|author||Daniel Dunbar <email@example.com>||Sun Nov 05 16:36:56 2017 -0800|
|committer||GitHub <firstname.lastname@example.org>||Sun Nov 05 16:36:56 2017 -0800|
Merge pull request #193 from ddunbar/core-wait-for-tasks-on-cancellation [Core] Fix a race condition in task cancellation.
A low-level build system.
llbuild is a set of libraries for building build systems. Unlike most build system projects which focus on the syntax for describing the build, llbuild is designed around a reusable, flexible, and scalable general purpose build engine capable of solving many “build system”-like problems. The project also includes additional libraries on top of that engine which provide support for constructing bespoke build systems (like
swift build) or for building from Ninja manifests.
llbuild currently includes:
[x] A flexible core engine capable of discovering new work on the fly.
[x] Scalability for dependency graphs reaching millions of nodes.
[x] Support for building Ninja manifests (e.g., for building LLVM, Clang, and Swift).
[x] An llbuild-native build description format designed for extensibility.
[x] Library-based design intended to support embedding and reuse.
The project currently produces three top-level products;
llbuild tool provides a command line interface to various feature of the llbuild libraries. It has several subtools available, see
llbuild --help for more information. The most important subtool is the Ninja build support:
You can use
llbuild to build Ninja-based projects using:
$ llbuild ninja build
This tool supports a subset of the command line arguments supported by Ninja itself, to allow it to be used as a compatible replacement, even by tools like CMake that depend on particular Ninja command line flags during their configuration process.
As a convenience, if you invoke
llbuild via a symlink called
ninja then it will automatically use this subtool. This supports installing llbuild as
ninja into your
PATH and then using it as an alternative to Ninja for building arbitrary projects (like LLVM, Clang, and Swift). This is also how we self-host
llbuild (via the CMake Ninja generator).
llbuild ninja subtool also provides additional commands which are primarily only useful for developers interested in working on the Ninja support. These commands allow testing the lexer, parser, and manifest loading components independently and are used as part of the test suite.
swift-build-tool product is the command line interface to the build system used by the Swift Package Manager. It is built as part of the Swift project build and incorporated into the Swift language snapshots.
This tool is built on top of the BuildSystem library.
libllbuild library exposes a C API for the llbuild libraries, which can be used directly by third-parties or to build additional language bindings. See bindings for example Swift and Python bindings that use this library.
This API is what is used, for example, in Xcode as the basis for the new build system introduced in Xcode 9.
The design of llbuild is a continuation of the LLVM philosophy of applying library-based design to traditional developer tools. Clang has followed this approach to deliver a high performance compiler and assembler while also enabling new tools like clang-format or the libclang interfaces for code completion and indexing. However, the rigid command line interface between traditional build systems and the compiler still limits the optimizations and features which can be implemented in Clang.
llbuild is designed to allow construction of more feature rich build environments which integrate external tools -- like the compiler -- using APIs instead of command line interfaces. By allowing the build system and tools to communicate directly and to be co-designed, we believe we can unlock additional optimization opportunities and create more robust, easy-to-use build systems.
For more information, see A New Architecture for Building Software from the 2016 LLVM Developer's Conference.
In the abstract, build systems are used to perform a task while also being:
Incremental: Outputs should be efficiently rebuilt given a small change to the inputs, by leveraging the ability to save partial outputs from a prior build.
Consistent: Equivalent inputs should always produce the same result as building from clean.
Persistent: Results should be stored so that builds can be interrupted and resumed after failure without needing to redo the full computation.
Parallel & Efficient: It must be possible to perform independent elements of the computation in parallel, in order to compute the result as efficiently as possible.
When viewed in this light, it is clear that the core technology of a build system is applicable to any complex, long-running computation in which it is common for the user to only modify a small portion of the input before wanting the recompute the result. For example, a movie editor application will commonly need to rerender small portions of the overall movie in response to interactive edits in order to support preview of the final result. However, such applications frequently do not take full advantage of the ability to store and partially recompute the results because of the complexity of correctly managing the dependencies between parts of the computation.
Part of the goal in designing llbuild around a general purpose build engine is to allow its use in contexts which are not traditionally thought of as requiring a “build system”.
Technical documentation is available at llbuild.readthedocs.io.
Bug reports should be filed in the Swift OSS Jira in the
llbuild is a work in progress. Some of the more significant open projects which we hope to tackle are:
[ ] Support for using file signatures instead of timestamps for change detection.
[ ] Support richer data types for communication between tasks.
Tasks currently only compute a single scalar value as their result. We would like to support richer data types for tasks results, for example tasks should be able to compute sets of results, and have the engine automatically communicate the addition or removal of individual items in the set to downstream consumers.
[ ] Support a more sophisticated database implementation.
The current implementation uses a SQLite3 database for storing build results. This was a pragmatic choice for bring up, but it can be a performance bottleneck for some applications, and we do not need the flexibility of a full SQL database. We would like to evaluate the tradeoffs of designing a custom solution for llbuild.
[ ] Support transparent distributed builds.
We would like llbuild to have facilities for transparently distributing a build across an array of worker machines.
[ ] Support automatic auditing of build consistency.
Few build systems diagnose problems effectively. Frequently, undeclared inputs or misbehaving tools can cause inconsistent build results. We would like llbuild to automatically diagnose these problems, for example by periodically or speculatively rebuilding items which are not expected to have changed and comparing the results.
[ ] Performance tuning of core engine queues.
The core build engine does its work using a number of queues of work items, and locking for the subset which support concurrent manipulation. We would like to investigate moving the shared queues to using a lock-free data structure and to micro-optimize the queues in general, in order to support very fine-grained task subdivisions without negatively impacting performance.
Q. Why does llbuild include some parts of LLVM?
A. As a low-level, embeddable component, we want llbuild itself to have a simple build process without any significant build time dependencies. However, we also wanted to take advantage of some of the data structures and support facilities that have been developed for LLVM. For now, our solution is to incorporate some parts of LLVM's Support libraries into the repository, with the hope that over time LLVM will either factor out those libraries in a way that makes it easier to reuse them, or that we will develop our own exclusive set of support data structures and utilities and drop use of the LLVM ones.
Q. Why does llbuild include Ninja support?
A. llbuild includes a Ninja compatibility layer which allows building projects which use Ninja manifests using the llbuild core engine. We developed this support as a proof of concept for the core engine, and as a way to bootstrap ourselves (we develop llbuild using the CMake Ninja generator and llbuild to build itself). This support is also valuable for allowing direct benchmarking comparisons of llbuild.
Our implementation of Ninja support also includes a separate library for programmatically loading Ninja manifests, which may prove useful to other projects wishing to use or manipulate Ninja files.
We intend to continue to maintain the Ninja support to keep compatibility with the main project.
llbuild is heavily influenced by modern build systems like Shake, Buck, and Ninja. We would particularly like to thank Neil Mitchell for his work describing the Shake algorithm which provided the inspiration for the mechanism llbuild uses to allow additional work to be discovered on the fly.
Copyright (c) 2014 - 2017 Apple Inc. and the Swift project authors. Licensed under Apache License v2.0 with Runtime Library Exception.
See http://swift.org/LICENSE.txt for license information.
See http://swift.org/CONTRIBUTORS.txt for Swift project authors.