Struct Context

#[repr(C)]
pub struct Context(/* private fields */);

A Context stores a ContextFn’s state of execution, for it to be resumed later.

If we have 2 or more Context instances, we can thus easily “freeze” the current state of execution and explicitely switch to another Context. This Context is then resumed exactly where it left of and can in turn “freeze” and switch to another Context.

Examples

See examples/basic.rs

Implementations

impl Context

pub unsafe fn new(stack: &Stack, f: ContextFn) -> Context

Creates a new Context prepared to execute f at the beginning of stack.

f is not executed until the first call to resume().

It is unsafe because it only takes a reference of Stack. You have to make sure the Stack lives longer than the generated Context.

Examples found in repository

examples/basic.rs (line 31)

fn main() {
    // This method will always `resume()` immediately back to the
    // previous `Context` with a `data` value incremented by one starting at 0.
    // You could thus describe this method as a "natural number generator".
    extern "C" fn context_function(mut t: Transfer) -> ! {
        for i in 0usize.. {
            print!("Yielding {} => ", i);
            t = unsafe { t.context.resume(i) };
        }

        unreachable!();
    }

    // Allocate some stack.
    let stack = ProtectedFixedSizeStack::default();

    // Allocate a Context on the stack.
    let mut t = Transfer::new(unsafe { Context::new(&stack, context_function) }, 0);

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        // The first and every other call will return references to the actual `Context` data.
        print!("Resuming => ");
        t = unsafe { t.context.resume(0) };

        println!("Got {}", t.data);
    }

    println!("Finished!");
}

examples/fibonacci.rs (line 36)

fn main() {
    // This method will always `resume()` immediately back to the
    // previous `Context` with a `data` value of the next number in the fibonacci sequence.
    // You could thus describe this method as a "fibonacci sequence generator".
    extern "C" fn context_function(mut t: Transfer) -> ! {
        let mut a = 0usize;
        let mut b = 1usize;

        loop {
            print!("Yielding {} => ", a);
            t = unsafe { t.context.resume(a) };

            let next = a + b;
            a = b;
            b = next;
        }
    }

    // Allocate some stack.
    let stack = ProtectedFixedSizeStack::default();

    // Allocate a Context on the stack.
    let mut t = Transfer::new(unsafe { Context::new(&stack, context_function) }, 0);

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        // The first and every other call will return references to the actual `Context` data.
        print!("Resuming => ");
        t = unsafe { t.context.resume(0) };

        println!("Got {}", t.data);
    }

    println!("Finished!");
}

examples/how_to_ontop.rs (line 135)

pub fn run() {
    // Allocate some stack.
    let mut some_stack = Some(ProtectedFixedSizeStack::default());
    let stack_ref = stack_ref_from_some_stack(&mut some_stack);

    // Allocate a Context on the stack.
    let mut ctx = unsafe { Context::new(some_stack.as_ref().unwrap(), context_function) };

    // Yield to context_function(). This important since the returned `Context` reference is
    // different than the one returned by `Context::new()` (since it points to the entry function).
    // It's important that we do this first or else calling `Context::resume_ontop()` will crash.
    // See documentation of `Context::resume_ontop()` for more information.
    // Furthermore we pass a reference to the Option<ProtectedFixedSizeStack> along with it
    // so it can delete it's own stack (which is important for stackful coroutines).
    let Transfer { context, data } = unsafe { ctx.resume(stack_ref) };
    ctx = context;

    // Store the pointer to the Carrier for `unwind_stack`.
    let carrier_ptr = data;

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        print!("Resuming => ");
        let Transfer { context, data } = unsafe { ctx.resume(0) };
        ctx = context;

        println!("Got {}", data);
    }

    // Resume `context_function()` with the ontop function `unwind_stack()`.
    // Before it returns from it's own call to `resume()` it will call `unwind_stack()`.
    println!("Resuming context with unwind_stack() ontop!");
    unsafe { ctx.resume_ontop(carrier_ptr, unwind_stack) };

    match some_stack {
        Some(..) => println!("Stack is still there (this should not happen here)!"),
        None => println!("Stack has been deleted!"),
    }

    println!("Finished!");
}

pub unsafe fn resume(self, data: usize) -> Transfer

Yields the execution to another Context.

The exact behaviour of this method is implementation defined, but the general mechanism is: The current state of execution is preserved somewhere and the previously saved state in the Context pointed to by self is restored and executed next.

This behaviour is similiar in spirit to regular function calls with the difference that the call to resume() only returns when someone resumes the caller in turn.

The returned Transfer struct contains the previously active Context and the data argument used to resume the current one.

It is unsafe because it is your responsibility to make sure that all data that constructed in this context have to be dropped properly when the last context is dropped.

Examples found in repository

examples/basic.rs (line 21)

fn main() {
    // This method will always `resume()` immediately back to the
    // previous `Context` with a `data` value incremented by one starting at 0.
    // You could thus describe this method as a "natural number generator".
    extern "C" fn context_function(mut t: Transfer) -> ! {
        for i in 0usize.. {
            print!("Yielding {} => ", i);
            t = unsafe { t.context.resume(i) };
        }

        unreachable!();
    }

    // Allocate some stack.
    let stack = ProtectedFixedSizeStack::default();

    // Allocate a Context on the stack.
    let mut t = Transfer::new(unsafe { Context::new(&stack, context_function) }, 0);

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        // The first and every other call will return references to the actual `Context` data.
        print!("Resuming => ");
        t = unsafe { t.context.resume(0) };

        println!("Got {}", t.data);
    }

    println!("Finished!");
}

examples/fibonacci.rs (line 24)

fn main() {
    // This method will always `resume()` immediately back to the
    // previous `Context` with a `data` value of the next number in the fibonacci sequence.
    // You could thus describe this method as a "fibonacci sequence generator".
    extern "C" fn context_function(mut t: Transfer) -> ! {
        let mut a = 0usize;
        let mut b = 1usize;

        loop {
            print!("Yielding {} => ", a);
            t = unsafe { t.context.resume(a) };

            let next = a + b;
            a = b;
            b = next;
        }
    }

    // Allocate some stack.
    let stack = ProtectedFixedSizeStack::default();

    // Allocate a Context on the stack.
    let mut t = Transfer::new(unsafe { Context::new(&stack, context_function) }, 0);

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        // The first and every other call will return references to the actual `Context` data.
        print!("Resuming => ");
        t = unsafe { t.context.resume(0) };

        println!("Got {}", t.data);
    }

    println!("Finished!");
}

examples/how_to_ontop.rs (line 100)

extern "C" fn context_function(t: Transfer) -> ! {
    println!("Entering context_function...");

    // Take over the stack from the main function, because we want to manage it ourselves.
    // The main function could safely return after this in theory.
    let mut some_stack = take_some_stack_from_transfer(&t);
    let stack_ref = stack_ref_from_some_stack(&mut some_stack);

    let (result, context) = {
        let mut carrier = Carrier { context: Some(t.context) };

        let carrier_ptr = &mut carrier as *mut _ as usize;

        // Use `std::panic::catch_unwind()` to catch panics from `unwind_stack()`.
        let r = panic::catch_unwind(|| {
            // We use an instance of `Dropper` to demonstrate
            // that the stack is actually being unwound.
            let _dropper = Dropper;

            let carrier = unsafe { &mut *(carrier_ptr as *mut Carrier) };

            // We've set everything up! Go back to `main()`!
            println!("Everything's set up!");
            let context = carrier.context.take().unwrap();
            let Transfer { context, .. } = unsafe { context.resume(carrier_ptr) };
            carrier.context = Some(context);

            for i in 0usize.. {
                print!("Yielding {} => ", i);
                let context = carrier.context.take().unwrap();
                let Transfer { context, .. } = unsafe { context.resume(i) };
                carrier.context = Some(context);
            }
        });

        (r, carrier.context.take().unwrap())
    };

    match result {
        Ok(..) => println!("Finished loop without panicking (this should not happen here)!"),
        Err(..) => println!("Recovered from a panic!"),
    }

    // We own the stack (`main()` gave it to us) and we need to delete it.
    // Since it would be unsafe to do so while we're still in the context function running on
    // that particular stack, we defer deletion of it by resuming `main()` and running the ontop
    // function `delete_stack()` before `main()` returns from it's call to `resume_ontop()`.
    println!("Defer stack deallocation by returning to main()!");
    unsafe { context.resume_ontop(stack_ref, delete_stack) };

    unreachable!();
}

pub fn run() {
    // Allocate some stack.
    let mut some_stack = Some(ProtectedFixedSizeStack::default());
    let stack_ref = stack_ref_from_some_stack(&mut some_stack);

    // Allocate a Context on the stack.
    let mut ctx = unsafe { Context::new(some_stack.as_ref().unwrap(), context_function) };

    // Yield to context_function(). This important since the returned `Context` reference is
    // different than the one returned by `Context::new()` (since it points to the entry function).
    // It's important that we do this first or else calling `Context::resume_ontop()` will crash.
    // See documentation of `Context::resume_ontop()` for more information.
    // Furthermore we pass a reference to the Option<ProtectedFixedSizeStack> along with it
    // so it can delete it's own stack (which is important for stackful coroutines).
    let Transfer { context, data } = unsafe { ctx.resume(stack_ref) };
    ctx = context;

    // Store the pointer to the Carrier for `unwind_stack`.
    let carrier_ptr = data;

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        print!("Resuming => ");
        let Transfer { context, data } = unsafe { ctx.resume(0) };
        ctx = context;

        println!("Got {}", data);
    }

    // Resume `context_function()` with the ontop function `unwind_stack()`.
    // Before it returns from it's own call to `resume()` it will call `unwind_stack()`.
    println!("Resuming context with unwind_stack() ontop!");
    unsafe { ctx.resume_ontop(carrier_ptr, unwind_stack) };

    match some_stack {
        Some(..) => println!("Stack is still there (this should not happen here)!"),
        None => println!("Stack has been deleted!"),
    }

    println!("Finished!");
}

pub unsafe fn resume_ontop(self, data: usize, f: ResumeOntopFn) -> Transfer

Yields the execution to another Context and executes a function “ontop” of it’s stack.

This method identical to resume() with a minor difference:

The argument f is executed right after the targeted Context, pointed to by self, is woken up, but before it returns from it’s call to resume(). f now gets passed the Transfer struct which would normally be returned by resume() and is allowed to inspect and modify it. The Transfer struct f returns is then finally the one returned by resume() in the targeted Context.

This behaviour can be used to either execute additional code or map the Transfer struct to another one before it’s returned, without the targeted Context giving it’s consent. For instance it can be used to unwind the stack of an unfinished Context, by calling this method with a function that panics, or to deallocate the own stack, by deferring the actual deallocation until we jumped to another, safe Context.

It is unsafe because it is your responsibility to make sure that all data that constructed in this context have to be dropped properly when the last context is dropped.

Examples found in repository

examples/how_to_ontop.rs (line 124)

extern "C" fn context_function(t: Transfer) -> ! {
    println!("Entering context_function...");

    // Take over the stack from the main function, because we want to manage it ourselves.
    // The main function could safely return after this in theory.
    let mut some_stack = take_some_stack_from_transfer(&t);
    let stack_ref = stack_ref_from_some_stack(&mut some_stack);

    let (result, context) = {
        let mut carrier = Carrier { context: Some(t.context) };

        let carrier_ptr = &mut carrier as *mut _ as usize;

        // Use `std::panic::catch_unwind()` to catch panics from `unwind_stack()`.
        let r = panic::catch_unwind(|| {
            // We use an instance of `Dropper` to demonstrate
            // that the stack is actually being unwound.
            let _dropper = Dropper;

            let carrier = unsafe { &mut *(carrier_ptr as *mut Carrier) };

            // We've set everything up! Go back to `main()`!
            println!("Everything's set up!");
            let context = carrier.context.take().unwrap();
            let Transfer { context, .. } = unsafe { context.resume(carrier_ptr) };
            carrier.context = Some(context);

            for i in 0usize.. {
                print!("Yielding {} => ", i);
                let context = carrier.context.take().unwrap();
                let Transfer { context, .. } = unsafe { context.resume(i) };
                carrier.context = Some(context);
            }
        });

        (r, carrier.context.take().unwrap())
    };

    match result {
        Ok(..) => println!("Finished loop without panicking (this should not happen here)!"),
        Err(..) => println!("Recovered from a panic!"),
    }

    // We own the stack (`main()` gave it to us) and we need to delete it.
    // Since it would be unsafe to do so while we're still in the context function running on
    // that particular stack, we defer deletion of it by resuming `main()` and running the ontop
    // function `delete_stack()` before `main()` returns from it's call to `resume_ontop()`.
    println!("Defer stack deallocation by returning to main()!");
    unsafe { context.resume_ontop(stack_ref, delete_stack) };

    unreachable!();
}

pub fn run() {
    // Allocate some stack.
    let mut some_stack = Some(ProtectedFixedSizeStack::default());
    let stack_ref = stack_ref_from_some_stack(&mut some_stack);

    // Allocate a Context on the stack.
    let mut ctx = unsafe { Context::new(some_stack.as_ref().unwrap(), context_function) };

    // Yield to context_function(). This important since the returned `Context` reference is
    // different than the one returned by `Context::new()` (since it points to the entry function).
    // It's important that we do this first or else calling `Context::resume_ontop()` will crash.
    // See documentation of `Context::resume_ontop()` for more information.
    // Furthermore we pass a reference to the Option<ProtectedFixedSizeStack> along with it
    // so it can delete it's own stack (which is important for stackful coroutines).
    let Transfer { context, data } = unsafe { ctx.resume(stack_ref) };
    ctx = context;

    // Store the pointer to the Carrier for `unwind_stack`.
    let carrier_ptr = data;

    // Yield 10 times to `context_function()`.
    for _ in 0..10 {
        // Yield to the "frozen" state of `context_function()`.
        // The `data` value is not used in this example and is left at 0.
        print!("Resuming => ");
        let Transfer { context, data } = unsafe { ctx.resume(0) };
        ctx = context;

        println!("Got {}", data);
    }

    // Resume `context_function()` with the ontop function `unwind_stack()`.
    // Before it returns from it's own call to `resume()` it will call `unwind_stack()`.
    println!("Resuming context with unwind_stack() ontop!");
    unsafe { ctx.resume_ontop(carrier_ptr, unwind_stack) };

    match some_stack {
        Some(..) => println!("Stack is still there (this should not happen here)!"),
        None => println!("Stack has been deleted!"),
    }

    println!("Finished!");
}

Trait Implementations

impl Debug for Context

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.