Folly (acronymed loosely after Facebook Open Source Library) is a
library of C++11 components designed with practicality and efficiency
in mind. It complements (as opposed to competing against) offerings
such as Boost and of course
std. In fact, we embark on defining our
own component only when something we need is either not available, or
does not meet the needed performance profile.
Performance concerns permeate much of Folly, sometimes leading to
designs that are more idiosyncratic than they would otherwise be (see
SmallLocks.h). Good performance at large
scale is a unifying theme in all of Folly.
Folly is a collection of relatively independent components, some as simple as a few symbols. There is no restriction on internal dependencies, meaning that a given folly module may use any other folly components.
All symbols are defined in the top-level namespace
folly, except of
course macros. Macro names are ALL_UPPERCASE. Namespace
defines other internal namespaces such as
code should not depend on symbols in those namespaces.
At the top level Folly uses the classic "stuttering" scheme
folly/folly used by Boost and others. The first directory serves as
an installation root of the library (with possible versioning a la
folly-1.0/), and the second is to distinguish the library when
including files, e.g.
The directory structure is flat (mimicking the namespace structure),
i.e. we don't have an elaborate directory hierarchy (it is possible
this will change in future versions). The subdirectory
contains files that are used inside folly and possibly at Facebook but
not considered stable enough for client use. Your code should not use
folly/experimental lest it may break when you update Folly.
folly/folly/test subdirectory includes the unittests for all
components, usually named
ComponentXyzTest.cpp for each
folly/folly/docs directory contains
folly has been tested on gcc 4.6 on 64-bit installations
of Fedora 17, Ubuntu 12.04, and Debian wheezy. It might work unmodified
on other 64-bit Linux platforms.
Below is a list of Folly components in alphabetical order, along with a brief description of each.
Simple arena for memory allocation: multiple allocations get freed all at once. With threaded version.
High-performance atomic hash map with almost lock-free operation.
A small framework for benchmarking code. Client code registers benchmarks, optionally with an argument that dictates the scale of the benchmark (iterations, working set size etc). The framework runs benchmarks (subject to a command-line flag) and produces formatted output with timing information.
Various bit manipulation utilities optimized for speed; includes functions that wrap the ffsl(l) primitives in a uniform interface.
An implementation of the structure described in A Provably Correct Scalable Concurrent Skip List by Herlihy et al.
A variety of data conversion routines (notably to and from string), optimized for speed and safety.
boost::variant, but restricted to pointers only. Uses the
highest-order unused 16 bits in a pointer as discriminator. So
sizeof(DiscriminatedPtr<int, string, Widget>) == sizeof(void*).
Dynamically-typed object, created with JSON objects in mind.
Endian conversion primitives.
Escapes a string in C style.
Wrapper around the
A drop-in implementation of
std::string with a variety of optimizations.
A mostly drop-in implementation of
std::vector with a variety of
Pseudo-statements (implemented as macros) for iteration.
Python-style formatting utilities.
Group Varint encoding for 32-bit values.
Various popular hash function implementations.
A simple class for collecting histogram data.
Convenience type definitions for using
JSON serializer and deserializer. Uses
Memory allocation helpers, particularly when using jemalloc.
Helpers for finding items in associative containers (such as
A highly specialized data structure consisting of a pointer, a 1-bit spin lock, and a 15-bit integral, all inside one 64-bit word.
Necessarily evil stuff.
Pretty-printer for numbers that appends suffixes of unit used: bytes (kb, MB, ...), metric suffixes (k, M, G, ...), and time (s, ms, us, ns, ...).
Lock free single-reader, single-writer queue.
Defines only one function---
Boost-style range facility and the
Fast and compact reader-writer spin lock.
C++11 incarnation of the old ScopeGuard idiom.
Very small spin locks (1 byte and 1 bit).
Vector with the small buffer optimization and an optional embedded
Collections similar to
std::map but implemented as sorted vectors.
STL allocator wrapping a simple allocate/deallocate interface.
String utilities that connect
Subprocess library, modeled after Python's subprocess module.
High-level synchronization library.
Demangling and errno utilities.
High-performance atomic increment using thread caching.
Improved thread local storage for non-trivial types.
Queue with per-item timeout.
Type traits that complement those defined in the standard C++11 header