An embeddable, thread-safe implementation of the cubescript language
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 

6.9 KiB

libcubescript

Build Status

CubeScript REPL

Overview

Cubescript is a minimal scripting language first introduced in the Cube FPS and carried over into derived games and game engines such as Sauerbraten. Originally being little more than a few hundred lines of code, serving primarily as the console and configuration file format of the game, it grew more advanced features as well as a bytecode VM.

Nowadays, it is a minimal but relatively fully featured scripting language based around the concept that everything can be interpreted as a string. It excels at its original purpose as well as things like text preprocessing. It comes with a Lisp-like syntax and a variety of standard library functions.

Libcubescript is a project that aims to provide an independent, improved, separate implementation of the language, available as a library, intended to satisfy the needs of the OctaForge project. It was originally forked from Cubescript as present in the Tesseract game/engine and gradually rewritten; right now, very little of the original code remains. At language level it is mostly compatible with the other implementations (although with a stricter parser and extra features), while the standard library does not aim to be fully compatible. Some features are also left up to the user to customize, so that it is not tied to game engines feature-wise.

Like the codebase it is derived from, it is available under the permissive zlib license, and therefore compatible with just about anything.

Benefits and differences

There's a variety of things that set this implementation apart:

  • It's independent and can be embedded in any project
  • There is no global state, so you can have as many Cubescripts as you want, in one program
  • Written in C++20, following modern language conventions, both internally and at API level
  • That means the ability to use lambdas as commands, including captures, type inference and so on
  • There is a robust allocator system in place, and all memory the library uses is allocated through it; that gives you complete control over its memory (for tracking, sandboxing, limits, etc.)
  • A large degree of memory safety, with no manual management
  • Strings are interned, with a single reference counted instance of any string existing at a time, which lowers memory usage and simplifies its management
  • Minimal stack memory usage, which means no artificial limits on recursion depth as well as safe usage from threads and coroutines with small stacks
  • Errors will no longer cause the interpreter to march on, instead acting like real errors
  • Protected calls allow you to catch errors in a similar way to exceptions, and nearly every error can be caught
  • Stricter parsing, with things like unfinished strings being caught
  • Loops now have break and continue statements
  • Customizable integer and floating point types
  • Full support for symbol visibility in API
  • Highly portable and cross-platform, no dependencies other than a compiler
  • Clean codebase that is easy to pick up and contribute to

More features and enhancements are planned, such as:

  • Improved support for debugging information (line information tracking at runtime rather than just compile-time)
  • Thread safety

Right now, the codebase is unstable, but quickly approaching production readiness. You are encouraged to test things and report bugs; contributions of any kind are also welcome (you can use pull requests in our Gitea instance as well as the GitHub mirror).

Our primary means of communication is the #octaforge IRC channel on Freenode.

Threads

The API provides a concept of threads. The first created thread is the main thread, which owns all variables and most state. Based on the main thread you can create side threads, which share a lot of state with the main thread but have their own call stack.

In the future, accesses to "global" state (the state shared between threads) will be made thread safe.

That means you will be able to use the library in multithreaded contexts, as long as you make sure to only use any Cubescript thread from at most one real thread at a time (accesses to thread state will not be thread-safe).

Right now, this at least means the library is coroutine-safe. You can call into a Cubescript thread inside a coroutine, yield somewhere mid-command, and still be able to access the state safely through other Cubescript threads. Once you resume the coroutine, it will continue where it left off, without anything being wrong.

Since strings are interned and reference counted, this is also geared towards thread safety - any API returning a string will give you your own reference, which means nothing can free it while you are still using it. Similarly, things taking string references will generally increment the count for their own purposes. This all happens automatically thanks to C++'s scoped value handling.

Building and usage

The library has absolutely no dependencies other than a C++20 compiler, similarly there are no dependencies on system or architecture specific things, so it should work on any OS and any CPU.

The C++20 support does not have to be complete. These are the baselines (which are ensured by the CI):

  • GCC 10
  • Clang 10 (with libstdc++ or libc++)
  • Microsoft Visual C++ 2019

Older versions of either of these are known not to work.

You will need Meson to build the project. Most Unix-like systems have it in their package management, on Windows there is an installer available on their website. Being written in Python, you can also use pip to get an up to date version on any OS.

Once you have it, compiling is simple, e.g. on Unix-likes you can do:

mkdir build && cd build
meson ..
ninja all

Refer to Meson's manual for how to customize whether you want a shared or static library and so on. By default, you will get a shared library plus a REPL (interactive interpreter). The REPL also serves as an example of how to use the API.

If you don't want the REPL, use -Drepl=disabled. When compiled, it can have support for line editing and command history. This is provided through linenoise (which is a minimal single-file line editing library bundled with the project, and is the default). In case you're on a platform that linenoise does not support (highly unlikely), there is a fallback without any line editing as well. Pass -Dlinenoise=disabled to use the fallback.

The version of linenoise bundled with the project is cpp-linenoise, available at https://github.com/yhirose/cpp-linenoise. Our version is modified, so that it builds cleanly with our flags, and so that it supports the "hints" feature available in original linenoise. Other than the modifications, it is baseed on upstream git revision a927043cdd5bfe203560802e56a7e7ed43156ed3. The reason we use this instead of upstream linenoise is Windows support.