bpt Design and Rationale

bpt has been designed from the very beginning as an extremely opinionated hybrid build system and package manager. Unlike most build systems however, bpt has a strong focus on a particular aspect of software development: C and C++ libraries.

This may sound pointless, right? Libraries are useless unless we can use them to build applications!

Indeed, applications are essential, and bpt is able to build those as well.

Another design decision is that bpt is built to be driven by automated tools as well as humans. bpt is not designed to entirely replace existing build systems and package management solutions. Rather, it is designed to be easy to integrate with existing systems and tools.

Background

I’m going to say something somewhat controversial: C and C++ don’t need “package management.” At least, not generalized “package management.” C++ needs library “package management.”

The C and C++ compilation model is inherently more complex than almost any other language in use today. This isn’t to say “bad,” but rather than it is built to meet extremely high and strange demands. It also comes with a large burden of legacy. Meeting both of these requirements simultaneously presents incredible implementation challenges.

Despite the vast amount of work put into build systems and tooling, virtually all developers are using them incorrectly and/or dangerously without realizing it, and we seem to be still a great distance from a unified library package distribution and consumption mechanism.

Tabula Rasa

bpt attempts to break from the pattern of legacy demands and strange usage demands in a few ways. The major differences between bpt and other build systems like CMake, Meson, build2, SCons, MSBuild, etc. is that of tradeoffs. If you opt-in to have your library built by bpt, you forgo customizability in favor of simplicity and ease.

bpt takes a look at what is needed to build and develop libraries and optimizes for that use case. It is also built with a very strong, very opinionated idea of how libraries should be constructed and used. These prescriptions are not at all arbitrary, though. They are built upon the observations of the strengths and weaknesses of build systems in use throughout industry and community.

There is some ambiguity on the term “build system.” It can mean one of two things:

  1. A proper noun “Build System,” such as CMake, Meson, Autotools, or even Gulp, WebPack, and Mix. These are specific tools that have been developed for the implementation of the second definition:

  2. A general noun “build system” refers to the particular start-to-finish process through which a specific piece of software is mapped from its raw inputs (source code, resource libraries, toolchains) to the outputs (applications, appliances, libraries, or web sites).

For example, LLVM and Blender both use the CMake “Build System,” but their “build system” is not the same. The “build system” for each is wildly different, despite both using the same underlying “Build System.”

bpt takes a massive divergence at this point. One project using bpt as their build system has a nearly identical build process to every other project using bpt. Simply running bpt build should be enough to build any bpt project.

In order to reach this uniformity and simplicity, bpt drops almost all aspects of project-by-project customizability. Instead, bpt affords the developer a contract:

If you play by the rules, you get to play in this space.

The Rules

We’ve talked an awful lot about the “rules” and “restrictions” that bpt imposes, but what are they?

bpt Is not Made for Complex Applications

Alright, this one isn’t a “rule” as much as a recommendation: If you are building an application that needs some build process functionality that bpt does not provide, bpt is only open to changes that do not violate any of the other existing rules.

Note

However: If you are a library author and you find that bpt cannot correctly build your library without violating other rules, we may have to take a look. This is certainly not to say it will allow arbitrary customization features to permit the rules to be bent arbitrarily: Read on.

bpt does contain functionality for building applications, but they must also play by the rules.

If you want to build a complex application with bpt that uses lots of platform-specific sources, code generation, and conditional components, a good option is to use an external build script that prepares the project tree before invoking bpt.

Your Code Should Be Changed Before bpt Should Be Changed

The wording of this rule means that the onus is on the developer to meet the expectations that bpt prescribes in order to make the build work.

If your project meets all the requirements outlined in this document but you still find trouble in making your build work, or if you cannot see any possible way for your project to be built by bpt regardless of what changes you make, then it this is grounds for change in bpt, either in clarifying the rules or tweaking bpt functionality.

Library Projects Must Meet the Layout Requirements

This is a very concrete requirement. bpt prescribes a particular project structure layout with minimal differing options. bpt prescribes the Pitchfork layout requirements.

Note

These prescriptions are not as draconian as they may sound upon first reading. Refer to the Projects & Packages page for more information.

A Library Build Must Successfully Compile All Source Files

Almost all Build Systems have a concept of conditionally adding a source file to a build. bpt elides this feature in place of relying on in-source conditional compilation.

All Code Must Be in Place Before Building

bpt does not provide code-generation functionality. Instead, any generated code should be generated by separate build steps before bpt is executed.

No Arbitrary #include Directories

Only src/ and include/ will ever be used as the basis for header resolution while building a project, so all #include directives should be relative to those directories. Refer to Source Root Kinds.

All Files Compile with the Same Options

When bpt compiles a project, every source file will be compiled with an identical set of options. Additionally, when bpt compiles a dependency tree, every library in that dependency tree will be compiled with an identical set of options. Refer to the Toolchains page for more information.

Currently, the only exception to this rules is for flags that control compiler diagnostics: Dependencies will be compiled without adding any warnings flags, while the main project will be compiled with warnings enabled by default.