1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
|
.. _coroutines:
Coroutines
----------
Coroutines are used to break up large and computationally expensive
tasks into smaller tasks, where control is relinquished to the main
loop after each smaller task. Coroutines are also very useful in
constructing state machines. In the event where blocking is
unavoidable, and the duration of the block is unknown (for example,
connecting to a remote host, or scaling a very large image), threads
can be used. These two different approaches are unified with a very
similar API.
A function or method is designated a coroutine by using the @kaa.coroutine
decorator. A coroutine allows a larger tasks to be broken down into smaller
ones by yielding control back to the "scheduler" (the :ref:`notifier
<notifier>`), implementing a kind of cooperative multitasking. More usefully,
coroutines can yield at points where they may otherwise block on resources
(e.g. disk or network), and when the resource becomes available, the coroutine
resumes where it left off. With coroutines and :ref:`InProgress <inprogress>`
objects, it is possible to construct non-trivial state machines, whose state is
modified by asynchronous events, using a single coroutine. Without coroutines,
this is typically implemented as a series of smaller callback functions. (For
more information on coroutines, see `Wikipedia's treatment of the subject
<http://en.wikipedia.org/wiki/Coroutine>`_.)
Any function decorated with coroutine will return an InProgress object, and the
caller can connect a callback to the InProgress object in order to be notified
of its return value or any exception.
When a coroutine yields kaa.NotFinished, control is returned to the
main loop, and the coroutine will resume after the yield statement
at the next main loop iteration, or, if an interval is provided with the
decorator, after this time interval. Following the cooperative multitasking
analogy, yielding kaa.NotFinished can be thought of as the coroutine releasing
a "time slice" so that other tasks may run.
When a coroutine yields any value other than kaa.NotFinished (including None),
the coroutine is considered finished and the InProgress returned to the caller
will be :ref:`emitted <emitting>` (i.e. it is finished). As with return, if no value is
explicitly yielded and the coroutine terminates, the InProgress is finished
with None.
There is an important exception to the above rule: if the coroutine yields an
:class:`~kaa.InProgress` object, the coroutine will be resumed when the
InProgress object is finished. This allows a coroutine to be "chained" with
other InProgress tasks, including other coroutines.
To recap, if a coroutine yields:
* `kaa.NotFinished`: control is returned to the main loop so that other tasks
can run (such as other timers, I/O handlers, etc.) and resumed on the next
main loop iteration.
* an :class:`~kaa.InProgress` object: control is returned to the main loop and
the coroutine is resumed with the yielded InProgress is finished. Inside
the coroutine, the yield call "returns" the value that InProgress was finished
with.
* any other value: the coroutine terminates, and the InProgress the coroutine
returned to the caller is finished with that value (which includes None, if
no value was explicitly yielded and the coroutine reaches the end naturally).
Here is a simple example that breaks up a loop into smaller tasks::
import kaa
@kaa.coroutine()
def do_something():
for i in range(10):
do_something_expensive()
yield kaa.NotFinished
yield 42
def handle_result(result):
print "do_something() finished with result:", result
do_something().connect(handle_result)
kaa.main.run()
A coroutine can yield other coroutines (or rather, the InProgress
object the other coroutine returns)::
@kaa.coroutine()
def do_something_else():
try:
result = yield do_something()
except:
print "do_something failed"
yield
yield True if result else False
(Note that the above syntax, in which the yield statement returns a value,
was introduced in Python 2.5. kaa.base requires Python 2.5 or later.)
Classes in kaa make heavy use of coroutines and (to a lesser extent) threads
when methods would otherwise block on some resource. Both coroutines and
:ref:`@threaded <threaded>`-decorated methods return InProgress objects and behave identically.
These can be therefore yielded from a coroutine in the same way::
@kaa.coroutine()
def fetch_page(host):
"""
Fetches / from the given host on port 80.
"""
socket = kaa.Socket()
# Socket.connect() is implemented as a thread
yield socket.connect((host, 80))
# Socket.read() and write() are implemented as single-thread async I/O.
yield socket.write('GET / HTTP/1.1\n\n')
print (yield socket.read())
In the above example, the difference between threaded functions
(:meth:`kaa.Socket.connect`) and coroutines (:meth:`~kaa.IOChannel.write` and
:meth:`~kaa.IOChannel.read`) is transparent. Both return InProgress objects. (As
an aside, we didn't really need to yield socket.write() because writes are
queued and written to the socket when it becomes writable. However, yielding a
write means that when the coroutine resumes, the data has been
written.)
To more clearly see the benefit of implementing the above example as a coroutine,
consider the following code, which is rewritten using the more traditional approach
of connecting callbacks at the various stages of the task::
def fetch_page(host):
socket = kaa.Socket()
socket.connect((host, 80)).connect(finished_connect, socket)
def finished_connect(result, socket):
socket.write('GET / HTTP/1.1\n\n').connect(finished_write, socket)
def finished_write(len, socket):
socket.read().connect(finished_read)
def finished_read(data):
print data
In practice then, coroutines can be seen as an alternative approach to the
classic signal/callback pattern, allowing you to achieve the same logic but
with a much more intuitive and readable code. This means that if you design
your application to use signals and callbacks, it might not be clear where
coroutines would be useful.
However, if you make use of the asynchronous plumbing that kaa offers early on
in your design -- and that includes healthy use of :class:`~kaa.InProgress`
objects, either explicitly or implicitly through the use of the @coroutine
and :ref:`@threaded <threaded>` decorators -- you should find that you're able
to produce some surprisingly elegant, non-trivial code.
Decorator
=========
.. autofunction:: kaa.coroutine
|
