=undefined or any of the other flags from
http://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html

FWIW: Chrome defines the same preprocessor symbol:
https://chromium.googlesource.com/chromium/src/+/ac18f489dca8c902e4dfaa1a28d716b7914121d0%5E%21/build/common.gypi

Reviewed By: andrewjcg

Differential Revision: D2885167

fb-gh-sync-id: e1129c0863bfde5d032c32e7d5cea7c43d82009f

8 years agoImplementing callback functionality for exception routines
Dmitry Pleshkov [Mon, 1 Feb 2016 23:35:18 +0000 (15:35 -0800)]
Implementing callback functionality for exception routines

Summary: Depends on D2865911

Reviewed By: luciang

Differential Revision: D2742158

fb-gh-sync-id: 3e7866a742575ee4f7501cff0abbd5c21e26a46e

8 years agofolly: ubsan: Benchmark: avoid division by zero (fsanitize=float-divide-by-zero)
Lucian Grijincu [Mon, 1 Feb 2016 23:19:55 +0000 (15:19 -0800)]
folly: ubsan: Benchmark: avoid division by zero (fsanitize=float-divide-by-zero)

Reviewed By: philippv

Differential Revision: D2886132

fb-gh-sync-id: a845d657fb920847df52fe5ec077e91554544f5b

8 years agofolly: fix one more -Wunused-parameter
Igor Sugak [Mon, 1 Feb 2016 22:35:10 +0000 (14:35 -0800)]
folly: fix one more -Wunused-parameter

Summary: The usage is guarded by a preprocessor macro, and I didn't noticed this earlier.

Reviewed By: yfeldblum

Differential Revision: D2887016

fb-gh-sync-id: 791c4d16475aab77235792953997a281354018e9

8 years agoRemove unnecessary includes from Optional.h
Scott Wolchok [Mon, 1 Feb 2016 21:17:55 +0000 (13:17 -0800)]
Remove unnecessary includes from Optional.h

Summary:
I want to be able to include this in
fbobjc/xplat/folly:headers_only, but it currently pulls in a boost
header and Portability.h. Fortunately, neither of those includes are
needed.

Reviewed By: ddrcoder, yfeldblum, mhorowitz

Differential Revision: D2880443

fb-gh-sync-id: 409561b8fb555ac3946d59cd6657c6c643b67c86

8 years agoSupport constexpr_strlen under MSVC.
Christopher Dykes [Mon, 1 Feb 2016 20:36:43 +0000 (12:36 -0800)]
Support constexpr_strlen under MSVC.

Summary: MSVC doesn't support evaluating strlen at compile time, so implement our own version instead.

Reviewed By: yfeldblum, lbrandy

Differential Revision: D2856926

fb-gh-sync-id: 22222350b57d9eff6a06c9d0f37d43a3cb1f2534

8 years agofolly: Hash: use loadUnassigned to avoid UB in hsieh_hash32_buf (fsanitize=alignment)
Lucian Grijincu [Mon, 1 Feb 2016 19:51:51 +0000 (11:51 -0800)]
folly: Hash: use loadUnassigned to avoid UB in hsieh_hash32_buf (fsanitize=alignment)

Reviewed By: philippv

Differential Revision: D2886152

fb-gh-sync-id: 64c3543db831c72c7f4cf307867223270293066e

8 years agofolly: ubsan: HashTest: avoid invalid shift (sanitize=shift)
Lucian Grijincu [Mon, 1 Feb 2016 19:45:53 +0000 (11:45 -0800)]
folly: ubsan: HashTest: avoid invalid shift (sanitize=shift)

Summary:
  [ RUN      ] Hash.TWang_Unmix64
  folly/test/HashTest.cpp:125:20: runtime error: shift exponent 32 is too large for 32-bit type 'unsigned int'

Reviewed By: philippv

Differential Revision: D2886144

fb-gh-sync-id: 8d7963c087c9db34b08c07451d35e5568c750520

8 years agofolly: fix clang -Wmissing-field-initializers
Igor Sugak [Mon, 1 Feb 2016 18:44:42 +0000 (10:44 -0800)]
folly: fix clang -Wmissing-field-initializers

Summary: Fix a few `-Wmissing-field-initializers` exposed by clang.

Reviewed By: yfeldblum

Differential Revision: D2881902

fb-gh-sync-id: 67fdffd39d3ccca64f84055adae1f3b47fdec633

8 years agoComments for SSLVerifyPeerEnum
Neel Goyal [Mon, 1 Feb 2016 13:52:01 +0000 (05:52 -0800)]
Comments for SSLVerifyPeerEnum

Summary:
Document what the enum settings mean since they can be
somewhat confusing.

Reviewed By: shamdor-fb

Differential Revision: D2882929

fb-gh-sync-id: 74ec30132bf5d2dce42f51a0b7b30cf2fae12dbf

8 years agofolly: fix -Wunused-parameter in opt mode
Igor Sugak [Fri, 29 Jan 2016 23:53:26 +0000 (15:53 -0800)]
folly: fix -Wunused-parameter in opt mode

Summary: Fix a few remaining unused parameters, that are exposed only in opt build.

Reviewed By: markisaa

Differential Revision: D2878865

fb-gh-sync-id: d0d9761362220973cda14d99ab7342fbe8b1a469

8 years agofolly: fix clang build with -Wunused-const-variable
Igor Sugak [Fri, 29 Jan 2016 23:33:50 +0000 (15:33 -0800)]
folly: fix clang build with -Wunused-const-variable

Summary: Fix a few unused const variables exposed by clang's `-Wunused-const-variable`.

Reviewed By: yfeldblum

Differential Revision: D2878944

fb-gh-sync-id: f5500fda4782eac2964761c0398d016d57716269

8 years agoAdjust AsyncServerSocket to not use getsockname before it's connected
Orvid King [Fri, 29 Jan 2016 22:37:19 +0000 (14:37 -0800)]
Adjust AsyncServerSocket to not use getsockname before it's connected

Summary:
Winsock doesn't like it when you try to call getsockname on a socket that hasn't yet been connected or bound, because that socket doesn't have a name yet.
This only occurred because we were trying to get the family of the socket.
To solve this, I just passed the family in from the parent methods that already knew what the family was.

Reviewed By: yfeldblum

Differential Revision: D2871859

Pulled By: Orvid

fb-gh-sync-id: 7674f565a968aa0258355fc977c185a416e4fbe4

8 years agoHandle tmp dir on Android
Michael Lee [Fri, 29 Jan 2016 18:32:31 +0000 (10:32 -0800)]
Handle tmp dir on Android

Summary:
There is no universally accessible tmpdir on Android which
means mkstemp fails. This diff calls a function that the test runner
should provide to fill a valid temporary directory.

Reviewed By: yangchi

Differential Revision: D2842034

fb-gh-sync-id: 9b826757bd750af016a18adccd5a21174be644d6

8 years agoMake advanced init function for test optional
Denis Samoylov [Fri, 29 Jan 2016 18:18:01 +0000 (10:18 -0800)]
Make advanced init function for test optional

Summary: In order to avoid mandatory dependncy on elf, dwarf, unwind made new init function optional

Reviewed By: markisaa

Differential Revision: D2878545

fb-gh-sync-id: f66d3884a531dcf56fc1432330325ab45b149d7f

8 years agoExclude based on __APPLE__
Michael Lee [Fri, 29 Jan 2016 15:23:19 +0000 (07:23 -0800)]
Exclude based on __APPLE__

Summary: This was showing up on a Linux compile to my sadness.

Reviewed By: francis-ma

Differential Revision: D2876010

fb-gh-sync-id: f638d593712c0aeb50177d96aaacb26575820359

8 years agoOptimize getline(istream&, fbstring&) implementation
Giuseppe Ottaviano [Fri, 29 Jan 2016 15:02:58 +0000 (07:02 -0800)]
Optimize getline(istream&, fbstring&) implementation

Summary:
Current `getline` implementation in `fbstring` always allocates, even if the passed string is already large enough. Furthermore, the growing strategy relies on outdated assumptions about the allocator.
This implementation reuses the existing allocation as much as possible, then uses exponential growth.

Reviewed By: luciang, philippv

Differential Revision: D2871976

fb-gh-sync-id: 8db9512030be3f4953efa8f008747827504c032c

8 years agoA common init function for binaries and a default main function for tests
Denis Samoylov [Thu, 28 Jan 2016 19:18:54 +0000 (11:18 -0800)]
A common init function for binaries and a default main function for tests

Summary: Added initialization routines to test main function that can help with debugging tests

Reviewed By: markisaa, yfeldblum

Differential Revision: D2839759

fb-gh-sync-id: 71cad45f3747336c8c7f8706db139cd060e1442b

8 years agofolly: ubsan: replace undefined call through reinterpret-cast fn-pointer with std...
Lucian Grijincu [Thu, 28 Jan 2016 19:07:44 +0000 (11:07 -0800)]
folly: ubsan: replace undefined call through reinterpret-cast fn-pointer with std::function

Summary:
This code casts function pointers to void(*fn)(void*) and calls
functions through that type. Calling functions through a different
pointer type is undefined behavior. Casts were used to avoid memory
allocations.

`std::bind` needs a memory allocation - so it's avoided in EventBase.

  std::function<void()> x = std::bind(fn, args);

`std::function` should use small object optimizations when possible and embed the functor in the body of `std::function`.

On GCC 5.0 and forward small lambdas don't need memory allocations
(lambdas being tiny structures - two pointers in this case - and a
function operator).

  std::function<void()> y = [=] { fn(args); };

On GCC 4.7 .. 4.9 functors for which __is_location_invariant<Func> is
true also don't need memory allocations.

Remove undefined behavior by using a `SmallFunctor` which leverages `std::function`'s small object optimization.

Reviewed By: philippv

Differential Revision: D2864895

fb-gh-sync-id: ab40f60b3519ce38f43fecebf88ccdbf09d9bea9

8 years agoAdd (void)ssl back to SSLContext for -Wunused-parameter
Michael Lee [Thu, 28 Jan 2016 18:52:42 +0000 (10:52 -0800)]
Add (void)ssl back to SSLContext for -Wunused-parameter

Summary:
I'm not sure what accepted practice is to make the
-Wunused-parameter not balk.

Reviewed By: igorsugak

Differential Revision: D2874705

fb-gh-sync-id: 8765bae504fd90dfd8896857f4bf865ca8f64a8a

8 years agoFix unused parameter errors under -Werror
Marcus Holland-Moritz [Thu, 28 Jan 2016 16:05:32 +0000 (08:05 -0800)]
Fix unused parameter errors under -Werror

Summary:
D2872406 enables -Werror=unused-parameter, which in conjunction with mode/opt
(which eliminates asserts) causes two files in folly/io/async to fail to compile.
This change works around the error.

Reviewed By: vchalyshev

Differential Revision: D2874625

fb-gh-sync-id: 97104679f964390c5df88ee7831af7df243a152a