README.md

coop

Coop is an experimental asynchronous multitasking scheduler that tries to run everywhere. It has literally no run time dependencies except for a proper x86_64/amd64 machine set to long mode. It's only build-time dependency is a C compiler supporting GCC inline assembly. Technically, coop should run, and work, on everything you can compile to, including baremetal and even Windows.

Additionally, if you want to use coop while staying sane and stack overflow-less you will need some memory allocation function like malloc() and if you want to regain some of you precious memory also one like free().

Usage example

#include <stdio.h>
#include <stdlib.h>

#include "lib/coop.h"

void my_async_func() {
    puts("Func1: 1");
    yield;
    puts("Func1: 2");
    dump;
}

void my_async_func2() {
    puts("Func2: 1");
    yield;
    puts("Func2: 2");
    dump;
}

int main() {
    coop_context ctx = make_context();
    task_t t1 = {0}, t2 = {0};
    coop_stack s1 = {.size = 4096, .stack_bottom = malloc(4096)};
    coop_stack s2 = {.size = 4096, .stack_bottom = malloc(4096)};
    add_task(&ctx, &t1, s1, my_async_func);
    add_task(&ctx, &t2, s2, my_async_func2);
    start(&ctx);
    free(s1.stack_bottom);
    free(s2.stack_bottom);
    return 0;
}

Pitfalls

Most importantly coop isn't tested further than what's in test.c

coop does have two more semi-runtime dependency. Your compiler has to push it's rbp value to the stack. If it doesn't, you will get at least the last variable created before yielding control, or, if you don't have a variable, the return pointer, overwritten, both of which can cause very bad things to happen. Additionally, the ABI you are using can't consider RAX and RDX nonvolatile (callee-saved), since coop a) uses these registers internally and therefore b) doesn't save them. Luckily this is a) true for the two major ABI's and b) nobody would create such an ABI because these registers are used by both the div and idiv instruction.