Summary:
Added missing installation of UTF8String.h
Closes https://github.com/facebook/folly/pull/813

Reviewed By: yfeldblum

Differential Revision: D7506038

Pulled By: Orvid

fbshipit-source-id: 9821a0040c44715cfd68852743b64a32df0dcf57

Summary: MSVC 2015 Update 3 can't parse these with curly-brace init'd, so swap the assignment. The result is the exact same.

Reviewed By: simpkins

Differential Revision: D7508756

fbshipit-source-id: 5e60607f978a87ae8bfed22086d1dccef15759ac

Summary:
Optimized builds of //folly/experimental/coro test Coro.Sleep
would almost always fail because the compiler would generate
code that would only sleep for 0.98 seconds. Perform some
simple rounding in order to ignore slight optimizations like
these.

Reviewed By: wqfish

Differential Revision: D7496416

fbshipit-source-id: f81ff30922fbf3e5d2086fde50f4a469bdf72e51

Summary:
F14HashToken encapsulates the work of hash_function(), which
is the same across F14 containers that have the same key_type and equal
hash_function() (same type and operator==).  This diff adds documentation
that callers may rely on this fact, which means they can use this feature
to reduce work if a single key is access in multiple maps/sets.

(Note: this ignores all push blocking failures!)

Reviewed By: shixiao

Differential Revision: D7506737

fbshipit-source-id: 321032d619434501b1e0544962e54732bfc4c970

Summary:
Allow the user to give their own pair implementation. This allows
using a packed pair.

Reviewed By: yfeldblum

Differential Revision: D7504521

fbshipit-source-id: 11888594b8b3ef3f530ebd95b8006ef6f5afbc65

Summary: Terabytes aren't Web Scale.

Reviewed By: ot, philippv, luciang

Differential Revision: D7479752

fbshipit-source-id: fd353933e642b1ee024c28107c8fcb6a55fe755e

Summary: Until I do a PR on `double_conversion`, we need to work around this by catching the trailing 'e' case.

Reviewed By: luciang

Differential Revision: D7481130

fbshipit-source-id: 2734dbe834e6cd69c6dfe41d6b4e9a7a548a7da8

Summary:
When `startFail()` is called, it sets the state to StateEnum::ERROR.
There is a chance that immediate read callback is scheduled to run in
next event loop. The callback calls handleRead() then triggers
`assert(state_ == StateEnum::ESTABLISHED)` in `AsyncSokcet::handleRead`.

Cancel any immediate read callback when startFail is called.

Reviewed By: yfeldblum

Differential Revision: D7051956

fbshipit-source-id: 0d6693d851b272355a4f36ab8d001969fa47750e

Summary:
At `-O0` each coroutine call results in a heap allocation. This
combined with the increased memory pressure when using ASAN
results in an invalid memory access beyond the maximum address.
Avoid the problem by limiting the recursion to 5,000 frames deep,
instead of 10,000. This still fulfills the intent of the test,
but without pushing the limits of ASAN.

Reviewed By: wqfish

Differential Revision: D7503976

fbshipit-source-id: 7850a9d9db23328d42c25b74a1290caa7773e6a7

Summary: sendRequestSync should only be implemented by channels, which own threads looping EventBases.

Reviewed By: phoad

Differential Revision: D7464392

fbshipit-source-id: bde6edc85438bb406592291330869972f88c787a

Summary:
Using an array of prehash tokens is quite inconvenient without
the ability to default construct them.

Reviewed By: shixiao

Differential Revision: D7490113

fbshipit-source-id: f802a7dc6dc2c603c4682e5f384d9f135220cd54

Summary:
Apple Clang 9 (clang-900.0.39.2) on macOS 10.13.3, when configured with
`FOLLY_USE_LIBCPP`, fails to build `SynchronizedTest.cpp` with the following
error:

    In file included from /Users/awang/code/folly/folly/test/SynchronizedTest.cpp:20:
    /Users/awang/code/folly/folly/Synchronized.h:511:9: error: no matching constructor for
          initialization of 'folly::Synchronized<int, NonDefaultConstructibleMutex>'
          : Synchronized{std::piecewise_construct,
            ^           ~~~~~~~~~~~~~~~~~~~~~~~~~~
    /Users/awang/code/folly/folly/test/SynchronizedTest.cpp:539:25: note: in instantiation
          of function template specialization 'folly::Synchronized<int,
          NonDefaultConstructibleMutex>::Synchronized<int &&, int &&>' requested here
      auto&& synchronized = folly::Synchronized<int, NonDefaultConstructibleMutex>{
                            ^
    /Users/awang/code/folly/folly/Synchronized.h:499:12: note: candidate constructor not
          viable: no known conversion from 'const std::__1::piecewise_construct_t' to
          'in_place_t' (aka 'folly::in_place_tag (&)(folly::in_place_tag)') for 1st argument
      explicit Synchronized(in_place_t, Args&&... args)
               ^
    /Users/awang/code/folly/folly/Synchronized.h:764:3: note: candidate template ignored:
          could not match 'std::__1::integer_sequence' against 'folly::integer_sequence'
      Synchronized(
      ^
    <other constructors not viable due to number-of-parameter mismatches>

The proximate cause of the error is that `make_index_sequence` is
generating a `folly::index_sequence`, but the delegatee constructor only
accepts `std::index_sequence`.

Changing the delegatee constructor to take an unqualified `index_sequence`
solves the problem, since the return type for an unqualified
`make_index_sequence` should be the same as the unqualified
`index_sequence`.

This appears to have been introduced in
8581e9e0.

Clang ends up using the Folly implementation of
`make_integer_sequence` because `__cpp_lib_integer_sequence` is
still not defined, even though `integer_sequence` support was added
to libc++ way back in mid-2013, in revision 185343. There is a bug open
 on the LLVM bug tracker (#18584 [1]) to add the
`__cpp_lib_integer_sequence` and other feature test macros,
but the last activity on it was in late 2014. This results in
the compiler successfully finding the definition for `std::index_sequence`
(unlike what would happen with C++11 compilers), but using the folly
implementation for `make_index_sequence`.

Qualified `std::index_sequence` is used in a few other places, but `std::`
was not removed from those usages because they reside in functions that
have C++14 features that would prevent a C++11 compiler from compiling
them at all, let alone with a `std::` v.s. `folly::`
mismatch:

    - folly/test/FormatBenchmark.cpp
        - decltype(auto) is used for format20Numbers()
    - folly/container/Foreach-inl.h
        - decltype(auto) and generic lambdas are used in a few of the
          for_each_range_impl() overloads
        - decltype(auto) is used in some of the impl() overloads
        - decltype(auto) is used as the return type for fetch()
    - folly/container/Iterator.h
        - decltype(auto) is used as the get_emplace_arg() and unwrap_emplace_arg() return type

A different way of solving the problem would be to add a special case
for Clang to the `integer_sequence` check, similar to what is done for
MSVC, but that seemed like partially papering over the problem. The best
solution would be to get those feature macros into libc++, but that
wouldn't help for already-released versions of LLVM.

    [0]: https://blogs.msdn.microsoft.com/vcblog/2014/06/06/c14-stl-features-fixes-and-breaking-changes-in-visual-studio-14-ctp1/
    [1]: https://bugs.llvm.org/show_bug.cgi?id=18584
Closes https://github.com/facebook/folly/pull/810

Reviewed By: aary

Differential Revision: D7482813

Pulled By: yfeldblum

fbshipit-source-id: 36310acbc227bfe646d6fad877a406b8d57407e5

Summary:
Continuing from D7482813, ensure everywhere in folly is using the folly versions of integer and index sequences.
Also implement index_sequence_for and use it.

Reviewed By: aary

Differential Revision: D7490483

fbshipit-source-id: 082209dd6c6ca11f67b38dbe4529d46e9dce57ce

Summary:
This has functionality not overlapping with `FBString` or `folly::fbstring` or `folly::StringPiece`; and a pretty expensive dependency on `boost/regex`.

Much of the thrift-gen `_types.h` code comes from this inclusion chain:

* `_types.h` uses strings (for `str` fields in thrift structs); these are represented by `folly::fbstring`, from...
* `folly/String.h` includes `FBString.h` and other things to support string-related stuff; simple enough.  But it has a class `UTF8StringPiece`, which uses an iterator, from Boost called `boost::u8_to_u32_iterator<char const *>`.  So then we have...
* `boost/regex/pending/unicode_iterator.hpp` includes tons of things in the usual boost header avalanche: iterator facades, lots of `mpl` code.

Reviewed By: meyering

Differential Revision: D7373577

fbshipit-source-id: efdb920b6fe8c7b7c57d3becf2bf0710975fcf73

Summary:
Upon `erase`, F14 vector containers destroys the value and erases the item
(i.e., index into the `values_` vector) from the underlying hashtable. However,
the item still needs to be hashable when erasing from the hashtable, so we have
to destroy the value _afterwards_.

This diff fixes the bug.

There are a couple of reasons that this was previous undetected / did not cause
a problem:
  * for POD types, `allocator_traits::destroy` is a no-op.
  * this code path is only hit if the chunk of the destroyed item has hosted
    overflowed items
  * the use was immediately after free
  * our test coverage did not have vector policy + non-SSO string keys

Reviewed By: nbronson

Differential Revision: D7488050

fbshipit-source-id: ea29e875a0c7a39b8deed40a15777a6983438836

Summary:
A library that provides gflags-like functionality, with the addition of:
- Threadsafe read/writes for all flags
- Arbitrary C++ types
- No declare/define split, all setting definitions are declaration and are safe to place in headers
- Runtime introspection of all flags
- Compile time and runtime collision checks
- Namespaces per project

Down the road:
- Overlay support (change a set of settings based on some context)
- Construct arbitrary user types from string
- Support multiple library instances with distinct settings in a single process
- Callbacks on setting updates

Reviewed By: aary, mnv104

Differential Revision: D6217956

fbshipit-source-id: 7c03d79c7601ce7b266df78e62b7193712d8331e

Summary: This diff change the `FOLLY_USE_SYMBOLIZER` gates to check for a value of `1`.

Reviewed By: yfeldblum

Differential Revision: D7471272

fbshipit-source-id: 74dbb4d56e1adc56ec75af5b4950ca2c027851cc

Summary:
We used `link_whole` to make sure the `setStackTraceGetter` constructor is
always linked when using Singleton's to setup symbolizer for stack traces.
However, using `link_whole` means we always link in the singleton library, even
if it's never used.

This diff merges `SingletonStackTrace.cpp` into `Singleton.cpp` to avoid the
need for `link_whole` and to link the former if and only if we use the latter.

Reviewed By: pixelb

Differential Revision: D7230733

fbshipit-source-id: de9be51a4eefee201563523d25800ab19ea777e3

Summary: These were being mangled as C++ functions rather than C functions.

Reviewed By: yfeldblum

Differential Revision: D7462668

fbshipit-source-id: e4e83e16aafad3d888599dd78bdbb4ac2711ad7f

Summary: It's unused, remove it.

Reviewed By: yfeldblum

Differential Revision: D7462683

fbshipit-source-id: 604c5691d69f2576c83c25350d29515efbc77a7b

Summary: `timespec` is defined in `sys/time.h`, so don't rely on `<chrono>` including it.

Reviewed By: yfeldblum

Differential Revision: D7462731

fbshipit-source-id: 590954e2230a0da322e2f5eb966bb49b18aa56f7

Summary:
Add API to AsyncUdpSocket to support three options on the socket.

- SO_RCVBUF and SO_SNDBUG options are responsible for the size of the queues that contain received and sent messages, respectively.
- SO_BUSY_POLL sets the approximate time in microseconds to busy poll on a blocking receive when there is no data

Details in socket man: http://man7.org/linux/man-pages/man7/socket.7.html

Reviewed By: yfeldblum

Differential Revision: D7372123

fbshipit-source-id: 5fd92489cbe86879d63df208c3821febdadb0020

Summary: MSVC calls this out explicitly as a non-standard cast and errors.

Reviewed By: yfeldblum

Differential Revision: D7451404

fbshipit-source-id: 9261b15fdc47b6ffbc1b53e83cbfbe9f9af894a9

Reviewed By: ouj

Differential Revision: D7424982

fbshipit-source-id: 3190564d3e79d10d7d64c1822a53e058a201f87d

Summary:
Remove hazptr_obj_batch feature.  Not required after D7429494:

After timer runs out, *all* thread caches are cleaned up, not just the current one,
so it should be safe to push anything to local thread cache, even if thread isn't used often,
as long as *some* thread is used.

Reviewed By: magedm

Differential Revision: D7433978

fbshipit-source-id: 01508fd199fcd267550efc08d2695b53f1569c84

Summary:
Continued reclamation issues from lots of UnboundedQueue batches, or lots of threads.

In addition to the current number of outstanding objects, cleanup *all threads* on a timer
basis.

Reviewed By: magedm

Differential Revision: D7429494

fbshipit-source-id: 7e9bdae9ae49927ca0b6e3f24963a4d4b23a4f0b

Summary:
I would like to store a folly::Poly I'm defining in an ordered
container (boost::containers::flat_set). While I could implement operator<
for exactly my type, it seems nicer to add it as another interface all
Poly users can reuse.

Reviewed By: ericniebler

Differential Revision: D7415861

fbshipit-source-id: 1912eaa009acea5c243db3e14fc0715a0829f251

Summary:
[Replaceable] Move `Replaceable` mixin param defaults to base template.

Saves a few lines overall.

Reviewed By: WillerZ

Differential Revision: D7421561

fbshipit-source-id: f56d1ffdd081b0a8a20cb7a492f06c9647445782

Summary: This check is causing some issues in github commits. Figuring out exactly which versions of 4.9 have the bug is a pain, so let's just get rid of it. See D6695110 for more context.

Reviewed By: yfeldblum

Differential Revision: D7431713

fbshipit-source-id: 17d1b025edf52592495c5f69c5a15025b33b3895

Summary: Windows doesn't have a nice function to do this, so we have to do it ourself, throwing the right exception with a struct that no public header defines, because Windows.

Reviewed By: akrieger

Differential Revision: D5100900

fbshipit-source-id: 4f6840e50c5c1aef36adade098739201c97b9af7

Summary: Actually remove it. All uses of it (outside of thrift) were removed in the previous diff.

Reviewed By: ot

Differential Revision: D7422614

fbshipit-source-id: 6a59c34b3cbc0ee7c50f8cbf8c5e641352f2db1a

Summary:
There is a missing `#include <ostream>`.  Add it.
Closes https://github.com/facebook/folly/pull/807

Reviewed By: shixiao

Differential Revision: D7425693

Pulled By: yfeldblum

fbshipit-source-id: cb645f7da94ca5fe5f8d4314b99621557109f360

Summary:
Python is considered a Safe Enivronment, where things like aborts and segfaults
can not happen.

If clients where used outside of "async with" we use to print a error message
and then C++ would abort.

This fix makes sure C++ no longer aborts but we still print the message
that the "async with" should be used.

All futures are now shielded so cancel will not produce strange exceptions when
C++ decides to fullfill the canceled future.

Also we now pass a tuple of client and future to all bridgeFutureWith calls.
This has the affect of making sure the client can not GC before the future
is completed.

We now use a unique_ptr for RequestChannel so there is no more race to delete
the shared pointer.

Reviewed By: yfeldblum

Differential Revision: D7389380

fbshipit-source-id: 6a3ca019190907e726c550c9b379a2a2610d0b0a

Summary: `-Wextra-semi` became more strict. This fixes folly build with the latest clang.

Reviewed By: yfeldblum

Differential Revision: D7359992

fbshipit-source-id: ce7caf05aa536f35e9af4ed33a86072afd09f387

Summary: Revert changes accidentally slipped into D7405057.

Reviewed By: yfeldblum, Orvid

Differential Revision: D7419691

fbshipit-source-id: ab9bea3393eb33aa19b3009263a53ac91e30692c

Summary: folly::ThreadLocal already maintains this list for us.  Add an accessor in folly::SingletonThreadLocal, and use it

Reviewed By: yfeldblum

Differential Revision: D7264843

fbshipit-source-id: 45278f7a074e50b96b451437c327de5d6713363a

Summary: [Folly] Fix `retrying` when the policy throws.

Differential Revision: D7392852

fbshipit-source-id: 3cb1d19697f111bf19047dc382056b39ad241035

Summary: Every compiler we currently support supports the `[[deprecated]]` attribute, so use that instead.

Reviewed By: yfeldblum, ot

Differential Revision: D7405895

fbshipit-source-id: 541c537cd7a0be0d901378c0e79f5cd0f55b85e3

Summary: Adding more details to the documentation and a unit test which checks various executors.

Reviewed By: yfeldblum, djwatson

Differential Revision: D7410461

fbshipit-source-id: 2ae6926f43862ec45a06ff76dbacae681fe6feaa

Reviewed By: wqfish

Differential Revision: D7205067

fbshipit-source-id: 2236c132b2150e38da6d5a5c9970347320e19a60