Tkinter wait_variable problem: hangs at termination

[Russell E. Owen]

I want to support execution of simple user-written scripts in a Tkinter
application. The scripts should be able to wait for data and such
without hanging the GUI (and without having to write the script as a
bunch of asynchronously called subroutines).

I decided to use Tkinter's wait_variable. I built a "script runner"
object that has suitable wait methods. Internally each of these methods
registers a callback that sets a variable when the wait condition is
satisfied, then calls wait_variable to wait until the variable is set.

The resulting scripts are simple and seem to work well, e.g.:

myscript(sr):
# do some normal python stuff that is fast
sr.waitForData(...)
# more normal fast python stuff
sr.waitForMS(time)
# etc.

Unfortunately, if a user closes the root window while wait_variable is
waiting, the application never fully quits. On my unix box the root
window closes but the command-line prompt never returns and ^C is
ignored. The process isn't using excess cpu cycles; it's just not
listening.

I have tried registering an atexit handler and adding a __del__ method
so that the variable being waited on is toggled at shutdown, but it
makes no difference.

Here is an example:
<ftp://ftp.astro.washington.edu/pub/users/rowen/python/WaitBug.py>
Press "Start" to start the script (which simply prints a number to
sys.stdout every second, then quits. Close the root window while the
script is waiting to print the next number, or after pausing the script,
and you'll see the problem.

Any suggestions for how to avoid/work around this problem? If it's a
Tkinter bug I'll report it.

-- Russell

P.S. I saw that some perl/Tk users got around a different problem by
faking wait_variable by running a tight "do one event, see if the wait
condition is satisfied" loop.

I don't think Tkinter allows one to execute a single event, and I
suspect it would be inefficient even if it were possible. I might be
able to use update (though the nice thing about wait_variable is most of
the action happens down at the tcl/tk level, presumably making it
maximally efficient).

(This is one of those rare times I wish Tkinter worked at the C level
the way perl's tk interface does.)

 

[Peter Otten]

<veröffentlicht & per Mail versendet>

I want to support execution of simple user-written scripts in a Tkinter
application. The scripts should be able to wait for data and such
without hanging the GUI (and without having to write the script as a
bunch of asynchronously called subroutines).

I decided to use Tkinter's wait_variable. I built a "script runner"
object that has suitable wait methods. Internally each of these methods
registers a callback that sets a variable when the wait condition is
satisfied, then calls wait_variable to wait until the variable is set.

The resulting scripts are simple and seem to work well, e.g.:

myscript(sr):
# do some normal python stuff that is fast
sr.waitForData(...)
# more normal fast python stuff
sr.waitForMS(time)
# etc.

Unfortunately, if a user closes the root window while wait_variable is
waiting, the application never fully quits. On my unix box the root
window closes but the command-line prompt never returns and ^C is
ignored. The process isn't using excess cpu cycles; it's just not
listening.

I have tried registering an atexit handler and adding a __del__ method
so that the variable being waited on is toggled at shutdown, but it
makes no difference.

Here is an example:
<ftp://ftp.astro.washington.edu/pub/users/rowen/python/WaitBug.py>
Press "Start" to start the script (which simply prints a number to
sys.stdout every second, then quits. Close the root window while the
script is waiting to print the next number, or after pausing the script,
and you'll see the problem.

Any suggestions for how to avoid/work around this problem? If it's a
Tkinter bug I'll report it.

-- Russell

P.S. I saw that some perl/Tk users got around a different problem by
faking wait_variable by running a tight "do one event, see if the wait
condition is satisfied" loop.

I don't think Tkinter allows one to execute a single event, and I
suspect it would be inefficient even if it were possible. I might be
able to use update (though the nice thing about wait_variable is most of
the action happens down at the tcl/tk level, presumably making it
maximally efficient).

(This is one of those rare times I wish Tkinter worked at the C level
the way perl's tk interface does.)

Without any deeper insight in your script - could the following meet your
needs?

# instead of atexit.register():
def dw():
self.cancel()
self._tk.destroy()
self._tk.protocol("WM_DELETE_WINDOW", dw)

and further down:

root = Tkinter.Tk()
ScriptRunner._tk = root

That way your runner would get notified if the window shall be closed.

Peter

 

[Russell E. Owen]

<veröffentlicht & per Mail versendet>

Without any deeper insight in your script - could the following meet your
needs?

# instead of atexit.register():
def dw():
self.cancel()
self._tk.destroy()
self._tk.protocol("WM_DELETE_WINDOW", dw)

and further down:

root = Tkinter.Tk()
ScriptRunner._tk = root

That way your runner would get notified if the window shall be closed.

This fix does, indeed work! Also, there is an even easier solution: it
turns out to be sufficient to bind to <Destroy>. The callback sets the
variable being waited on and the application no longer hangs.

However, wait_variable proved to unsuitable because if multiple users
use wait_variable, they stack up. If a new routine executes
wait_variable while an existing one is waiting, the new routine has to
fully finish executing before the first one resumes. This is
understandable in hindsight, but disappointing. As a result:
- only one script could be run at a time
- if wait_variable is used elsewhere (and some dialog boxes are reported
to do so) then that would also pause an executing script

I ended up implementing user-written scripts as generators, instead. The
user's script ends up looking like this:
def myscript(sr):
yield sr.doCmd(...)
...
yield sr.waitMS(...)

where sr is a ScriptRunner object. sr's doCmd, waitMS etc. set up a
termination condition that causes the script generator's next() method
to be executed, until the script is finished.

This works fully normally with Tk's event loop. No wait_variable magic
is involved and multiple scripts can peacefully run at the same time.

The big problem is that users are likely to forget the "yield"
statement. To help with this I increment one counter each time a wait
subroutine begins and another counter each time the script generator is
executed. Any discrepancy indicates a missing yield.

-- Russell