qubes-installer-qubes-os/docs/threads.txt
2011-01-18 04:24:57 -05:00

107 lines
3.5 KiB
Plaintext

Threads in anaconda? No!
David Cantrell <dcantrell@redhat.com>
INTRODUCTION
Threads make a lot of people run screaming. That's entirely
understandable because thread concurrency can be a pain. In this short
document, I want to explain why threads are in anaconda and how to work
with them in the code.
MISCONCEPTIONS
Just to make sure everyone is on the same page, threads are similar to
processes. The big advantage we get is easier shared data structures.
Threads can communicate over more methods than just signals. But,
multithreaded does not mean that we are taking every operation out to
a separate thread.
ANACONDA THREADS
The idea for anaconda threads is to simplify life for things that can
or need to run parallel to other operations. So we will reserve the
use of threads for tasks that fit in to this category well (the logging
system, for instance) and keep the bulk of the installer in the main
thread.
THREADS AND PYTHON
Python has a nice model for threading. Threads in Python extend the
threading.Thread class. So an easy way to identify something that will
run or can be run as a thread is seeing a class definition like this:
class SomeClass(threading.Thread):
You still have your __init__ method for the constructor, but threads
also have a run() method and a join() method (there are others, but
I will just discuss these).
The run() method is called when you start the thread. This is where
you want to do the work. Normally this happens in the class
constructor, but in threads we need that separated out to a different
method.
The join() method is to block execution of other threads. Whatever you
put in the join() method will run and other threads will be blocked
while it runs. Now, this method is only run when you call it explicitly
from another thread, so think of it as similar to waitpid().
Python has the thread and threading modules. Use threading as it's
built on top of thread and provides a threading system similar to the
POSIX thread model.
More information:
http://docs.python.org/lib/module-threading.html
THREAD NAMES
Threads have names in Python. They are automatically assigned or you
can specify the name. For anaconda it probably makes more sense to
name our threads since we won't be launching more than one for the
same task.
The convention I'm using is: NameThr
For example: RelNotesThr
The name is arbitrary, but we should probably have some sort of
consistency.
PYGTK AND THREADS
GTK+ presents the biggest challenge for threads, but it's not
impossible. We will be allowing GTK+ calls from any thread, so we
have to call threads_init() in gui.py as the first thing:
gtk.gdk.threads_init()
After this, you can use Python threads as you normally would. When
you call gtk.main(), you need to call it like this:
gtk.threads_enter()
gtk.main()
gtk.threads_leave()
Calls to PyGTK methods or fiddling with GTK objects...all that has to
be wrapped in threads_enter/threads_leave calls. The suggested syntax
is:
gtk.threads_enter()
try:
# do stuff
finally:
gtk.threads_leave()
Suggested reading:
http://www.async.com.br/faq/pygtk/index.py?req=show&file=faq20.006.htp
http://developer.gnome.org/doc/API/2.0/gdk/gdk-Threads.html
I will try to expand this document as I move through more threading code.
Email me if you have any questions.
--
David Cantrell
<dcantrell@redhat.com>