Summary:
- Add support for private lists of retired objects
- Add an option for one domain
- More scalable thread cache managemnt
- hazptr_rec alignment
- FOLLY_ALWAYS_INLINE
- Refactor management of retired objects in hazptr_domain
- Refactor benchmarks

Reviewed By: davidtgoldblatt

Differential Revision: D5322646

fbshipit-source-id: 9d31582b9a8216861e7e78e2e1cd08dc11cf7f88

Summary: So we can use writeFile in projects which compile with -fno-rtti

Reviewed By: mzlee

Differential Revision: D5341469

fbshipit-source-id: bb77baf9576604ff862d54d59bf382862bd0dd58

Summary:
writeFileAtomic variants all have a mode, lets just hoist the mode parameter
up and use what was hardcoded as the default.

Reviewed By: Orvid

Differential Revision: D5341481

fbshipit-source-id: 6976915dd73d2382b42bd991782730601b918978

Summary: There's no reason these utilities should only be used by folly.

Reviewed By: mzlee

Differential Revision: D5317894

fbshipit-source-id: 5a9bdf4c5efaa5bcbe78e6723a03a468f2fe5e32

Summary:
Added getter and setter methods for the RSA struct in OpenSSL. This is
needed for compatibility between OpenSSl 1.1.0 and other versions.

Reviewed By: yfeldblum

Differential Revision: D5331948

fbshipit-source-id: ab52ffd38bb5e0bd59e058bcbc6ec6122839844e

Summary:
GCC 7 has issues with `this->` in trailing return type declarations. This takes the approach of eliminating the trailing return types entirely.

Closes https://github.com/facebook/folly/pull/621

Reviewed By: yfeldblum

Differential Revision: D5324763

fbshipit-source-id: e38ae76c13ff60bc227146c1ec25afbac8b6982f

Summary:
[Folly] Fixes for `Try::withException`.

* Fix the overload with the exception type specified so that the specified exception type is checked. A mismatch between the specified exception type and the parameter type of the passed invokable will result in the expected compiler error.
* Add an overload where the exception type is deduced intentionally, as opposed to accidentally.

Of course, these simply forward to their corresponding `exception_wrapper::withException` overloads.

Reviewed By: ericniebler

Differential Revision: D5216691

fbshipit-source-id: 7e85c906c3aa17dfede6e0980c6ac8bf75034073

Summary: As title.

Reviewed By: WillerZ, yfeldblum, ivmaykov

Differential Revision: D5312308

fbshipit-source-id: 91ad4bb39df0619f2885e13c1fda1e018292a40e

Summary:
Add attachEventBase() and detachEventBase() methods to AsyncSignalHandler,
similar to the methods of AsyncSocket.

The main benefit of this is that it allows creating an AsyncSignalHandler with
an initially null EventBase, and then attaching it to an EventBase at some
later point in time.

Reviewed By: yfeldblum

Differential Revision: D5315325

fbshipit-source-id: 8a4ca483a62ca86837ea0bb54fa9a70d59f2f04e

Summary:
Clang's UndefinedBehaviorSanitizer flagged an issue that pushAtMost() could
call `memcpy(dest, nullptr, 0)` if the input was an empty `ByteRange`.  A
default constructed `ByteRange` (or `StringPiece`) will be empty and both its
begin and end pointers will be null.  Unfortunately it is undefined behavior to
call `memcpy()` with a null source pointer even if the length is 0.

This updates the `Cursor` and `Appender` code to avoid calling `memcpy()` at
all when the input length is 0.

Reviewed By: yfeldblum

Differential Revision: D5322917

fbshipit-source-id: 67fce9579f97e7e93a5767b11cc5e43ff7be5876

Summary:
`FormatValue` holds non-integral objects by reference, which
can cause subtle problems if a formatter is constructed with temporary
arguments and used beyond the expression that constructs it.

With this diff the arguments are perfectly forwarded into a tuple, so
the formatter will take ownership of the temporaries while holding
references to lvalues as before.

The only downside is that now `FormatValue` objects are constructed
every time the formatter is used, rather than only at formatter
construction. This should not be noticeable as those objects'
constructors should just take a reference to the argument.

Note that the format string is still held by reference, but this is
fine because it should almost always be a string literal.

Reviewed By: ericniebler

Differential Revision: D5317382

fbshipit-source-id: ef8355194b634d3751ef1ccca32dd1db29e27c48

Summary: Previously, only marking it as non-blocking was supported.

Reviewed By: simpkins

Differential Revision: D5307769

fbshipit-source-id: 03ed7c35632dbc9551552b21c401ea709bd6eaaa

Summary: Edited the example function for withWLock

Reviewed By: simpkins

Differential Revision: D5322000

fbshipit-source-id: 6d95b173087dec33cfdc0d0cee9c605cdf4e2593

Summary:
[Folly] Let `make_exception_wrapper` construct on-heap objects in-place.

Currently, it constructs on-heap objects on the stack and then move-constructs them into on-heap storage.

Reviewed By: ericniebler

Differential Revision: D5315104

fbshipit-source-id: cc0493e7d98aacadd342eb56601028ced4a19bb3

Fix SharedPromise::isFulfilled doesn't get back correct value when it's running in a different thread of SharedPromise::setValue

Summary: The implementation of `SharedPromise::isFulfilled` return `hasValue_` directly without taking a lock. `isFulfilled` can be accessed concurrently, but `hasValue_` is not synchronized. Adding a lock fix the issue.

Reviewed By: yfeldblum

Differential Revision: D5319030

fbshipit-source-id: a94b12ed277aa64254680ae84cb18946226cceea

Summary: There was a circular dependency between the `detail/ThreadLocalDetail.(cpp|h)` files and `ThreadLocal.(cpp|h)`, this resolves that circular dependency by moving a couple of declarations into the detail file.

Reviewed By: yfeldblum

Differential Revision: D5318875

fbshipit-source-id: 61de7c4b3ffa30f9eb11bd97e0493e9de12c5804

Summary: This reverts commit f16634ce92ee6c299613d3db22f0537bd31ac14d

Differential Revision: D5312574

fbshipit-source-id: 1557ac17f79fb8e3530a09d69fa978ddd38e5093

Summary:
`toDelim` with a single parameter of the same type as the target type was
previously copy-constructing its result. It will now construct it with perfect
forwarding which is more efficient if the input is a temporary.

This brings `toDelim` into line with the similar implementations of `to` and
`tryTo`

Reviewed By: yfeldblum

Differential Revision: D5301427

fbshipit-source-id: 843a2d93384de88cce42f26da6562a1a6ed0dc9c

Summary: In the parameter declarations for the `gettimeofday` portability header implementation. More details on exactly why can be found in the newly added comment on the function.

Reviewed By: yfeldblum

Differential Revision: D5281301

fbshipit-source-id: 1b246adc7743b5470201e452c008418429f7f142

Summary: It's an example, not part of the main Folly library.

Reviewed By: yfeldblum

Differential Revision: D5304985

fbshipit-source-id: 7cfb95da5795af61e9307882f71836ef7fa5a463

Summary: They are intended to be run as separate sets of tests, and the logging tests define multiple mains.

Reviewed By: yfeldblum

Differential Revision: D5307503

fbshipit-source-id: ed38a762177cec3c600791b861584dd47019b5a9

Summary:
Resolve two issues with folly's clock_gettime() shim on Apple platforms (iOS, macOS, etc.):

When detecting whether folly should define clock_gettime() on Apple platforms, don't assume clock_gettime() is declared in Apple's headers just because MAC_OS_X_VERSION_10_12/__IPHONE_10_0 is defined. Someone might use a newer macOS or iOS SDK (which defines MAC_OS_X_VERSION_10_12/__IPHONE_10_0), but not have clock_gettime() (because _DARWIN_FEATURE_CLOCK_GETTIME isn't set).

When defining a shim for clock_gettime(), don't rely on Apple's clock_gettime() declaration. Instead, shadow Apple's declaration with our own. This prevents folly from inheriting Apple's availability and visibility attributes for clock_gettime().

Reviewed By: yfeldblum

Differential Revision: D5312574

fbshipit-source-id: f16634ce92ee6c299613d3db22f0537bd31ac14d

Summary: So that it is available on other transports.

Reviewed By: Orvid

Differential Revision: D5302039

fbshipit-source-id: cbfdadd158061ed9a2b3ed3e0960ce66f0d545fd

Summary: Fix UBSAN failure in atomic_shared_ptr_test by using reinterpret_cast instead of static_cast.

Reviewed By: yfeldblum

Differential Revision: D5314116

fbshipit-source-id: 8479c9c4bdeeb155067a3de75efe4e97c36519e5

Summary: Same as `boost::noncopyable` but it does not disable move constructor/assignment.

Reviewed By: luciang

Differential Revision: D5311043

fbshipit-source-id: 44fe95712169b95a00e474385be43fa857cfd8ec

Summary: `long` is only 32-bits on Windows, which means this call to `std::min` was ambiguous, which caused a compile error.

Reviewed By: simpkins

Differential Revision: D5305964

fbshipit-source-id: 7e4152006e163a0301652be46318f0fc937e5ead

Summary:
exception_wrapper's non-allocating variant was depending
on the exception reference being a reference to a heap object,
which is not the case on Windows. All thrown exceptions are
copied to the stack first.

So, we abuse internals of exception_ptr to get the internally
heap allocated copy of the object. What could possibly go wrong.

Reviewed By: yfeldblum, ericniebler, Orvid

Differential Revision: D5270634

fbshipit-source-id: 2983a6af9a2c3cb66cbee8a382707a76b582c489

Summary: By moving the input shared_ptr instead of copying it, we avoid doing an unnecessary ref count increment + decrement.

Reviewed By: djwatson, yfeldblum

Differential Revision: D5298467

fbshipit-source-id: b9f0b6999278609417bb4dc062030ca2388ba20a

Summary:
It was missing a portability header needed for `setenv`, and was assuming that the `system_clock` was represented as nanoseconds.
On Windows `std::chrono::system_clock` is represented as 100s of nanoseconds, ie. `FILETIME`.

Reviewed By: simpkins

Differential Revision: D5308182

fbshipit-source-id: facb922ebe1587723358b38499dadac683e3517b

Summary: MSVC complains about `<:` being a digraph and that it isn't interpreting it as one (C4628). Make it be quiet by making it not look like a digraph.

Reviewed By: simpkins

Differential Revision: D5305162

fbshipit-source-id: 7385164e8f73226ce18bc04c32894c2f27f8ae25

Summary: These were using `boost::filesystem::path.native()` in contexts where an `std::string` was the expected output. `boost::filesystem::path.native()` returns a `std::wstring` on Windows, so use `.string()` instead.

Reviewed By: simpkins

Differential Revision: D5305649

fbshipit-source-id: 3c7cd53d2f33d2dbb222731655cdab69d1c4a187

Summary: I failed to realize that this helper function also needs to be guarded on `!_WIN32`, because the assertions are false on Windows.

Reviewed By: ot, luciang

Differential Revision: D5304914

fbshipit-source-id: f2564cb2fcbfdc73a2529d4a41df2995748bc3f6

Summary: This enables auto-deps support in more places in Folly. It also fixes a couple of auto-deps declarations that were missing an `s`.

Reviewed By: yfeldblum

Differential Revision: D5289542

fbshipit-source-id: 7d34b29bcc172fb31f91f1ba880864e2d5869ffd

Summary:
Add appropriate specializations for detail::Extract() so that you can pass a
plain function pointer to `onError()`.  Previously the code only worked with
member function pointers and functor-style objects.

Reviewed By: yfeldblum, wez

Differential Revision: D5286773

fbshipit-source-id: 67b44d1d7573eb1da501475045fd24ad4ab1c074

Summary:
Update the FB_LOG() and XLOG() macros so that FATAL log messages are correctly
detected as not returning, by both clang and gcc.

We have to ensure that both sides of the log statement check (log message
enabled or disabled) evaluate to `[[noreturn]]` expressions.  I did try
updating the log check itself so that it could be constexpr detected as always
passing, but this was not sufficient.

Reviewed By: wez

Differential Revision: D5290780

fbshipit-source-id: 773a56a8392dfd7c310d5d84fc9311e66edf99cb

Summary:
Currently when calling connect with AsyncSSLSocket in unencrypted
mode it still attempts to perform a SSL handshake. Add check to not
do SSL. Calling sslConnect will always run the SSL handshake

Reviewed By: jrahman

Differential Revision: D5153456

fbshipit-source-id: 4d9164115be72c8ee76e383535561e3083a327e3

Summary:
This test has run into occasional failures on continuous build test runs.
Unfortunately when something goes wrong it crashes in the std::thread
destructor due to this thread still being joinable when it is destroyed, which
hides information about what actually failed in the test.

This updates the test to immediately detach the thread, so that on error we
will be able see the real failure reason.

This also increases the size of the message that we write, which will hopefully
help ensure that this write always blocks.

Reviewed By: wez

Differential Revision: D5295574

fbshipit-source-id: ea8cfa855613398f88f9f982c600ec661018a31c

Summary:
Switch all code in the logging library from using `ERROR` to `ERR`,
and remove the `ERROR` LogLevel entirely, even if it is not already
defined as a macro.

Previously the code kept `ERROR` available as a LogLevel name if it had
not already been defined as a macro (which is done by common Windows header
files).  However, this made for inconsistent behavior, and made it easy to
write code that would not be portable to Windows.

This diff fully drops the `ERROR` name for consistency across platforms.

Reviewed By: wez

Differential Revision: D5288600

fbshipit-source-id: 8d2d52e955959c278345fc9c2086c7cacf9660f9

Summary:
Add some tests for fatal log messages during static initialization and
destruction.  Ideally most programs won't actually do much real work before or
after main, but make sure the logging framework behaves sanely in these
situations.

This also fixes a bug in the `test_none()` test, which previously caused it to
always succeed without testing the desired behavior.

Reviewed By: wez

Differential Revision: D5281718

fbshipit-source-id: ca838c098886e99418264acf9a4d651ea9e7502c

Summary:
- Support empty hazptr_holder, move constructor and assignment operator
- Limit thread caching to the default domain to improve performance of thread caching
- Fix unnecessary calls to stats singleton
- Use the mprotect version of AsymmetricMemoryBarrier for reducing the overhead of bulkReclaim().
- Update read-side benchmark results

Reviewed By: djwatson

Differential Revision: D5292885

fbshipit-source-id: bc5713ac95492a7114e1e467e71d2278e64b165d