revival of the async-io functions, #90

This commit is contained in:
Chris Boesch
2026-04-02 10:38:45 +02:00
parent 3b22bfd898
commit ffde357f30
3 changed files with 58 additions and 79 deletions

View File

@@ -1,87 +1,57 @@
//
// Remember how a function with 'suspend' is async and calling an
// async function without the 'async' keyword makes the CALLING
// function async?
// When multiple async tasks access shared data, you need
// synchronization! Io provides a Mutex for this:
//
// fn fooThatMightSuspend(maybe: bool) void {
// if (maybe) suspend {}
// }
// var mutex: std.Io.Mutex = .init;
//
// fn bar() void {
// fooThatMightSuspend(true); // Now bar() is async!
// }
// // In a task:
// try mutex.lock(io); // blocks until lock is acquired
// defer mutex.unlock();
// // ... critical section: safe to modify shared data ...
//
// But if you KNOW the function won't suspend, you can make a
// promise to the compiler with the 'nosuspend' keyword:
// Without the mutex, concurrent tasks could read and write the
// same memory simultaneously, causing a data race — the result
// would be unpredictable.
//
// fn bar() void {
// nosuspend fooThatMightSuspend(false);
// }
// mutex.lock() is a cancellation point — it can return
// error.Canceled. There's also tryLock() which returns
// immediately (true if acquired, false if not).
//
// If the function does suspend and YOUR PROMISE TO THE COMPILER
// IS BROKEN, the program will panic at runtime, which is
// probably better than you deserve, you oathbreaker! >:-(
// Fix this program so the counter is correctly synchronized.
// Without the fix, the final count would be unpredictable.
// With it, four tasks incrementing 100 times each = 400.
//
const print = @import("std").debug.print;
const std = @import("std");
const print = std.debug.print;
pub fn main() void {
const SharedState = struct {
counter: u32 = 0,
mutex: std.Io.Mutex = .init,
};
// The main() function can not be async. But we know
// getBeef() will not suspend with this particular
// invocation. Please make this okay:
var my_beef = getBeef(0);
pub fn main(init: std.process.Init) !void {
const io = init.io;
var state = SharedState{};
print("beef? {X}!\n", .{my_beef});
var group: std.Io.Group = .init;
group.async(io, increment, .{ io, &state, 100 });
group.async(io, increment, .{ io, &state, 100 });
group.async(io, increment, .{ io, &state, 100 });
group.async(io, increment, .{ io, &state, 100 });
try group.await(io);
print("Counter: {} (expected: 400)\n", .{state.counter});
}
fn getBeef(input: u32) u32 {
if (input == 0xDEAD) {
suspend {}
fn increment(io: std.Io, state: *SharedState, times: u32) void {
for (0..times) |_| {
// Acquire the lock before modifying shared state.
// What Mutex method blocks until the lock is acquired?
state.mutex.??? catch return;
defer state.mutex.unlock(); // <-- what's missing here?
state.counter += 1;
}
return 0xBEEF;
}
//
// Going Deeper Into...
// ...uNdeFiNEd beHAVi0r!
//
// We haven't discussed it yet, but runtime "safety" features
// require some extra instructions in your compiled program.
// Most of the time, you're going to want to keep these in.
//
// But in some programs, when data integrity is less important
// than raw speed (some games, for example), you can compile
// without these safety features.
//
// Instead of a safe panic when something goes wrong, your
// program will now exhibit Undefined Behavior (UB), which simply
// means that the Zig language does not (cannot) define what will
// happen. The best case is that it will crash, but in the worst
// case, it will continue to run with the wrong results and
// corrupt your data or expose you to security risks.
//
// This program is a great way to explore UB. Once you get it
// working, try calling the getBeef() function with the value
// 0xDEAD so that it will invoke the 'suspend' keyword:
//
// getBeef(0xDEAD)
//
// Now when you run the program, it will panic and give you a
// nice stack trace to help debug the problem.
//
// zig run exercises/090_async7.zig
// thread 328 panic: async function called...
// ...
//
// But see what happens when you turn off safety checks by using
// ReleaseFast mode:
//
// zig run -O ReleaseFast exercises/090_async7.zig
// beef? 0!
//
// This is the wrong result. On your computer, you may get a
// different answer or it might crash! What exactly will happen
// is UNDEFINED. Your computer is now like a wild animal,
// reacting to bits and bytes of raw memory with the base
// instincts of the CPU. It is both terrifying and exhilarating.
//