1 files changed, 236 insertions, 0 deletions
diff --git a/tasks/main.go b/tasks/main.go
new file mode 100644
index 0000000..13832c4
--- /dev/null
+++ b/tasks/main.go
@@ -0,0 +1,236 @@
+// Package tasks is a simple package for tracking the state of particular asynchronous, or
+// background tasks. Simply define a "Run" method to satisfy the Task interface, then
+// they can be added to a group for tracking.
+package tasks
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"log"
+	"reflect"
+	"strings"
+	"sync"
+)
+
+var (
+	EAlreadyRegistered = errors.New("a task has already been registered with this name")
+	EAlreadyStarted    = errors.New("specified task has already been started")
+	ENotRegistered     = errors.New("task has not been registered")
+
+	// logger = log.New(os.Stderr, "", log.LstdFlags|log.Lshortfile)
+	logger = log.New(io.Discard, "", 0)
+)
+
+const (
+	_ taskState = 1 << iota
+	sTodo
+	sRunning
+	sFinished
+	sError
+)
+
+type taskState int
+
+// Task interface is wraps anything with a Run(), it's assumed
+// that the Run() is blocking and will return when the task
+// is complete.
+type Task interface {
+	Run() error
+}
+
+// Group represents a group of tasks, and will keep track of their various states
+// as well as any errors. While this library does not concurrently access the
+// tasks it's likely you'll want them to be safe for concurrent use if you're
+// accessing the tasks elsewhere.
+//
+// Tasks can be run more than once if Start or Run is called more than once.
+// no effort is made to prevent that other than not allowing jobs to stack.
+// ( They will return an error if they're currently running )
+type Group struct {
+	tasks map[string]*taskEntry
+	m     *sync.RWMutex
+}
+
+func NewGroup() *Group {
+	return &Group{
+		tasks: map[string]*taskEntry{},
+		m:     &sync.RWMutex{},
+	}
+}
+
+type TaskList map[string]Task
+
+func (tl TaskList) String() string {
+	bld := &strings.Builder{}
+	for name, entry := range tl {
+		bld.Write([]byte{' '})
+		bld.WriteString(fmt.Sprintf("%s->%s(%+v)", name,
+			reflect.TypeOf(entry), entry))
+	}
+	return bld.String()
+}
+
+// Running returns all the running Tasks. These are not copies, so care must
+// be taken if they're not thread safe.
+func (g *Group) Running() TaskList {
+	return g.getForCondition(
+		func(state taskState) bool {
+			return state&sRunning > 0
+		})
+}
+
+// Todo returns all the tasks in the Todo state. Run or Start simply has not
+// been called.
+func (g *Group) Todo() TaskList {
+	return g.getForCondition(
+		func(state taskState) bool {
+			return state&sTodo > 0
+		})
+}
+
+func (g *Group) Finished() TaskList {
+	return g.getForCondition(
+		func(state taskState) bool {
+			return state&sFinished > 0
+		})
+}
+
+func (g *Group) Error() TaskList {
+	return g.getForCondition(
+		func(state taskState) bool {
+			return state&sError > 0
+		})
+}
+
+func (g *Group) GetError(name string) error {
+	var err error
+	g.m.RLock()
+	defer g.m.RUnlock()
+	entry, ok := g.tasks[name]
+	if !ok {
+		return fmt.Errorf("task: %s not found: %w", name, ENotRegistered)
+	}
+	err = entry.Error
+	return err
+}
+
+// Registered returns all the registered tasks
+func (g *Group) Registered() TaskList {
+	return g.getForCondition(
+		func(state taskState) bool {
+			return true
+		})
+}
+
+func (g *Group) getForCondition(condition func(state taskState) bool) TaskList {
+	g.m.Lock()
+	defer g.m.Unlock()
+	out := TaskList{}
+	for name, entry := range g.tasks {
+		if condition(entry.State) {
+			out[name] = entry.Task
+		}
+	}
+	return out
+}
+
+// AddAndStart does exactly what ti says on the tin
+func (g *Group) AddAndStart(name string, task Task) error {
+	err := g.Add(name, task)
+	if err != nil {
+		return err
+	}
+	return g.Start(name)
+}
+
+// AddAndRun does exactly what it says on the tin
+func (g *Group) AddAndRun(name string, task Task) error {
+	err := g.Add(name, task)
+	if err != nil {
+		return err
+	}
+	return g.Run(name)
+}
+
+// Add will register a task with the group, however it does not run it
+// immediately. Allowing you to call Run or Start separately.
+func (g *Group) Add(name string, task Task) error {
+	g.m.Lock()
+	defer g.m.Unlock()
+	_, ok := g.tasks[name]
+	if ok {
+		return fmt.Errorf("%s already registered: %w", name, EAlreadyRegistered)
+	}
+	g.tasks[name] = &taskEntry{
+		Task:  task,
+		State: sTodo,
+	}
+	return nil
+}
+
+// Start will start the task but not wait around for it to finish, returns
+// an error if the task isn't registered or is already started.
+func (g *Group) Start(name string) error {
+	g.m.Lock()
+	entry, ok := g.tasks[name]
+	if !ok {
+		g.m.Unlock()
+		return fmt.Errorf("'%s' not registered %w", name, ENotRegistered)
+	}
+	logger.Printf("Start %s -> state: %d condition: %v",
+		name, entry.State, entry.State&sRunning > 0)
+	if entry.State&sRunning > 0 {
+		g.m.Unlock()
+		return fmt.Errorf("'%s' already running %w", name, EAlreadyStarted)
+	}
+	go func() {
+		g.m.Unlock()
+		_ = entry.run(g.m, name)
+	}()
+	return nil
+}
+
+// Run will do the same thing as start, except that it will block and
+// wait for the task to finish, and return the task's error.
+func (g *Group) Run(name string) error {
+	g.m.Lock()
+	entry, ok := g.tasks[name]
+	g.m.Unlock()
+	if !ok {
+		return fmt.Errorf("'%s' not registered %w", name, ENotRegistered)
+	}
+	return entry.run(g.m, name)
+}
+
+type taskEntry struct {
+	Task  Task
+	State taskState
+	Error error
+}
+
+func (e *taskEntry) getState(m *sync.RWMutex) taskState {
+	m.RLock()
+	defer m.RUnlock()
+	return e.State
+}
+
+func (e *taskEntry) run(m *sync.RWMutex, name string) error {
+	s := e.getState(m)
+	if s&sRunning > 0 {
+		return fmt.Errorf("'%s' already running %w", name, EAlreadyStarted)
+	}
+	m.Lock()
+	e.State = sRunning
+	m.Unlock()
+
+	err := e.Task.Run()
+	m.Lock()
+	e.State = sFinished
+	e.Error = err
+	if err != nil {
+		e.State = e.State | sError
+	}
+	m.Unlock()
+	return err
+}