Add annotations to `opensource/fbcode_builder`

Reviewed By: shannonzhu

Differential Revision: D35512626

fbshipit-source-id: db449b0f2dadcbb1ea8917adfde65f2f890ee988

defer sync write behaviour to LogWriter

Summary: A simple write+flush is a reasonable default behaviour for synchronous writes, but for some custom async implementations of log writers, we may want to avoid locking the entire queue for a full flush(). This may require different implementations, different signatures, etc. So let's defer the decision of how to do synchronous writes to the writer, with write+flush as default behaviour.

Reviewed By: Gownta

Differential Revision: D35060544

fbshipit-source-id: cb20eaecdccee7b390e4270ddbcd4c0ea4b7a649

Avoid recursive wrapping of reused buffers

Summary:
Existing reuse logic recursively wraps the tail buffers, which may result in a deep recursion on cleanup and lead to stack overflow (especially when executed on fibers).

This diff adds an optimization to avoid doing this chaining and instead always clone original `IOBuf` for the purpose of ownership.

Reviewed By: ot

Differential Revision: D35167875

fbshipit-source-id: 824026c783f7ff19db31a06734699fd1fdea2b05

Remove Faulty Assert from JemallocHugePageAllocator

Summary:
This assert is just wrong, and is breaking in production. Jemalloc can ask for
non-multiple-of-2MB chunks. This code is already resilient to that, AFAICT.

Reviewed By: davidtgoldblatt

Differential Revision: D34936556

fbshipit-source-id: e4ff49a37ce8c4ac7c378af995a58a17eb21020c

Revert D29856163: Use Huge Page Allocator in Zstd Context Pool

Differential Revision:
D29856163 (facebook@04c2275)

Original commit changeset: 38e5f2e399c1

Original Phabricator Diff: D29856163 (facebook@04c2275)

fbshipit-source-id: 172a495033b26880cc6a9fd5bd906a20cf20c66e

let coro::collectAllRange reserve its tasks vector

Summary:
It is tricky to get the size of an input range of unknown type. We must avoid iterating the range twice since iteration may be expensive.

But we do expect the input range to work with range-for, giving us options. Effectively, if the natural iterator type of the input range is a random-access iterator, then we use that as the signal that getting the size of the input range will be free.

Reviewed By: iahs

Differential Revision: D34540902

fbshipit-source-id: 3ac40ab19461e4222168d9db3e241d07b1404bc4

further optimize relaxed atomic-bit-fetch-op by removing volatile

Summary: The compiler does not move or remove `asm volatile` blocks even if it otherwise believes that all side-effects are unused or do not matter to surrounding instructions. But a relaxed operation ought to be subject to reordering.

Reviewed By: Gownta

Differential Revision: D34015943

fbshipit-source-id: 255716922faa32d95cdf9e49310db2dea1150035

cut class-qual in addrof fun-ptr in AtForkList

Summary: No need for class-qualification within the same class.

Differential Revision: D34370055

fbshipit-source-id: 53f382a5c6696aaa6f2e592834da297106477822

avoid the name I in some places

Summary:
There is a collision with `#define I` in glibc's `math/complex.h`.

https://github.com/bminor/glibc/blob/glibc-2.35/math/complex.h#L53

Differential Revision: D34202940

fbshipit-source-id: 15c6f6cd1d01e8f84f228f6ef51b9c2a42f27071

condense the atomic-fetch-bit-op implementations

Summary: Coalesce the inline-asm. Use invokers.

Reviewed By: Gownta

Differential Revision: D32302511

fbshipit-source-id: 6edf75c6ef820ed80e751d965174e5acfaac957e