Summary:
Compile out instead of erroring.

In an ideal world, we'd have a fallback that would work across platforms
(i.e. no SSE, no unaligned 32-bit writes etc.) and compile some version
of GroupVarint in all environments. I actually tried this; the SSE stuff
is all behind #if __SSSE3__ already, so I thought it could work (modulo
the unaligned-writes problem). I ran into problems with the
SSSE3-vs.-not distinction that @simpkins alluded to in D652764, and
decided I'd rather not open that can of worms at the moment.

Test Plan:
fbmake runtests. Manually force the #ifs to false and make
sure fbmake runtests still passes (although GroupVarintTest is empty).

Reviewed By: delong.j@fb.com

FB internal diff: D747150

Summary:
Go with the fat header approach. Merge these two into Memory.h. We could, potentially, include Malloc.h as well, but it fbstring header uses the once define for some special magic. Leave it alone for now.

An alternate approach might be moving all three leaner headers into a `memory/` subdir with `folly/Memory.h` just #including the three.

Test Plan:
fbconfig folly/tests && fbmake runtests_opt

Reviewed By: delong.j@fb.com

FB internal diff: D745873

Summary:
Certain operations should not be performed on infinite sequences
(e.g. sorting, left-folds, summation).  In some cases, we can
detect that a sequence is infinite at compile-time and provide
a static_assert to prevent such dangerous operations.

Test Plan:
Manually created cases where the operation should
be disallowed.  Compiler correctly raised an error.

Reviewed By: tjackson@fb.com

FB internal diff: D740011

Summary: See title.

Test Plan: Inspection.

Reviewed By: delong.j@fb.com

FB internal diff: D745883

Summary:
Generate an `fbvector` of `struct iovec` suitable for using with `writev` or
`sendmsg`.

This code is pretty straightforward, but Adam pointed out that something along
these lines has already been done in thrift, so I followed that code closely.
http://fburl.com/11586814

Test Plan:
fbmake runtests

I am using also this in a prototype and it's working there.

Reviewed By: agartrell@fb.com

FB internal diff: D744055

Summary: allow to dequeue the first IOBuf in an IOBufQueue

Test Plan: throughly tested with some dependent code in proxygen

Reviewed By: tudorb@fb.com

FB internal diff: D732484

Summary:
Aligned loads are faster on some architectures. Let's refactor
qfind_first_of so that it uses aligned loads when possible.

Also modify the benchmarks to account for begin/end-of-string
logic.

Test Plan:
Tests pass. Updated benchmarks.  Generally seeing a 5-20%
speed-up, depending on the situation.

Reviewed By: philipp@fb.com

FB internal diff: D720369

Summary:
We don't do this today, but it looks like std::string does. For longer, similar strings, this is a big win.

Before:

```lang=text
============================================================================
./folly/test/FBStringTestBenchmarks.cpp.h       relative  time/iter  iters/s
============================================================================
BM_equality_string(65536)                                    5.13ms   194.87
BM_equality_fbstring(65536)                                 11.34ms    88.18
============================================================================
```

After:

```lang=text
============================================================================
./folly/test/FBStringTestBenchmarks.cpp.h       relative  time/iter  iters/s
============================================================================
BM_equality_string(65536)                                    5.01ms   199.74
BM_equality_fbstring(65536)                                  6.63ms   150.78
============================================================================
```

Test Plan: Benchmark, unit  tests

Reviewed By: tudorb@fb.com

FB internal diff: D737482

Summary: Otherwise you get errors like `error: array used as initializer ./folly/experimental/StringGen-inl.h: In constructor ‘folly::gen::detail::UnsplitBuffer<Delimiter, OutputBuffer>::UnsplitBuffer(const Delimiter&, OutputBuffer*) [with Delimiter = char [3] ...]`, since literal strings bind as reference to fixed-length character arrays. Providing an explicit overload for `const char*` fixes this.

Test Plan: Unit tests

Reviewed By: tulloch@fb.com

FB internal diff: D737117

Summary:
template <typename SomeInt>
void foo(SomeInt x) {
// yields an error in clang when SomeInt is unsigned and -Werror is used
if(x <= 0) {
//...
}
}

Test Plan: added unit tests

Reviewed By: andrei.alexandrescu@fb.com

FB internal diff: D735735

Summary:
std::hash is not awesome and not configurable.  Typical cases you might want to customize are:
string: I happen to know that fnv isn't super awesome, for example, and that's what folly uses for std::hash fbstring.
pointers: you may want to hash the contents of the pointer instead of the address for certain types.

This is a very simple diff that lets you do that.  It provides StdHasher that passes through to std::hash and uses that for hash_combine, so this should be 100% backward compatible.

Test Plan: test_hash.  I will add another test for using a hasher besides StdHasher shortly.

Reviewed By: delong.j@fb.com

FB internal diff: D733899

Summary:
GENERATOR was broken if you didn't `using namespace folly::gen`, and
we're still copying quite a few functors where we could move them.

Test Plan: Unit tests

Reviewed By: chaoyc@fb.com

FB internal diff: D731253

Summary:
See new test.  Under GCC 4.6 (which is what the folly tests build with) the old code works fine but under 4.7 building fails with:

folly/test/OptionalTest.cpp: In member function ‘virtual void Optional_AssignmentContained_Test::TestBody()’:
folly/test/OptionalTest.cpp:122:14: error: use of deleted function ‘ContainsOptional& ContainsOptional::operator=(const ContainsOptional&)’
folly/test/OptionalTest.cpp:106:21: note: ‘ContainsOptional& ContainsOptional::operator=(const ContainsOptional&)’ is implicitly deleted because the default definition would be ill-formed:
folly/test/OptionalTest.cpp:106:21: error: use of deleted function ‘folly::Optional<int>& folly::Optional<int>::operator=(const folly::Optional<int>&)’
In file included from folly/test/OptionalTest.cpp:17:0:
./folly/Optional.h:84:7: note: ‘folly::Optional<int>& folly::Optional<int>::operator=(const folly::Optional<int>&)’ is implicitly declared as deleted because ‘folly::Optional<int>’ declares a move constructor or move assignment operator
folly/test/OptionalTest.cpp:129:30: error: use of deleted function ‘ContainsOptional& ContainsOptional::operator=(ContainsOptional&&)’
folly/test/OptionalTest.cpp:108:21: note: ‘ContainsOptional& ContainsOptional::operator=(ContainsOptional&&)’ is implicitly deleted because the default definition would be ill-formed:
folly/test/OptionalTest.cpp:108:21: error: non-static data member ‘ContainsOptional::opt_’ does not have a move assignment operator or trivial copy assignment operator
folly/test/OptionalTest.cpp:137:14: error: use of deleted function ‘ContainsOptional& ContainsOptional::operator=(const ContainsOptional&)’

Test Plan:
(1) fbconfig folly/test && fbmake dbg && _build/dbg/folly/test/optional_test
(2) Remove folly/PLATFORM to build with gcc 4.7, then repeat (1).  Without the code fix, the new test fails to build.  With the fix, the test builds and runs fine.

Reviewed By: tjackson@fb.com

FB internal diff: D732402

Summary:
When compiling folly with clang, the compiler warns about our
use of GCC-specific pragmas to silence incorrect compiler warnings:

folly/Optional.h:79:33: warning: unknown warning group '-Wpragmas', ignored [-Wunknown-pragmas]
folly/Optional.h:80:33: warning: unknown warning group '-Wmaybe-uninitialized', ignored [-Wunknown-pragmas]

Thankfully, those incorrect compiler warnings are not emitted by
clang, so we can just disable the pragmas in clang.

Test Plan:
Built folly in gcc and ran it through clang. Warning above
is gone.

Reviewed By: andrei.alexandrescu@fb.com

FB internal diff: D733323

Test Plan: async_io_test

Reviewed By: philipp@fb.com

FB internal diff: D730100

Summary:
In hopes of keeping 'busywork' helpers out of folly, I've moved this
closer to the code that needed to do this.

Test Plan: Unit tests

Reviewed By: andrei.alexandrescu@fb.com

FB internal diff: D729194

Blame Revision: D726916

Summary:
Now that we have truthy files, it would be nice to be able to simply
//try// opening files and return a possibly-initialized File.

Test Plan: Unit tests

Reviewed By: chaoyc@fb.com

FB internal diff: D726916

Summary:
File has a default constructor so it can be initialized late, but it
doesn't have a good canonical way to see if it's been initialized. This adds an
//explicit// operator bool so you can test files like `if (file) ...`.

Test Plan: Unit tests

Reviewed By: chaoyc@fb.com

FB internal diff: D726914

Summary: Added Adaptor(size_t, const value_type&)

Test Plan: test added

Reviewed By: soren@fb.com

FB internal diff: D726358

Summary:
'open() failed' isn't too helpful. Seeing a filename there is. So I
added it.

Test Plan: Unit test

Reviewed By: mohittalwar@fb.com

FB internal diff: D725204

Summary:
Add a --bm_min_iters flag to control the minimum number of iterations
that the benchmark code starts with on each epoch.

This can be used on benchmarks that test very cheap operations, but take
a long time to set up.  Otherwise the benchmark code may have to retry
many times before it hits a large enough number of iterations to get a
meaningful result, and each time it still pays the fixed setup cost.

This also helps with benchmarks when some of the setup cost cannot be
hidden with BenchmarkSuspender for some reason.  --bm_min_iters can be
set to a large enough value so that the extra startup cost will not
affect the measurements too much.

Test Plan:
Used this with the thread local stats benchmark.  During setup/cleanup,
this benchmark starts and synchronizes with many threads.  The entire
setup time cannot be reliably hidden with BenchmarkSuspender; the
synchronization between the threads takes a relatively long time
compared to the cost of the operation being benchmarked.  --bm_min_iters
allows a relatively high number of iterations to be used, masking this
cost.

Reviewed By: rajat@fb.com

FB internal diff: D723304

Summary:
Helper method for creating a folly::Optional<T> by just passing
in a T reference.  Analogous to boost::make_optional.

Test Plan: Added test case

Reviewed By: tjackson@fb.com

FB internal diff: D721762

Summary: MemoryMapping::data() returns StringPiece ::range() returns ByteRange

Test Plan: tests

Reviewed By: philipp@fb.com

FB internal diff: D720985

Summary:
There are platforms other than clang that don't support ifuncs. (The one
I'm concerned about is ARM.) I changed the ifdef __clang__ around the
ifunc attributes to be more abstract, so we can can pass in this flag on
the command line, or use autoconf to detect it.

Test Plan:
fbmake runtests. Manually define HAVE_IFUNC 0 and make sure the
popcount() and popcountll() functions get compiled as calls to
popcount_builtin.

Run autoreconf, ./configure, make sure the feature gets detected
properly by looking at config.h.

Reviewed By: andrewjcg@fb.com

FB internal diff: D712192

Summary:
This flag conflicts with the flag with the same purpose in
`common/files/MemoryMappedFile.cpp`. Just renaming it for now.

Test Plan: Build

Reviewed By: lucian@fb.com

FB internal diff: D717067

Summary: MemoryMapping is a C++ wrapper object around mmap. It works with `folly::File`s, and provides bitwise-range access for reading and writing files.

Test Plan: Unit test

Reviewed By: lucian@fb.com

FB internal diff: D452384

Summary: Change State to enum class, and add debugging helper to AsyncIOOp and AsyncIO.

Test Plan: n/a

Reviewed By: philipp@fb.com

FB internal diff: D715676

Summary: Moving File into `/folly`.

Test Plan: Same unit tests, rebuild affected code outside folly.

Reviewed By: philipp@fb.com

FB internal diff: D714462

Summary:
Our SSE version of find_first_of was reading past the end of
the StringPiece in some cases, which (very rarely) caused a seg-fault
when we were reading outside of our allotted virtual address space.

Modify the code to never read past the end of the underlying buffers
except when we think it's "safe" because we're still within the same
page. (ASSUMPTION: if a process is allowed to read a byte within a
page, then it is allowed to read _all_ bytes within that page.)

Test Plan:
Added tests that verify we won't go across page boundaries.

Sadly, this code hurts our benchmarks -- sometimes by up to 50% for
smaller strings.

Reviewed By: philipp@fb.com

FB internal diff: D707923

Blame Revision: D638500

Summary:
AsyncIOOp no longer requires derivation to be able to use callbacks; the
callback is passed in.  This makes composition easier (see AsyncIOQueue, added
in this diff).

Test Plan: async_io_test, test added

Reviewed By: lucian@fb.com

FB internal diff: D709648

Test Plan: No

Reviewed By: philipp@fb.com

FB internal diff: D709795

Summary: Compilation is failing in 'opt' build

Test Plan: compile

Reviewed By: andrewjcg@fb.com

FB internal diff: D707869

Test Plan: No

Reviewed By: philipp@fb.com

FB internal diff: D707792

Summary: Minor clang compiler fixes

Test Plan: - fbmake runtest

Reviewed By: andrewjcg@fb.com

FB internal diff: D707663

Summary: It is set once, and is never changed afterwards.

Test Plan: compiled, unit tests

Reviewed By: tudorb@fb.com

FB internal diff: D706125

Summary: codemod with 'Yes to all'.

Test Plan: None really.

Reviewed By: abirchall@fb.com

FB internal diff: D693769

Test Plan: folly/io/test, both dbg and opt, standalone and in valgrind

Reviewed By: philipp@fb.com

FB internal diff: D702755

Summary: TSIA

Test Plan: Unit tests

Reviewed By: tulloch@fb.com

FB internal diff: D701890

Test Plan: async_io_test

Reviewed By: lucian@fb.com

FB internal diff: D699175

Summary:
- Add CursorBase::operator-() implementations for Cursor and BufType;
useful for figuring out the distance between two objects

Test Plan: - Used in some other code

Reviewed By: simpkins@fb.com

FB internal diff: D690046