| Name | License | Description |
|---|---|---|
| libctx | Boost v1.0 | A cooperative multitasking library for C/C++ |
libctx began life as a port of Boost::Context to C and ended up a substantial rewrite for speed (extremely minor), flexibility and new architecture support. This library enables functions like co-routines, exceptions or other neat tricks in C. libctx attempts (and succeeds!) at being small and self- contained. While the features found here are fun and interesting, some care should be taken in their use since, without careful planning, things can go downhill quickly.
libctx has been tested in the following configurations:
| x86 | arm | sparc | ppc | mips | mipsel | |
|---|---|---|---|---|---|---|
| Debian | 32/64 | 32/64 | 32/64 | 32/64 | 32 | 32 |
| Fedora | 32/64 | 32/64 | ||||
| FreeBSD | 32/64 | |||||
| NetBSD | 32/64 | |||||
| OpenBSD | 32/64 | |||||
| Apple OS X | 32/64 | |||||
| Solaris | 32/64 |
There have long been functions for jumping around code in C. Several have seen wide adoption, but suffer from inconsistent implementations or other restrictions.
The earliest are setjmp()/longjmp(). These functions are available on pretty much any UNIX/Linux system, but come with a number of problems. Most severely, because you cannot execute functions on an independent stacks, you can only jump in the direction of stack unwinding. Secondarily, implementations provide inconsistent support for extended CPU state saving. One feature that setjmp has that others don't is support for saving/restoring signal mask state. While this sounds neat at first, mixing signals and cooperative multitasking is bad mojo. Just don't do it (there is a reason no one else implements this).
Next came an attempt at UNIX standardization via ucontext. This approach provides a mechanism for independent stacks, but still suffers from inconsistent implementations. Some implementations of ucontext also depend on memory allocation during state save, decreasing performance.
Of the many other attempts not worth mentioning, Boost::Context stands out. It is mature, consistent and widely-deployed. Its main problem is that it is written in C++ and I wished to use it in C applications. If you are already using Boost, you probably want to use Boost::Context. However, if you don't want to add a dependency on C++, or you don't plan on using Boost, you might find libctx useful. One other reason you may want to use libctx instead of Boost::Context is that libctx supports AARCH64 (a.k.a. ARMv8).
I don't currently plan on supporting Windows. Not that I'm against it, I just don't have systems to test it on. Boost::Context already provides Windows support. Feel free to use it. If you'd like to use libctx on Windows, it should be a fairly straightforward port from Boost::Context. I'd be happy to merge any such patches.