Summary:
Freshly opened TCP sockets can have a send buffer size bigger
than wmem_default, because connection parameters are cached in the
route cache.  Some of the async socket tests assumed that writing 8M of
data was enough to guarantee blocking behavior, but since wmem_max is
20M on our systems the resulting tests are flaky.

Reviewed By: marcinpe

Differential Revision: D4292201

fbshipit-source-id: ba5d606d080330e455eee2b17bcae6cf546bf981

Summary:
Add the default move constructor and move operator
to EvictingCacheMap so it can be moved.

Reviewed By: markisaa, luciang

Differential Revision: D4287472

fbshipit-source-id: 130e9d5467d6da14ba95a9e769cf8e8d541a704c

Summary: UBSAN instrumentation results in more stack being used. Applying the same logic we already have for ASAN there.

Reviewed By: igorsugak

Differential Revision: D4287455

fbshipit-source-id: 1a224cb5a3654c23b15fa298bf80476234a1418c

Summary: MacAddress only needs to include ostream and the test header does not need iostream.

Differential Revision: D4286405

fbshipit-source-id: e64fdd4e8428c62279ed7c185188f3dfc14096d3

Summary:
We do not need it and it adds a static initialization

This is the last extra one in liger. The other places actually need iostream to print things

Reviewed By: yfeldblum

Differential Revision: D4286095

fbshipit-source-id: 44a827e48e9fb9106626e784592eb194dff83bf6

Summary:
The benchmark API only supports `unsigned int` as the iteration number, so this being 64-bit means it has to be truncated for no reason.
This change gets us closer to being able to enable MSVC's implicit truncation warnings.

Reviewed By: yfeldblum

Differential Revision: D4281264

fbshipit-source-id: 27397e3f23bace20f3cc457665ea6f21bf994da5

Summary:
I was getting an error when running commands in
https://our.intern.facebook.com/intern/dex/fibers/gdb/

Reviewed By: andriigrynenko

Differential Revision: D4283229

fbshipit-source-id: fe45847d0ec6cadfbce6fe9818066cce3e275a46

Summary: overload constexpr_strlen to work with other character types

Reviewed By: yfeldblum, luciang, Orvid, ot

Differential Revision: D4269654

fbshipit-source-id: 2fcd666669efb51d77e847ff31bda3a301e763da

Reformulate constexpr_min and constexpr_max to achieve stability in sorting as described in http://stepanovpapers.com/notes.pdf

Summary: There is a famous long-standing "bug" in the standard library regarding the semantics of min and max wrt values that are equivalent wrt op< but not equal. Let's not make the same mistake with constexpr_min and constexpr_max.

Reviewed By: yfeldblum, luciang, Orvid, ot

Differential Revision: D4269635

fbshipit-source-id: 19b464c949dc0cf07afb08eaf657ae8b242ca42d

Summary:
Fix potential double close() on exception when SO_REUSEPORT fails.
The fd will be closed twice, once by close() and second by the scope guard above.

Reviewed By: yfeldblum

Differential Revision: D4272300

fbshipit-source-id: a163be822d7522cadb9dc3c3eddca10df453de14

Summary:
MSVC was getting thoroughly confused while trying to eval the constexpr function, so switch it to a template instead.
MSVC was also failing to wrap it's head around what a NodeRecycler is, so delay its resolution via typename.

Reviewed By: yfeldblum

Differential Revision: D3478755

fbshipit-source-id: f687f4538fb12ef8eee525557f4cc988a09e714d

Summary:
This makes the changes required to allow folly to compile cleanly with the sign/unsigned mismatch warnings 4388 and 4389, as well as with warnings 4804 and 4805, which are about comparisons between scalars and bool.
Most of the changes in this are to `DCHECK_*` calls which are translated to a templated call which does the comparison internally based on the inferred type of the argument, which for a literal `0` is `int`, causing the warnings to get generated when the comparison is done.

Reviewed By: yfeldblum

Differential Revision: D4253427

fbshipit-source-id: cd17973a78e948a62c886a2959f9abf40a69f9f5

Summary: This brings the EventBase functionality of getting a timer for the eventbase into the virtual interface

Differential Revision: D4264135

fbshipit-source-id: e5526610eca808e1d7b549a7cd6cd7b2c65d3ce1

Summary:
Add a utility function to more safely set a file's contents by writing to a
temporary file first, then renaming the temporary file into place.  On Linux
systems where renames are atomic, this ensures that the operation either
succeeds or that the old file state is left unchanged on failure.

Note that unlike most of the other APIs in FileUtil.h, I intentionally made
writeFileAtomic() throw an exception on failure.  This is implemented using a
lower-level writeFileAtomicNoThrow() version.  Callers who care about the
exception overhead can use the lower level version instead.  (Exception
overhead should be relatively low compared to the file I/O operations, though,
so I suspect most users will prefer the throwing APIs.)

Reviewed By: yfeldblum

Differential Revision: D4253964

fbshipit-source-id: 5301e2791b82c6f90b63bb509b0411841c266705

Summary:
[Folly] `Future<typename>::getTryVia`.

We have `waitVia` and `getVia` and `getTry`, but `getTryVia` was missing.

Reviewed By: fugalh

Differential Revision: D4254972

fbshipit-source-id: 403e1a3496ad5dfc93c4c55ab75a83bcc89e6c64

Summary:
Fix test for AsyncSSLSocket. When a timeout
occurs during a handshake it's possible that
invokeConnectError is called with sslState_ ==
CONNECTING. This just removes the assertion in
this case. This doesnt affect prod code, only
tests

Reviewed By: yfeldblum

Differential Revision: D4252677

fbshipit-source-id: 4db3a94ae28aa140546554c087bedcee7f06d201

Summary:
Switching ServiceRouter code to use VirtualEventBase instead of EventBase allows us to have same thread-pool shared between multiple ServiceRouter instances.

In future this allows us to replace EventBasePoolImpl with some generic thread-pool implementation (e.g. IOThreadPoolExecutor). It also makes it trivial to have ServiceRouter run on a any existing thread-pool if necessary.

Reviewed By: smichelson

Differential Revision: D3935283

fbshipit-source-id: 3275bbace15a0df8b06d12c970ccc098a631658d

Summary: folly::SingletonThreadLocal should be used for all thread-local singletons to avoid SDOF.

Reviewed By: yfeldblum

Differential Revision: D4253009

fbshipit-source-id: 5b4ca57a7a77b7b54c94630a2a29d13b84a3af27

Summary:
Truncations should be explicit, but for some reason, MSVC seems to be the only compiler that will warn you when you implicitly truncate integer or float values.
This allows Folly to be compiled with warnings 4018, 4242, 4244 and 4305 enabled.
Technically 4018 is a sign mismatch warning, but there was only one place it was being triggered so I included it anyways. The other 3 are warnings for implicit truncation.

There is one other implicit truncation warning that currently triggers in Folly, 4267, but there are a lot more places where that triggers so I'll do that in a separate diff.

Reviewed By: yfeldblum

Differential Revision: D4249471

fbshipit-source-id: e18a93d85856c998576934a6229c9edd1638a54e

Summary: Using newly added `StaticTracepoint.h`, add a Tracepoint for `RequestContext` switch that could be enabled by compiler flag.

Reviewed By: drarmstr

Differential Revision: D4238658

fbshipit-source-id: 043a035ba8251b48d7b29d63a2219ad2edda164f

Summary: More constexpr is better!

Reviewed By: yfeldblum, luciang, ot

Differential Revision: D4244996

fbshipit-source-id: 30a9b726c115a92bb18538d7f18e50eccb0a1ef6

Summary: If a write request is buffered after a partial write, then bytes written is not updated after subsequent write operations (`performWrite`) for the buffered write request (`BytesWriteRequest`). This results in a wrong value for totalBytesWritten_, which is reported in the error callback in case the write request fails.

Reviewed By: yfeldblum

Differential Revision: D4205743

fbshipit-source-id: f77ca55ccfdceda1008c45e72ec093b00bf250e4

Summary:
Make folly::Singleton fail if singleton is requested before registrationComplete. By doing this we disallow any folly::Singleton to be created before main().

Strict mode is still disabled for Python and Hs.

Reviewed By: yfeldblum

Differential Revision: D4121322

fbshipit-source-id: b66c23e24f6a7324cd12ddb77cad960e0950a1aa

Summary: Closes https://github.com/facebook/folly/pull/520

Reviewed By: pixelb

Differential Revision: D4243347

Pulled By: yfeldblum

fbshipit-source-id: cae0b36abdc7807afc188220f7f421f2a3bbc973

Summary: D4194834 broke the OSS build.  Ubuntu 14.4 has r114 in its repo, so support it.

Reviewed By: yfeldblum

Differential Revision: D4224129

fbshipit-source-id: c85e95716ee1a08b33455bfe6fc9f7712d226edf

Summary: Because MSVC isn't happy about it and generates a warning with a lot of context.

Reviewed By: yfeldblum

Differential Revision: D4242688

fbshipit-source-id: b1e26c9e3b58e7bac0e00ab8959358010fa610e8

Summary:
* include all dependencies
 * update `autoconf` & `configure` with correct params
 * include `make` & `make install` in the bootstrap for a better user experience
 * fixes #332
Closes https://github.com/facebook/folly/pull/513

Reviewed By: yfeldblum

Differential Revision: D4177362

Pulled By: Orvid

fbshipit-source-id: c62d6633c382fca57bb06db08724a7355b71bdb3

Summary:
[Folly] Canonicalize some aspects of the `folly/experimental/hazptr` test.

* Remove the custom `main()` and use the provided weak symbol, either from `folly/test/common/TestMain.cpp` or from `gtest` depending on the build.
* Extract some common setup/teardown (just debug logging in this case) to a test fixture.

Reviewed By: Orvid

Differential Revision: D4234154

fbshipit-source-id: 25613cba0ad6c9147714d98150e013fba1a64d46

Summary:
Thanks nbronson for suggesting this fix. I don't see why this should make any
difference (and nor does clang) but the important thing is that it works now.

Reviewed By: nbronson

Differential Revision: D4212670

fbshipit-source-id: 08b4313d736f237039a807fbc458d4d581a2ef35

Summary:
Otherwise it won't work under MSVC where we don't have Futexes.
This is achieved by using the Futex API already used elsewhere in Folly.

Reviewed By: yfeldblum

Differential Revision: D4233517

fbshipit-source-id: d6ae6a34b2287cf485f8e4a316445a2ba746d1a7

Summary: This reverts commit 636ed5ae8ad36a323054efaad96d3756f1b7b9d0

Differential Revision: D4214296

fbshipit-source-id: 41fe55a8fcbf41cce44fee7ba65eda5aadbf97b3

Summary:
When TFO falls back, it's possible that
the fallback can also error out.

We handle this correctly in AsyncSocket,
however because AsyncSSLSocket is so
inter-twined with AsyncSocket, we missed
the case of error as well.

This changes it so that a connect error on
fallback will cause a handshake error

Differential Revision: D4226477

fbshipit-source-id: c6e845e4a907bfef1e6ad1b4118db47184d047e0

Summary:
[Folly] Suppress deprecation warnings in related tests.

When a test specifically exercises a marked-deprecated library or a marked-deprecated part of a library, we do not need to see the deprecation warnings - they are noise.

Specifically, `Singleton` and `Subprocess` have tested but marked-deprecated APIs, so we suppress deprecation warnings in those specific test suites.

Reviewed By: andrewjcg

Differential Revision: D4230826

fbshipit-source-id: 937b078029cde492906088caa3a9e5eee449bc12

Summary: It's called `.native()` for a reason, and returns a `wstring` on windows, so call `string()` instead.

Reviewed By: yfeldblum

Differential Revision: D4229254

fbshipit-source-id: d5f9658b41f45fd719ca3e878a2e61c9dce24f10

Summary:
Derived from `sdt.h` in [[ https://sourceware.org/systemtap/ | SystemTap ]]
Performed changes and clean-ups including:
- Get rid of the `sdt-config.h`.
- Get rid of "Running in Assembler" code.
- Replace the `DTRACE_PROBE_1, ... DTRACE_PROBE_N` Macro family with a single Macro using the "count Macro arguments" trick
- Get rid of argument signed-ness calculation. A **lot** of painful code were there only for this, and it has no real usage except for displaying purposes.
- Re-structure the helper methods (Macros) so that the logic of the code is a bit more readable.
- Add some comments so that it would be easier for users and future developers to understand what this header file is actually doing.
- Update license banner.

Reviewed By: meyering

Differential Revision: D3792966

fbshipit-source-id: 564a1011f7946d30be78191f334e072d94fa264e

Summary:
Even when linked against the dynamic runtime.
In truth, the old way was far cleaner, but we do need to be able to close the file descriptors as there is a hard global limit set by the CRT when it's compiled of 8k file descriptors.
This closes it by applying a flag that prevents calls to `CloseHandle` from actually closing the handle, then we call `_close` to get the file descriptor itself closed, then we can call `_closesocket` to close the socket and handle.
This is complicated by the fact that attempting to close a handle that you've marked as being prevented from being closed will trigger an SEH exception, but only if you have a debugger attached. That means that, for anything closing sockets to be debuggable, we have to handle that exception (as a no-op) and then get to have fun trying to get `STATUS_HANDLE_NOT_CLOSABLE`, which for some absurd reason, is only defined as part of the driver developer kit's headers. Some fun ensued where VS was nice enough to point out that the bcrypt headers also defines `NTSTATUS`, and doesn't require building as a driver to use it. For that reason we borrow its definition of `NTSTATUS` rather than declaring it ourselves.
We also have to prevent the main windows headers from defining the base SEH exceptions, because those will be redefined by `ntstatus.h` otherwise :(

Reviewed By: yfeldblum

Differential Revision: D4222625

fbshipit-source-id: 6ed7e91b7a735b5506eb189e452983cd8cace34b

Summary: The helper exists because MSVC is a pain in this regard, so use it.

Reviewed By: yfeldblum

Differential Revision: D4224037

fbshipit-source-id: 6c6415877785f13ea91618a3042f2470dccd08c1

Summary:
There was a recent change in boost compatibility that changed fibers internals and thus broke this script.
This diff fixes gdb script to properly print stack trace for fibers.

Reviewed By: yfeldblum

Differential Revision: D4214575

fbshipit-source-id: 4bf69e9b297c7539a2b2b71093998c30d6a2d81c

Summary: Ubuntu 16.10 has a package for zstd 0.5.1 which is too old to be used by folly.  With this change autoconf will check that one of the newer functions is present, disable zstd if not.

Reviewed By: Orvid, philippv

Differential Revision: D4219964

fbshipit-source-id: 1e64dfb62f81f7e61b097cb7d4fd3810da476625

Summary: Because Windows disagrees with everything else about how to specify that you want the wildcard address. It's done with an empty string on Windows, but `nullptr` everywhere else.

Reviewed By: yfeldblum

Differential Revision: D4216970

fbshipit-source-id: b5dc136946d9677a96be3252e44d383a6abca800