Reviewed By: Gownta

Differential Revision: D33476592

fbshipit-source-id: 9e66fdf07ce88c507cfb32406b113e1c05131c8c

Summary: This opportunity was found with the combination of Infer's PULSE_UNNECESSARY_COPY and Strobelight cycle counts.

Reviewed By: Gownta

Differential Revision: D33304127

fbshipit-source-id: 7d3b54d26bc3e171cc7a0a831933fdbb9638f6a8

Summary: sizeof(SharedMutex) = 4, sizeof(std::mutex) = 40. This diff makes sure that we use smaller mutexes both in Observer Core and in Observables given that we may end up with many instances of these objects, yet there mutexes are accessed only on the update path.

Reviewed By: praihan

Differential Revision: D32271615

fbshipit-source-id: ba7d04632a677a9eb4ae398d681dadac3b06f033

Reviewed By: yfeldblum

Differential Revision: D33118848

fbshipit-source-id: c69b1f8f415b130d5d7a6487e7027d5c1e49865f

Summary: Add comments stating the expectations for Baton post and Futex wake native implementation to be async-signal-safe.

Reviewed By: ot, luciang

Differential Revision: D33461663

fbshipit-source-id: 2855427cc94f577b566b19203a4523f54a41d0aa

Summary:
A number of places were extracting dependencies from manifests, but only one was adding in the implicit dependencies for build tools.

Extract the logic to one place and use so that a change in a tool like cmake will now correctly affect all tools using cmake, as it will be taken into account as a dependency hash when the manifest's hash is computed.

Tests for this change revealed that install_dirs needed to be populated in reverse order from the manifest topo-sort, so have also addressed that

Reviewed By: wittgenst

Differential Revision: D32730717

fbshipit-source-id: 1b2a25e460de6085d274c99acfd391b3bd259264

Summary: We want to have a custom `IOExecutorThreadPool` that performs different logic for handling observers being adding or not. This change makes it so the functions and structs necessary to do so can be properly overriden.

Reviewed By: Gownta

Differential Revision: D32101852

fbshipit-source-id: 2f1c6bb4c00d41630891aa72b8e519708cb1a40f

Summary: When we build cmake on linux the openssl headers are needed

Reviewed By: fanzeyi

Differential Revision: D33432903

fbshipit-source-id: 2f869768f3a81b74e53391f5491781759c66b478

Summary:
If we already have pending application reads or writes when we have a
handshake error (eg when using a null connect callback), we should pass the
handshake error to finishFail to give a more helpful error message.

Reviewed By: mingtaoy

Differential Revision: D33415008

fbshipit-source-id: fc14ef774d09c0af975e177b24185f78c4e4139e

Summary: asm memset implementation based on the https://github.com/nadavrot/memset_benchmark implementation by Nadav Rotem

Reviewed By: luciang, magedm

Differential Revision: D32888087

fbshipit-source-id: 3678950ed5178d2a46ba833502d1a6aef4f2c731

Reviewed By: yfeldblum

Differential Revision: D32187345

fbshipit-source-id: d384aac90c5c0c10580824305ce108e7c0b0e577

Summary:
This avoids the following errors:

  folly/executors/ManualExecutor.h:42:3: error: '~ManualExecutor' overrides a destructor but is not marked 'override' [-Werror,-Wsuggest-destructor-override]

Reviewed By: r-barnes

Differential Revision: D33397861

fbshipit-source-id: f9bee89aeaa55d365155dc3fdc0b70b283980a2a

Summary:
This avoids the following errors:

  folly/io/async/ssl/OpenSSLTransportCertificate.h:31:11: error: '~OpenSSLTransportCertificate' overrides a destructor but is not marked 'override' [-Werror,-Wsuggest-destructor-override]

Reviewed By: r-barnes

Differential Revision: D33397487

fbshipit-source-id: 4237f5ad36166189b0bbd28dfcf268e0ef4b40f1

Summary:
This avoids the following errors:

  folly/experimental/coro/BlockingWait.h:301:3: error: '~BlockingWaitExecutor' overrides a destructor but is not marked 'override' [-Werror,-Wsuggest-destructor-override]

Reviewed By: r-barnes

Differential Revision: D33397860

fbshipit-source-id: 050c3c564e70dfc5dc1232e58b40f2c4cbcb660a

Summary:
This avoids the following errors:

  folly/io/async/ScopedEventBaseThread.h:49:3: error: '~ScopedEventBaseThread' overrides a destructor but is not marked 'override' [-Werror,-Wsuggest-destructor-override]

Reviewed By: ryanthomasjohnson

Differential Revision: D33397489

fbshipit-source-id: 29210c0de0b850bdda5d2c27a5be8971fc17b40e

Summary:
Update the README.md to reflect that getdeps.py build and test is the method tested in CI and thus preferred.

Where getdeps.py manifests encode dependencies I've removed the manually specified dependences from the README for consistency (e.g. googletest version in README was outdated, manifest is correct and is tested in CI)

Also referenced main vs master and improved markdown header formatting a little

Reviewed By: meyering

Differential Revision: D33295900

fbshipit-source-id: caec3b3795be0b37bf1efdf12f4dbf23a37a023c

Summary:
There is no current use case of this API throughout the codebase
and it's causing some confusions of how this class is used
so removing it for now. It's easy to add it back if we need it
in the future

Reviewed By: yfeldblum

Differential Revision: D33112001

fbshipit-source-id: c711f5ab0ad2324e478f6eb18155d367d2cce1f2

Summary: Looks like bad copy/pasta.

Differential Revision: D33279601

fbshipit-source-id: 24d3ad02422ac5af6a1624f09c719207ed4950a6

Reviewed By: aaronabramov

Differential Revision: D33358291

fbshipit-source-id: d22ff22b8c84ef7d60bafb41dd98acacea8f5ddb

Summary:
Adding methods to `folly::OpenSSLCertUtils` that allow to read extensions from X509 cert structures.
- `folly::OpenSSLCertUtils::getExtension()` allows to query extensions by name
- `folly::OpenSSLCertUtils::getAllExtensions()` returns everything including custom extensions.
No extension name translation mechanism is provided for custom extensions.

Reviewed By: mingtaoy

Differential Revision: D31852457

fbshipit-source-id: 22d34d2cf304ba135419c902d9c5c303f78cf781

Reviewed By: bhamodi

Differential Revision: D33330724

fbshipit-source-id: 8a798435742dedc96e2b6912179736b6a1c72491

Summary:
This test regularly times out under load.

I made a benchmark diff (prior to this in the stack, but not to be landed) which confirms (a) that the slowness of this test is due to spawning excessively-many threads, and (b) that the performance of ConcurrentSkipList doesn't unduly degrade with excessively-many threads. Conclusion: it is safe to simplify these tests.

To simplify the test, I ratcheted down the number of threads, but also bumped the number of iterations. Much more work is being done (so this test should retain its race-condition-hunting properties), but the time spent is much less, even when the system is under load.

Reviewed By: yfeldblum

Differential Revision: D33269899

fbshipit-source-id: 9f7a0ec5f3c1077c91e5d09c89f3752d552a6320

Summary: This test regularly fails on account of timing. Remove the timing upper bound.

Reviewed By: yfeldblum

Differential Revision: D33271470

fbshipit-source-id: 1369ef45bd49e500c35816bfdb21869f346b6f70

Summary: swap should be noexcept when possible

Reviewed By: yfeldblum

Differential Revision: D33198827

fbshipit-source-id: f8d52a21087b0346332359b66e1dc00f982ddbbb

Summary:
Externally this was failing on fedora 35's GCC 11.2.1

Add test so we know what platforms it's good on

Reviewed By: yfeldblum

Differential Revision: D33192636

fbshipit-source-id: 028255b6dbb34762b4e788122d948f26a449a57c

Summary: Try enabling this, need to handle linux ASM build case in cmake config

Reviewed By: Gownta

Differential Revision: D33191710

fbshipit-source-id: 6d49a08106c8a0044380f592fd2862080cbea454

Summary:
Having the tests is useful to be able to test the OSS builds. A few were failing for me locally so I've tagged them BROKEN in CMakeLists.txt which folly's cmake config filters out

Pull Request resolved: https://github.com/facebook/folly/pull/1691

Test Plan:
tested with:

```
./build/fbcode_builder/getdeps.py --allow-system-packages build
./build/fbcode_builder/getdeps.py --allow-system-packages test
...
100% tests passed, 0 tests failed out of 2736

Total Test time (real) =  14.95 sec
````

Reviewed By: yfeldblum, Gownta

Differential Revision: D33169409

Pulled By: ahornby

fbshipit-source-id: 9c781a84b8873c295af96368dd8315254a78f096

Summary:
Ensure that folly continues to work with platform010. Since folly is depended on my many projects, some on platform009 and some on platform010, we should make sure it continues to work with both platforms.

Since many platform010 projects depend on folly, folly's platform010 coverage is actually pretty good. However, changes to test files in folly have been observed to break with 010-incompatible changes (such as D33143441 (https://github.com/facebook/folly/commit/43ba86e231ab5cb158835433df656d3026c3c6cf) and D32930335 (https://github.com/facebook/folly/commit/cec3fcac4dded35478747e4d9b8e533b491140e1)); these are not caught by dependent projects' configs.

Reviewed By: yfeldblum, meyering

Differential Revision: D33172025

fbshipit-source-id: 94b46471bee604289b3b6a11763a27c6e3ed9fcf

Summary: sosorry

Differential Revision: D33218659

fbshipit-source-id: e7c77b503ef6c7144fd8da684b22f793d85b2b96

Summary: `ResultOf` is no longer the most-suitable name. This helper has long been a SFINAE helper and its actual output did not matter, but now that it is always `void` the name is even less clear. Rename it.

Reviewed By: luciang

Differential Revision: D33155200

fbshipit-source-id: a12842c6be38bd82bfffa85193ec265bc4d803ad

Summary:
C++14 introduces sized overloads of `operator delete` and C++17 introduces aligned overloads of `operator new` and `operator delete`. Offer wrappers which call the best available versions.

Using these overloads, asking for heap storage suitable for a non-over-aligned class can be done via these wrappers without requiring call-sites to test feature macros.

Most potential use-cases should use `std::allocator<T>::allocate` and `std::allocator<T>::deallocate`. Reasons why use-cases may turn to these wrappers instead are:
* These wrappers are not temnplates and using them does not instantiate class or function templates.
* `std::allocator<T>::deallocate` may, but is not specified to, use sized deallocation when sized deallocation is available.

Reviewed By: luciang

Differential Revision: D33180956

fbshipit-source-id: 944fe6a01ca4cfff72f3b92639158e4a5ac029b8

Summary: It is possible for return type names to be long. They are present in the ctor symbol names for SFINAE, but we can get around that by inserting a cast-to-void.

Reviewed By: luciang

Differential Revision: D32976770

fbshipit-source-id: d5a972543f4d97f94abea9631a4bed35fa67f8f3

Summary: Appears to be measurably faster to compile under clang as compared to `std::decay_t`.

Reviewed By: Alfus, luciang

Differential Revision: D32946561

fbshipit-source-id: 028c79633095ae9d6dc75d967282b53b32986736

Summary:
The channels framework provides most of its memory savings by allowing channels to be used without long-lived coroutine frames consuming them. However, coroutines are still used for short-lived transformation functions. While these coroutines should not increase steady state memory (as they are short-lived), it is possible that a large number of transformations occur at exactly the same time. This can cause a large memory spike (due to the size of the coroutine frames), and lead to heap fragmentation.

This diff solves this problem by adding support for rate limiting. An optional RateLimiter object can be provided to any transform or resumableTransform. A RateLimiter has a maximum concurrency specified on construction. For any transforms using the same rate limiter, the channels framework will ensure that the concurrency constraint is not violated. This limits the number of simultaneous coroutine frames (and therefore allows one to eliminate these spikes).

Reviewed By: aary

Differential Revision: D33037310

fbshipit-source-id: f7c5d30e08d155db7825d01f1b4c4b7354b14def

Summary:
Currently, resumableTransform takes a paremeter-less initializeTransform method for initialization, and re-initializes the transform when the transformValue function throws an OnClosedException. This requires users to capture any initialization parameters in the initializeTransform lambda, so they can re-initialize with the same parameters when needed.

However, it is often the case that the initialization arguments are stored elsewhere in the app, and passed along as part of each message coming through input receiver to the transform. (This is the case in all uses of resumableTransform so far.) If we could take advantage of already having the initialization arguments available, we would not need to waste memory capturing and storing them in the initializeTransform lambda.

This diff does exactly that. Instead of triggering a reinitialization by throwing an OnCloseException, reinitialization is triggered by throwing a ReinitializeTransformException<InitializationArg>. This initialization argument is then passed to initializeTransform. By doing this, we no longer need to store initialization arguments in resumable transforms.

(As a side benefit, this also makes the code slightly clearer to the reader. Rather than the reader having to know that throwing an OnClosedException triggers a reinitialization, throwing a ReinitializeTransform exception makes it clear.)

Reviewed By: aary

Differential Revision: D33037191

fbshipit-source-id: 891394d79d3fc43a8e253031472256bb852c7716

Summary: This diff adds a new function to FanoutChannel that closes all existing subscribers, without closing the FanoutChannel itself.

Reviewed By: aary

Differential Revision: D33035825

fbshipit-source-id: 6aa867344c280ad7a91bd8248e745aa2f27e7cc8

Summary:
This diff reduces the amount of memory that transforms use. Currently, transform explicitly takes an executor and a transform lambda (and resumableTransform takes an additional initialization lambda). Often, these lambdas end up taking pointers to some shared state object. In the case of resumableTransform, both lambdas need to capture a pointer to the same shared state object. Furthermore, the executor passed to transform is often present on the shared state object, making it redundant. This problem will get even worse when we add rate limiters (which also can be placed on the shared state object).

To reduce memory waste, this diff adds a new overload for transform and resumableTransform that takes in a Transformer object. This transformer object must implement getExecutor, transformValue, and (for resumableTransform) initializeValue. This allows a transformer object to contain just one 8-byte pointer to a shared state object, rather than duplicating pointers for different lambda captures and executors (and eventually rate limiters). It also slightly simplifies code in certain cases, as the shared state can be accessed as a field from any function in the class (rather than captured and passed down as a parameter to all helper functions).

We will still leave the existing overloads for cases where it is simpler for the user to not create a new object (when the user doesn't care as much about the memory savings).

Reviewed By: aary

Differential Revision: D33033365

fbshipit-source-id: e57c16e76ad3795a7f93b39565f553e76623beb6

Summary: This diff changes Transform's parameter names to conform to folly's style guidelines.

Reviewed By: Gownta

Differential Revision: D33033366

fbshipit-source-id: 3713ec7995df3dd913f4da11d21fbd9377c1f00f

Summary:
The current MergeChannel public interface has a few problems that have become more apparent with usage.

1. The output receiver does not contain the key associated with the input receiver that geneated the value. This often means that the user must transform input receivers before adding them to the merge channel to include the key, which wastes memory.
2. There is no way for a consumer of the output receiver to know if an input receiver was closed without an exception.
3. If an input receiver was closed with an exception, the entire merge channel is closed.

This diff changes the public API to fix these problems. Instead of returning an output receiver that just has values, the output receiver contains MergeChannelEvent objects. Each such event object has the key of the corresponding input receiver associated with the event, along with the event contents. There are four types of events:

1. A value was received through an existing input receiver.
2. A new input receiver was added.
3. An existing input receiver was removed.
4. An existing input receiver was closed (with or without an exception).

Note that we could theoretically avoid genering events for 2 and 3 (as they correspond with calls that the user makes to add and remove), but there are cases where it is useful to have 2 and 3. This way, an event is always received when an existing input receiver is removed, regardless of whether it was removed explicitly (through a call to removeReceiver) or implicitly (if the input receiver was closed).

Reviewed By: aary

Differential Revision: D33033374

fbshipit-source-id: 34f74cba3f765c76a80d448647d733ad38a2195c

Summary: This diff changes MergeChannel's template parameter names to adhere to folly's style guidelines. It also reverses the order of the template parameters, as it conceptually makes more sense when the key parameter precedes the value parameter.

Reviewed By: Gownta

Differential Revision: D33033369

fbshipit-source-id: 0b31d18baa26778ba9e8d17b9419cb180b1939d9