| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
Useful in i.e. loops that have to wait a bit before every execution.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
The reference used to become invalid during execution of the handler,
leading to subtle errors (the job just never finished). Unfortunately
I failed to write a test that catches this, as the test always seems
to work anyways....
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
The error is now propagated to the top-most (user-owned) Future. When
an error occurs, no further tasks are executed. The first error handler
function in the chain is also called, if there's any.
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
This is a simplified progress reporting, since it does not report progress
of ther overcall Job chain, but only of individual tasks, which makes it
only really useful on one-task Jobs.
TODO: propagate subjob progress to the Future user gets copy of
TODO: compound progress reporting (be able to report a progress of the overall
Job chain)
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Storing Future and current Job progress directly in Executors means
that we cannot re-execute finished job, or even execute the same Job
multiple times in parallel. To do so, we need to make Executors stateless
and track the state elsewhere.
This change does that by moving the execution state from Executor to Execution
class. Executors now only describe the tasks to execute, while Execution holds
the current state of execution. New Execution is created every time Job::exec()
is called.
Execution holds reference to it's result (Future) and Executor which created
the Execution. This ensures that Executor is not deleted when Job (which owns
Executors) goes out of scope while the execution is still running. At the same
time Future holds reference to relevant Execution, so that the Execution is
deleted when all copies of Future referring result from the respective Execution
are deleted.
|
| |
|
|
|
|
|
|
|
| |
The only reference to the top executor (the last in chain) is held by the
Job. If the Job goes out of scope after it's executed, the top Executor
is deleted (which in turn deletes the entire Executor chain). By having
the top Executor holding reference to itself during execution we ensure
that the Executor chain is not deleted until the job is finished, even
when the parent Job object is deleted in the meanwhile.
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
lifetime tests.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It is now possible use KJob-derived jobs with libasync without having to write
lambda wrappers.
auto job = Async::start<ReturnType, MyKJob, MyKJob::result, Args ...)
.then<ReturnType, OtherKJob, OtherKJob::result, PrevKJobReturnType>();
job.exec(arg1, arg2, ...);
The reason for this approach (instead of taking KJob* as an argument is that
we usually want the KJob ctor arguments to depend on result of previous job.
At least in case of Async::start() however it makes sense to support passing
KJob* as an argument (not yet implemented).
In future we should also support custom error handlers.
The KJob integration is build-time optional, but enabled by default (pass
-DWITH_KJOB=FALSE to CMake to disable).
Adds KCoreAddons dependency.
|
| |
|
|
|
|
|
|
| |
It is now possible to chain a job that takes no arguments after a job
that returns void.
Unfortunatelly it is not yet possible to disregard return value of a
previous job.
|
| |
|
|
|
|
|
|
|
| |
When user gets a Job (from a method call for instance), which is already running
or might have even finished already, they can still append a new Job to the chain
and re-execute it. The Job will internally chain up to the last finished Job, use
it's result and continue from the next Job in the chain. If a Job in the chain is
still running, it will wait for it to finish and pass the result to the next Job
in the chain.
|
| | |
|
| |
|
|
| |
multiple times
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Now it's possible to do something like
Job<int, int> job = createSomeJob();
auto main = Async::start<int>(....).then(job);
Previously the 'job' would have to be wrapped in a ThenTask-like lambda (which
is what we still do internally), but with this new syntax it's possible to append
another job chain to existing chain easilly. This syntax is available for all
task types.
|
| |
|
|
|
| |
The initial value can be passed in as argument to Job::exec(), or by another
job in case of job chaining.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
All jobs and executors now accept ErrorHandler argument which will be invoked
on error. Error handling itself has been moved to Executor::exec(), so that we
don't have to copy-paste the error handling code into every Executor implementation.
|
| |
|
|
|
|
|
|
|
| |
Sync executors don't pass Async::Future into the user-provided tasks, but
instead work with return values of the task methods, wrapping them into the
Async::Future internally. Sync tasks are of course possible since forever, but
not the API for those tasks is much cleaner, for users don't have to deal with
"future" in synchronous tasks, for instance when synchronously processing results
of an async task before passing the data to another async task.
|
| |
|
|
|
|
| |
We now hold executors in shared pointers. We cannot easilly delete them, as they
are referenced from two objects (the Job they belong to, and the next job), and
the lifetime of the jobs is unclear.
|
| |
|
|
|
|
| |
Error handlers don't have access to the future, so they can't mark it as finished,
so we do it after the error handler is run. This ensures that FutureWatchers will
finish.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
| |
The future is copied an the finished boolean has to be in the shared part,
otherwise the original copy never receives the updated value.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
