Summary:
This took a lot of cmake fuckery. I am guessing it will break the current travis-ci.org based build, since it no longer downloads gflags.

As I put this diff up, mvfst (a dependency of quic (a dependency of proxygen)) is not building right now due to another diff - a fix is in progress but isn't landed yet. I applied the fix locally for the test plan, but didn't include it in the diff.

Reviewed By: snarkmaster

Differential Revision: D25651851

fbshipit-source-id: b4f15ac7fa5293123710fe367c20d355c0970c42

Summary: `folly::MicroLock` is actually 4 locks, each using 2 bits. In cases we don't need all the slots, we can use the other bits to store some data which is loaded/stored atomically with locking and unlocking. There is no change in existing `MicroLock` usage behavior.

Reviewed By: davidtgoldblatt

Differential Revision: D25842148

fbshipit-source-id: 5f6ff705c35529d279629ccc256a9c210907ea5c

Summary: This was forgotten in a refactor.

Reviewed By: Mizuchi

Differential Revision: D25903164

fbshipit-source-id: 419131987a886f0d8d057b5b04b012d1162e564d

Summary:
`extra_cmake_defines` are extra flags that are passed to cmake when
compiling each one of the dependencies. For instance:
```
$ ./opensource/fbcode_builder/getdeps.py  build f4d --extra-cmake-define='{"CMAKE_CXX_FLAGS": "-mavx2 -mfma -mavx -mf16c -march=native"}'
```
It takes a json map as input, which can take a list of defines (key value
pairs).

Reviewed By: wez

Differential Revision: D25855781

fbshipit-source-id: 7f4fef2c66f4d12f23c8d7086d6a4f24fcc01ff7

Summary:
This patch performs SFINAE-friendly invocation in `FOLLY_CREATE_STATIC_MEMBER_INVOKER` in order for invocability tests to return `false` in the situation where the member is not even available. For example, given:
```
FOLLY_CREATE_STATIC_MEMBER_INVOKER(foo_invoker, foo);
```
With this diff, something like `std::is_invocable_v<foo_invoker, HasNoFoo>` yields `false` rather than a compilation error.
This brings `FOLLY_CREATE_STATIC_MEMBER_INVOKER` in line with `FOLLY_CREATE_FREE_INVOKER` and `FOLLY_CREATE_MEMBER_INVOKER`.

Reviewed By: yfeldblum

Differential Revision: D25878262

fbshipit-source-id: 150ae6b2c71941317bc02de20720fed014639bf5

Summary: The lazy initialization logic can now be simplified.

Reviewed By: andriigrynenko

Differential Revision: D25675031

fbshipit-source-id: 048328220a2b02c13a672bebe47b79521d5f5e7a

Summary: `folly::DelayedInit` serves a similar use case to `folly::Lazy` but is suitable for data members and is thread-safe. The summary in `DelayedInit.h` provides a good overview.

Reviewed By: andriigrynenko

Differential Revision: D25671587

fbshipit-source-id: 1933b15fdc7207a52067658188ca4d5724d9c40b

Summary: Migrate a couple of uses of Folly Format to fmt.

Reviewed By: simpkins

Differential Revision: D20152350

fbshipit-source-id: fb9798f5ff458ee03e1c01f839dc1c55796d0eaa

Summary:
This utility is meant for usage cases where
- the same `mallctl` command is called many times, and
- performance is important.

Reviewed By: yfeldblum

Differential Revision: D23638723

fbshipit-source-id: 6cab9acb1121ef4f46bf7acbe29d3877005b66b1

Summary: All changes were automated

Reviewed By: Orvid

Differential Revision: D25507587

fbshipit-source-id: fe757e6919793e9b8b28d6c1ce6f064d7e2623fb

Summary:
See docblock in new test. If the FiberManager weakref is created after the AsyncioExecutor weakref, then the native FiberManager object may be destroyed without being drained.

Unfortunately, there's no way to enforce an ordering between the destructors here (since weakrefs are cleaned up in inverse order of creation).

This diff adds a `__dealloc__` method to AsyncioExecutor to drain the native FiberManager

Reviewed By: yfeldblum

Differential Revision: D25797920

fbshipit-source-id: d1a548ebd67e7d9153eb43b169bf009a6a986e68

Summary: Update name to match usage of Try as tri-state, since this method also throws if the Try is empty

Reviewed By: yfeldblum

Differential Revision: D25737810

fbshipit-source-id: a4166153362f07353d212216fbaf7105867eef2a

Summary: Tweak `is_coroutine_handle` detail trait. No need for leading `_` and let it be a trait variable.

Reviewed By: iahs

Differential Revision: D25864424

fbshipit-source-id: 2f0723fcb3c97c6a0972056999f1d9574695c30b

Summary:
1. Patch the test to remove racy behavior
2. Added another test for a case where the timeout is hit.

Reviewed By: yfeldblum

Differential Revision: D25854609

fbshipit-source-id: c1694e5b46b9645dc3e71a12ddc8ed12e8cf270c

Summary: Fix typo

Reviewed By: yfeldblum

Differential Revision: D25861463

fbshipit-source-id: c24a521572229cbc5892a4da10999dcff94741c7

Reviewed By: igorsugak

Differential Revision: D25849195

fbshipit-source-id: 611df723a38b5c2a7ba5184c63fdeeed1ed27025

Summary: Adds a test for ConcurrentHashMap::reserve().

Reviewed By: magedm

Differential Revision: D25853643

fbshipit-source-id: 39404105c88d6dc38cdd0b399c646affad1b26e0

Summary:
Fixes a bug when a call to reserve leads to calling rehash with a buck_count value that is not a power of two and/or less than the current count.
The rehash algorithm depends on the bucket_count being a power of two and that the count does not decrease.

The fix: Ensure that reserve calls rehash with a power of two argument and that rehash is skipped if the new count is not greater than the old one.

Reviewed By: davidtgoldblatt

Differential Revision: D25852794

fbshipit-source-id: ae1ae6cfbcbd0659becc14c703920e9505181109

Summary: All changes were automated

Reviewed By: Orvid

Differential Revision: D25507562

fbshipit-source-id: ffed8f0e41d3581242b4e8e00bfd1fa6ee3e299b

Summary: Sets out the foundation of io_uring support into fizz and enables support in the fizz tool

Reviewed By: mingtaoy

Differential Revision: D23592810

fbshipit-source-id: 6f995c99541bc2aeca3cdfb7d84ebd2d25eb1f5e

Summary: Same code as in oneIn() explicit check for arg 0, 1.

Reviewed By: yfeldblum

Differential Revision: D25838957

fbshipit-source-id: 36ead588e435f116d0870ba45a181979c591c3f2

Reviewed By: yfeldblum

Differential Revision: D25689803

fbshipit-source-id: 0d5f6093d814c58e2d90f9042b7bad65ff8d8ade

Summary: Let `folly::FunctionRef` detect empty-callable ctor arguments like `folly::Function` does. Handles pointers which may be nullptr.

Reviewed By: luciang

Differential Revision: D25840538

fbshipit-source-id: 8cfccd57aa545ffc95807876d486ecf94aba8d62

Summary: `BlockingWaitExecutor::add` (and similarly `WaitExecutor`) posts the baton while holding the queue lock. If these happen in two different threads, the `drive()` thread wakes up while the lock is still held; the `add()` thread may be descheduled before the lock is released, causing the other thread to have to wait (and possibly sleep) for a while.

Reviewed By: andriigrynenko

Differential Revision: D25814620

fbshipit-source-id: 07d3e9afdc4647e68f996e3f0a1c007fd1596611

Summary: Pull Request resolved: https://github.com/facebook/folly/pull/1459

Reviewed By: yfeldblum

Differential Revision: D25730366

Pulled By: Orvid

fbshipit-source-id: d0fe94b1db6d81c0dabaa65de1483ea319669776

Summary:
Update AsyncSignalHandler to check if the events are still registered before
trying to unregister them during destruction.

Without this check the code crashes if an AsyncSignalHandler is registered,
and then the EventBase is destroyed before the AsyncSignalHandler is
destroyed.  When the EventBase is destroyed the AsyncSignalHandler's event
will be automatically unregistered, so `eb_event_del()` should not be called
again here, and calling it triggers a crash by attempting to dereference the
now-destroyed EventBase.

This change makes AsyncSignalHandler behave more like the AsyncTimeout and
EventHandler objects, which both support destroying the EventBase before these
objects.

Differential Revision: D25837541

fbshipit-source-id: a1c44ab442b9078bf84a51be2cced432d247f6cd

Summary: Restrict `FOLLY_HAVE_WEAK_SYMBOLS` by requiring not just that use of gnu weak symbols will compile but also that it will link and run when no definition is available for a weak declaration.

Reviewed By: luciang

Differential Revision: D25788660

fbshipit-source-id: 52447601257242cd713ac5cd70e900e3738a9770

Reviewed By: andriigrynenko

Differential Revision: D25814958

fbshipit-source-id: abc2448bf93c5016411629936a8c7f06bb0033e6

Summary: Add `kHasWeakSymbols` to `folly/Portability.h` so that downstream call-sites may detect availability of weak symbols at compile time in normal C++.

Reviewed By: luciang

Differential Revision: D25788659

fbshipit-source-id: 66768486317986396440099adecfc8dc8e553c7c

Summary:
The certificate identity verifier is meant to verify the end entity certificate
in a certificate chain. It assumes that OpenSSL has already validated that
the end entity has a valid signature by an issuer.

The test that I am removing is asserting buggy behavior in OpenSSL prior to
version OpenSSL 1.1.1i. The test scenario sets up a client *without a certificate
store* connecting to a server. When the client attempts to validate this
certificate chain, it will receive two errors:

1. Unable to find issuer locally. Because the test server's certificate is signed
   by an issuer that is not present in the client's (empty) certificate store,
   while trying to determine the issuer during chain building, OpenSSL fails as
   expected. Because the test scenario *intentionally* overrides this error by
   returning `true` in `handshakeVer`, OpenSSL continues its certificate
   validation procedure.

2. Unable to verify leaf signature. In order to validate the signature for a
   certificate, one needs to know the public key of the issuer. However, the
   client does not know about the issuer. Control flow would normally not reach
   here, but the client chose to continue the validation process without knowing
   the issuer in step 1. The test scenario forcefully returns "true" here, telling
   OpenSSL to continue.

Prior to OpenSSL 1.1.1i (specifically commit [2e06150e](https://github.com/openssl/openssl/commit/2e06150e3928daa06d5ff70c32bffad8088ebe58)),
OpenSSL would stop the verification procedure here. It would never signal the
completion of the verification procedure (by sending `preverify_ok=1`).

The test was asserting that this was expected. This meant that the test was
asserting behavior on an incorrect implementation. With OpenSSL 1.1.1i, if
all previous `handshakeVer` callbacks say that the certificate is fine, then
OpenSSL will perform the final call with `preverify_ok=1`, which will cause the
CertificateIdentityVerifier to be invoked.

Differential Revision: D25775428

fbshipit-source-id: 8a5f388ce2eb9a8d5667fd4e840b6b223e618d43

Summary:
Some tests were relying on internal OpenSSL behavior that changed in OpenSSL
1.1.1i.

OpenSSL commit [2e06150e](https://github.com/openssl/openssl/commit/2e06150e3928daa06d5ff70c32bffad8088ebe58)
causes additional `handshakeVer` callback invocations. If an UNABLE_TO_VERIFY_LEAF_SIGNATURE handshakeVer
callback overrides the result to *true*, OpenSSL will now continue to perform certificate
verifications **as if the overriden cert was the properly found root**. This means:

* It is subject to certificate expiration checks, and,
* A final `handshakeVer` callback is invoked with `preverifyOk=1` when it passes the expiration checks.

Several of our tests that relied on `handshakeVer` only being called *once* broke. Since the tests
are testing a different, higher level aspect of verification (i.e. it is not actually trying to
test how many times `handshakeVer` is being called), I've relaxed some of the expectations to allow
for multiple invocations.

CertificateVerifier semantics were also affected. Because there is a final
successful `handshakeVer` callback that is issued, it is not just sufficient
to check for depth and preverifyOk before invoking the CertificateVerifier.

Reviewed By: pixelb

Differential Revision: D25775429

fbshipit-source-id: 98d35249a623307fb1de366c498b1dbae4fabf82

Differential Revision: D25759768

fbshipit-source-id: c0d38b25ab74bfce82c3d5e1339bd4ceef98af14

folly: symbolizer: check if findSubProgramDieForAddress found the DW_TAG_subprogram for address & terminate DFS early

Summary:
It's legal that a CU has `DW_AT_ranges` listing all addresses, but not have a `DW_TAG_subprogram` child DIE for some of those addresses.

Check for `findSubProgramDieForAddress` success before reading potentially empty `subprogram` attributes in `eachParameterName`.

This is a bugfix. If `DW_TAG_subprogram` is missing, we read attributes for using offsets in the uninitialized DIE `subprogram`.

Initialize all struct members to avoid wasting time debugging uninitialized memory access (had fun with uninitialized `subprogram.abbr.tag` magically picking up `DW_TAG_subprogram` or garbage on the stack from previous calls).

 ---

Terminate `findSubProgramDieForAddress` DFS on match.

Before:
- when the DW_TAG_subprogram DIE was found we stopped scanning its direct siblings (the above "return false" when `pcMatch` or `rangeMatch` are true).
- but when we returned from recursion in the parent DIE we discarded the fact that we found the DW_TAG_subprogram and continued the DFS scan through all remaining DIEs.

After:
- after finding the DW_TAG_subprogram that owns that address the scan stops.

As reference when using `-fno-debug-types-section` type info is part of `.debug_info`.

```
namespace folly::symbolizer::test {
void function_A();
class SomeClass { void some_member_function() {} };
void function_B();
} // folly::symbolizer::test
```

```
0x0000004c:   DW_TAG_namespace [2] *
                DW_AT_name [DW_FORM_strp]       ( "folly")

0x00000051:     DW_TAG_namespace [2] *
                  DW_AT_name [DW_FORM_strp]     ( "symbolizer")

0x00000056:       DW_TAG_namespace [2] *
                    DW_AT_name [DW_FORM_strp]   ( "test")

0x0000012e:         DW_TAG_subprogram [7] *
                      DW_AT_name [DW_FORM_strp] ( "function_A")

0x0000022d:         DW_TAG_class_type [14] *
                      DW_AT_name [DW_FORM_strp] ( "SomeClass")

0x00000243:           DW_TAG_subprogram [16] *
                        DW_AT_name [DW_FORM_strp]       ( "some_member_function")

0x000002ed:         DW_TAG_subprogram [9] *
                      DW_AT_name [DW_FORM_strp] ( "function_B")
```

Before:
- if `address` corresponds to code inside `SomeClass::some_member_function` we stopped iterating over `SomeClass::some_member_function`'s sibling DIEs, and returned.
- but we still continued scanning over `SomeClass`'s sibling DIEs and all their children recursively (DFS), and then `SomeClass`'s parents siblings etc until reaching the end of the CU

After:
- we stop DFS on match

FWIW: trees are somewhat flatter when using `-fdebug-types-section`: member functions are extracted from their classes, but they're still part of namespace subtrees so the code unnecessarily explored siblings of the innermost namespace.

Differential Revision: D25799618

fbshipit-source-id: 9a0267e12a7d496ad00263bf30a12dc548764288

Summary:
This code implicitly casts a `long` to `double` but a `double`
can not accurately represent an integer above `2^53` as
the mantissa only stores that many digits. At `2^53 + 1`
`double`s increment by `2`. Thus this cast is lossy. Though,
casting a single value `LONG_MAX` to a `double` is and
being off by one doesn't cause any harm here.

Reviewed By: ispeters

Differential Revision: D25654006

fbshipit-source-id: 3d84092f1786d7879dd48d73af8edbbc80bb3585

folly::symbolizer: drop templated FrameArray from test functions and move test functions to separate compilation unit with minimal dependencies

Summary:
Eliminate `#include` dependencies of `SymbolizerTestUtils.cpp` so that the `.o` file with interesting debug info is tiny and easily inspectable.
- Instead of calling test functions that need to be inlined from the `SymbolizerTest.o` use `call_` trampolines so that the inlined debug info is generated in the minimal/tiny `SymbolizerTestUtils.o`.

 ---

Remove the frames template argument so that:
- removed dependencies from the SymbolizerTestUtils to minimize generated debug info (no longer need to include headers about FrameArray).
- all functions used in tests can be made regular functions
- all functions who's debug info is interesting to debug can be easily emitted in a single object file `SymbolizerTestUtils.o`

Reduce some of the copy-pasta that made test noisy by grouping together function name, file and lineno.

 ---

Test for either fullName/shortName (`DW_AT_linkage_name`/`DW_AT_name`)
- fullName: demangled function name with namespace, qualifiers, argument types, etc. and
- shortName: simple function name to support both cases when either both `DW_AT_linkage_name` and `DW_AT_name` or a single one of them is emitted.
- When `DW_AT_linkage_name` (full mangled name) and `DW_AT_name` (just the function name) match, DWARF emitters omit `DW_AT_linkage_name` (to save space). With `-fsplit-dwarf-inlining` only `DW_AT_name` is emitted in `.debug_info` and the `DW_AT_linkage_name` is emitted in the `.debug_info.dwo` sections.

Differential Revision: D25712808

fbshipit-source-id: 88d2ac22564e3d607fb3864d37e49892a5eaf002

Reviewed By: yfeldblum

Differential Revision: D25527140

fbshipit-source-id: 2a2721ca4418e1a4c57f621f41716e5c8f27869c

Summary:
Small change - adding  logs to connect error/success and eliminating the unnecessary ones.

(Note: this ignores all push blocking failures!)

Reviewed By: yfeldblum

Differential Revision: D25481273

fbshipit-source-id: 437942803bc895bba2eaa36e5eaf104503012d12

Summary:
Adding priorities to `ThreadManagerExecutorAdapter`. The API allows to pass single executor that would be used for all priorities (preserved behavior) but also allows to add custom executors per-priority. In this case, if no executor has been provided for priority, we will spin up default thread manager implementation (which supports upstream queues).

We soon will add upstream queues on top of user-provided executors as well.

Reviewed By: prshreshtha

Differential Revision: D25137444

fbshipit-source-id: db0740a705dd3cca69916f3d1b8c4e32c011aa43

Summary:
Allow callbacks to be attached to a SettingCore (via SettingWrapper),
with lifetime managed by a guard object which automatically unregisters
the callback at destruction. This provides an API for clients to react
to changes in settings without polling.

Reviewed By: yfeldblum

Differential Revision: D25592427

fbshipit-source-id: a11a98ed651ef56a0fc86f5f5e4ed235f2289746

Summary:
the n argument in XLOG_EVERY_N is not captured by it's lambda, so
expressions using local variables can not be passed as n. EdenFS has
a use case where we would like to use local variables here. We want to use a
dynamic config value for n, so that we can easily adjust how often we log
see: D25505654.

It seems like it should be ok to capture the n here, but I am by no means a
macro expert, so please let me know if there is a tricky cpp thing here lurking
that makes capturing n bad.

Reviewed By: simpkins

Differential Revision: D25551033

fbshipit-source-id: e82fa465823d3d2ee2ac26924cc4a93b7b7242dd