Summary:
[Folly] Fix a folly build failure under clang: MPMCQueueTest.cpp.

In clang-v3.4, there is a bug with the combination of a lambda expression inside a function template taking a template name (rather than a type name) as a template argument.

This diff, in the interest of building folly under clang-v3.4, extracts the lambda expression into a separate function so that the function template taking a template name as a template argument no longer has a lambda expression in it.

Test Plan: Build folly/test/MPMCQueueTest.cpp under clang.

Reviewed By: njormrod@fb.com

Subscribers: mathieubaudet, dougw

FB internal diff: D1446279

Tasks: 4723132

Summary:
[Folly] Fix a folly build failure under clang: FutexTest.cpp.

In clang-v3.4, there is a bug with the combination of a lambda expression inside a function template taking a template name (rather than a type name) as a template argument.

This diff, in the interest of building folly under clang-v3.4, extracts the lambda expression into a separate function so that the function template taking a template name as a template argument no longer has a lambda expression in it.

Test Plan: Build folly/test/FutexTest.cpp under clang.

Reviewed By: njormrod@fb.com

Subscribers: mathieubaudet, dougw

FB internal diff: D1446237

Tasks: 4723132

Summary:
This is D1422343, but with a one-line change to make it clang-compatible. I authored this diff by first copying the original, then updating, so it's easy to see what I had to change.

> on advice of @tudorb, move most of `folly::Formatter` into `folly::BaseFormatter` so that we can use compile-time polymorphism to provide different types of Formatters.

Test Plan:
```
fbmake clean
fbconfig -r thrift folly cold_storage && fbmake dbg && fbmake runtests
fbmake clean
fbconfig -r hphp --clang && fbmake dbgo && fbmake runtests
```

Reviewed By: tudorb@fb.com, pt@fb.com

Subscribers: mathieubaudet, tudorb

FB internal diff: D1440310

Tasks: 4624268, 4667712

Summary:
* The test computed nonaligned inputs but then copied them into temporary
buffers to compare different implementations. The temporary buffers
were word-aligned.
* This diff keeps the temp buffers but ensures that the data in them
keeps the original input's alignment.

Test Plan:
* Ran the test case with a modified String.cpp containing an assert to
catch unaligned reads.  The assert failed, as expected, on a copy
of the code without the fix from D1434585

Reviewed By: meyering@fb.com

Subscribers: ruibalp

FB internal diff: D1435028

Tasks: 4696800

Summary: There are a couple of places where this behavior is useful, and it's not 100% trivial to implement it from scratch. Adding it to Folly to save people code & bugs.

Test Plan: unit tests

Reviewed By: tudorb@fb.com

Subscribers: tjackson, folly@lists, tudorb

FB internal diff: D1297506

Summary: This way I can reuse it in Subprocess. It also makes it easy to make a bunch of other convenient tokenization routines (e.g. delimiter-preserving folly::gen tokenizers, file tokenizers, etc, etc).

Test Plan: fbconfig folly/gen/test && fbmake runtests

Reviewed By: tjackson@fb.com

Subscribers: vkatich, tjackson

FB internal diff: D1317973

Summary:
see task:
https://our.intern.facebook.com/intern/tasks/?t=4723861

Test Plan: unit tests

Reviewed By: marcelo.juchem@fb.com

FB internal diff: D1443223

Tasks: 4723861

Summary:
7.1.5 [dcl.constexpr] / 9 (N3337) requires type to be literal to use it
in `constexpr` object declaration (seems to be not enforced by GCC 4.8?)

Currently `folly::Range<>` fails one of the requirements:

* is an aggregate type, or has at least one constexpr constructor or
constructor template that is not a copy or move constructor,

Test Plan: fbconfig folly/test:range_test && fbmake runtests_opt -j32

Reviewed By: lucian@fb.com

Subscribers: chaoyc, search-fbcode-diffs@, unicorn-diffs@

FB internal diff: D1441646

Tasks: 4720575

Summary:
Check for self-assignment in move assignment. Otherwise
Optional<Neko> cat = whatever;
cat = std::move(cat);
cat.hasValue(); // always returns false

Test Plan: fbmake runtests

Reviewed By: tjackson@fb.com

FB internal diff: D1440633

Summary:
I modified the toAppend(args..., StringType* dest) so that
before appending it reserves enough space for data to be appended. It is
still work in progress (floats and doubles are really bad: we only do
very naive approximation of the size). On float like workload we gain
~10% perf, on strings/ints/chars we gain as much as 25% of perf.
Probably on bigger strings it will be even faster. We only modify the
case when toAppend() has more than 1 element to append as it would be
just overhead in case of one argument.

Test Plan:
with this change:

============================================================================
folly/test/ConvTest.cpp                         relative  time/iter
iters/s
============================================================================
preallocateTestNoFloat                                       1.59us 627.85K
preallocateTestFloat                                         1.09us 920.70K
----------------------------------------------------------------------------
============================================================================

without the change:

============================================================================
folly/test/ConvTest.cpp                         relative  time/iter
iters/s
============================================================================
preallocateTestNoFloat                                       2.12us 471.43K
preallocateTestFloat                                         1.22us 818.25K
----------------------------------------------------------------------------
============================================================================

Reviewed By: marcelo.juchem@fb.com

FB internal diff: D1420588

Tasks: 4632421

Summary:
To make MutableStringPiece more usable (and make the
implementation of replace* methods easier) I made sure that
all the const methods (like find) work for MutableStringPiece
in a fashion similar to how they work with StringPiece
(work with everything auto convertible to StringPiece). As
a result this change is pretty susbstatial in LOC, even
though now much happens here.

Test Plan: unit tests

Reviewed By: marcelo.juchem@fb.com

FB internal diff: D1420965

Tasks: 4632424

Summary:
* folly/String.cpp (toLowerAscii): Fix two errors: the most important
would cause unaligned accesses.  This would cause a performance loss
in general, but would also result in segfaults on ARM processes.
In addition, three conditionals were wrong, further limiting
the performance of this code: switch those "<" to "<=".

Test Plan:
Run this to exercise existing tests:
fbconfig folly/test:string_test && fbmake runtests_opt
Run this to generate timing stats (before and after this change), e.g.,
fbconfig folly/test:string_benchmark && fbmake opt
_bin/folly/test/string_benchmark > TIMING-toLower-old

These numbers show a 1.6% speed increase with this change:

--- TIMING-toLower-old  2014-07-14 16:51:12.793523778 -0700
+++ TIMING-toLower-new  2014-07-14 16:49:45.815119145 -0700
@@ -1,6 +1,6 @@
============================================================================
folly/test/StringBenchmark.cpp                  relative  time/iter  iters/s
============================================================================
-libc_tolower                                                 1.06us  941.91K
-folly_toLowerAscii                                          89.99ns   11.11M
+libc_tolower                                                 1.06us  941.90K
+folly_toLowerAscii                                          88.57ns   11.29M
============================================================================

Reviewed By: brianp@fb.com

Subscribers:

FB internal diff: D1434585

Tasks: 4696800

Blame Revision: D1421056

Reviewed By: lesha@fb.com
Test Plan: no

Test Plan: no

Reviewed By: lesha@fb.com

Subscribers: alandau, jhj, kma

FB internal diff: D1434163

Summary:
Because the version of lz4 that ships by default with Ubuntu 13.10 doesn't
define it (but has lz4_decompress_safe, so it's usable)

Test Plan: built on Ubuntu 13.10

Reviewed By: meyering@fb.com

Subscribers: jhj, lesha, kma

FB internal diff: D1433864

@override-unit-failures

Summary: IPAddress operator== throws an exception when comparing ip v6 mapped address with unitialized (default constructed) address. With this diff it returns false.

Test Plan: folly unit tests

Reviewed By: bmatheny@fb.com

FB internal diff: D1421280

Summary: Consider a case we found that the queue is empty and unlocked and before our setActive(false) from SCOPE_EXIT gets called (or gets the lock) putMessageImpl got the lock. Now putMessage thinks that we donot want to add another signal but actually we do.

Test Plan:
Running mcreplay2 without running into this problem on a box.
Benchmark:

Reviewed By: davejwatson@fb.com

FB internal diff: D1428249

Test Plan: folly/io/test, whatever contbuild dreams up for hphp

Reviewed By: meyering@fb.com

Subscribers: meyering, hphp-diffs@, ps, jhj, kma, sainbar, lesha

FB internal diff: D1429372

@override-unit-failures

Summary:
Folly is almost "out of the box" working with cygwin

This has the proper ifdefs to include cygwin in two areas and a workaround for a cygwin bug when including headers

Test Plan: fbconfig -r folly && fbmake runtests

Reviewed By: delong.j@fb.com

FB internal diff: D1413303

Summary:
MSVC has all the sockets and internet functionality required except for one missing item - which can safely be defined to a windows specific type
But it requires a different set of headers than on *nix systems

Test Plan: fbconfig -r folly && fbmake runtests

Reviewed By: delong.j@fb.com

FB internal diff: D1413265

Summary: Use msvc intrinsics for cpuid, popcount, byteswap, and bit scan functionality

Test Plan: fbconfig -r folly && fbmake runtests

Reviewed By: delong.j@fb.com

FB internal diff: D1413254

Summary:
MSVC 14 is still broken with expression sfinae - and the things that break are often strange
Moving this out into two expr templates solves the compilation issue

Test Plan: fbconfig -r folly && fbmake runtests

Reviewed By: tjackson@fb.com

FB internal diff: D1413297

Summary:
maelstrom destructs a Promise during an indirect call from
maybeCallback and deadlocks on itself unless we use a recursive mutex.

There may be a smarter way to proceed but at the moment I can't build
and deploy a package from trunk because the service is non-functional
:-/

Test Plan:
run

```
fbconfig -r folly/wangle messaging/maelstrom
fbmake runtests
```

Reviewed By: hannesr@fb.com

Subscribers: fugalh, fsilva

FB internal diff: D1428504

Summary:
No need to specialize std::hash for std::basic_string; the STL already does
this. Plus, the specializations were broken in multiple ways:

1. std::hash does not treat const char* specially, so it just hashes the
pointer value (contrasting the old __gnu_cxx::hash, which hashed a C-string)

2. Either way, using __gnu_cxx::hash<const char*> for std::string is broken,
as it won't hash anything past the first null byte.

Also, make sure fbstring gets the same (full) specializations as std::string
does in the standard.

Test Plan: fbconfig -r folly && fbmake runtests_opt, also tests in tupperware/agent which is still using __gnu_cxx::hash_map with string keys.

Reviewed By: pgriess@fb.com, andrei.alexandrescu@fb.com

Subscribers: njormrod, jhj, kma, lesha

FB internal diff: D1426479

Summary: When we are changing value of numActiveConsumers_ with setActive from handlerReady at the SCOPE_EXIT we have a race with reading of the same variable in putMessageImpl.

Test Plan: Had a local race which works fine now.

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent

FB internal diff: D1424674

Summary: per @rockyliu's suggestions

Test Plan: subprocess_test

Reviewed By: rockyliu4@fb.com

Subscribers: rockyliu, jhj, lesha, kma

FB internal diff: D1423157

Summary:
This reverts commit 5ea0eb0c705224b82a5c284dc5f7722e1f71199f.

This breaks the HHVM clang build test.

Test Plan: clang will pass now

Reviewed By: smith@fb.com

Subscribers: dmitrys

FB internal diff: D1426399

Tasks: 4667712

Summary:
on advice of @tudorb, move most of `folly::Formatter` into `folly::BaseFormatter` so that we can use compile-time polymorphism to provide different types of Formatters.

I wasn't able to get the recursive formatter to be polymorphic -- whenever I tried to convert `class FormatValue<Formatter<containerMode, Args...>, void>` into `class FormatValue<BaseFormatter...`, `FormatTest.cpp:Test(Format, Nested)` wouldn't compile because it couldn't find the template. @tudorb, if you have an easy fix for this, lmk, otherwise I'm (reluctantly) okay with requiring that `Formatter`s define their own nesting `FormatValue`.

phew. the last time I did this sort of metaprogramming was over 5 years ago in perl. Doing it in C++ is... interesting.

Test Plan: `fbconfig -r thrift folly cold_storage && fbmake dbg && fbmake runtests`

Reviewed By: tudorb@fb.com

Subscribers: tudorb, dgp

FB internal diff: D1422343

Tasks: 4624268

Summary:
Reported externally:

https://github.com/facebook/folly/issues/70
https://github.com/facebook/folly/issues/71
https://github.com/facebook/folly/issues/72
https://github.com/facebook/folly/issues/73

Note that I can't test on libc++ myself, but the reports suggested fixes which
sounded good.

Test Plan: fbconfig -r folly && fbmake runtests_opt

Reviewed By: marcelo.juchem@fb.com

Subscribers: jhj, ntv, lesha, kma, fugalh, jdelong

FB internal diff: D1421029

Summary:
We've wanted to use pipe2 in Subprocess for a while, but that's not supported
on glibc 2.5.1. This is not a problem in OSS (autoconf can solve this),
but, internally, we don't run autoconf; add an internal platform include file
with such per-platform differences.

Test Plan: fbconfig -r folly && fbmake runtests_opt

Reviewed By: meyering@fb.com

Subscribers: jhj, lesha, kma, agallagher

FB internal diff: D1422128

Summary: thatwaseasy

Test Plan: iloveunittests

Reviewed By: lucian@fb.com

Subscribers: philipp

FB internal diff: D1419848

Tasks: 4636617

Summary:
* Moved the fast toLower code into folly
* Updated the name to emphasize that it's ASCII-only

Test Plan: * Unit tests and benchmarks included.

Reviewed By: tudorb@fb.com

Subscribers: ruibalp, bmatheny

FB internal diff: D1421056

Summary: Not sure why this shadow error didn't turn up in my tests, but whatevs

Test Plan: traffic manager builds without warning-errors

Reviewed By: suhas@fb.com

Subscribers: hannesr, net-systems@, fugalh, exa

FB internal diff: D1421495

Tasks: 4653938

Summary: wangle/detail.h -> wangle/detail/State.h

Test Plan: OSS build

Reviewed By: meyering@fb.com

Subscribers: fugalh, lesha

FB internal diff: D1421490

Summary:
Use namespace gflags going forward, import namespace google into it if
backward.

Also added a few "checking for..." messages in autoconf.

Test Plan: fbconfig -r folly && fbmake runtests_opt, OSS build

Reviewed By: meyering@fb.com

Subscribers: fjargsto, ntv, jhj, lesha, kma, davejwatson

FB internal diff: D1420575

Summary:
Instead of returning a Later, `via` returns a cold future.
This works without keeping a backreference like Later does, because an inactive Future will always activate on destruction. Alternatively we could have an extra Promise, a la Later, and pass that along like Later does, and require launch() at the end (though, implicit launching on destruction would be an option there too).

If you think this approach is viable I'll clean it up on Wednesday: make sure all the calling sites work, etc.

Test Plan:
new unit test
This may fail in contbuild, I haven't done the codemod for calling sites, if there are any.

Reviewed By: hannesr@fb.com

Subscribers: jsedgwick, net-systems@, fugalh, exa

FB internal diff: D1412499

Tasks: 4480567

Summary:
Bring a bit more sanity to the lifetime. Now Future and Promise detach by calling the respective methods on State, and they do it during their destruction only. State manages its own lifetime.

Besides being cleaner, this also sets the stage for cancellation (where we'll need Future to hang on to its reference to State after setting the callback), and for folding in Later (we will have a bool for whether the future is hot and hold off executing the callback if it isn't).

Also cleans up various things I noticed while auditing the code for usages of `state_`.

Test Plan:
All the unit tests still pass.
Ran with ASAN (and found I had introduced a bug, then fixed it. yay)

Reviewed By: hannesr@fb.com

Subscribers: jsedgwick, net-systems@, fugalh, exa

FB internal diff: D1412038

Tasks: 4618297

Summary: Suggested by @aalexandre.

Test Plan: Compile.

Reviewed By: andrei.alexandrescu@fb.com

Subscribers: folly@lists, njormrod, aalexandre

FB internal diff: D1246180

Test Plan: copied from folly::json

Reviewed By: philipp@fb.com, soren@fb.com

FB internal diff: D1417992

Tasks: 4527315