Author: tack
Date: Tue Feb 12 20:35:01 2008
New Revision: 3053
Log:
Pretty much rewrote InProgress.throw, documenting gotchas. Fixed bug where
if InProgress A is waiting on InProgress B, if B raises an exception, A's
generator wouldn't get the exception (using the python 2.5 features).
Modified:
trunk/base/src/notifier/async.py
Modified: trunk/base/src/notifier/async.py
==============================================================================
--- trunk/base/src/notifier/async.py (original)
+++ trunk/base/src/notifier/async.py Tue Feb 12 20:35:01 2008
@@ -161,26 +161,67 @@
This function should be called when the creating function is
done because it raised an exception.
"""
- # store result
+ # This function must deal with a tricky problem. See:
+ #
http://mail.python.org/pipermail/python-dev/2005-September/056091.html
+ #
+ # Ideally, we want to store the traceback object so we can defer the
+ # exception handling until some later time. The problem is that by
+ # storing the traceback, we create some ridiculously deep circular
+ # references.
+ #
+ # The way we deal with this is to pass along the traceback object to
+ # any handler that can handle the exception immediately, and then
+ # discard the traceback. A stringified formatted traceback is attached
+ # to the exception in the formatted_traceback attribute.
+ #
+ # The above URL suggests a possible non-trivial workaround: create a
+ # custom traceback object in C code that preserves the parts of the
+ # stack frames needed for printing tracebacks, but discarding objects
+ # that would create circular references. This might be a TODO.
+
self._finished = True
self._exception = type, value, tb
- self._unhandled_exception = None
- # Wake any threads waiting on us
+ self._unhandled_exception = True
+
+ # Attach a stringified traceback to the exception object. Right now,
+ # this is the best we can do for asynchronous handlers.
+ trace = ''.join(traceback.format_exception(*self._exception)).strip()
+ value.formatted_traceback = trace
+
+ # Wake any threads waiting on us. We've initialized _exception with
+ # the traceback object, so any threads that call get_result() between
+ # now and the end of this function will have an opportunity to get
+ # the live traceback.
self._finished_event.set()
- if self.exception.emit_when_handled(type, value, tb) != False:
- # No handler returned False to block us from logging the exception.
- # Set a flag to log the exception in the destructor if it is
- # not raised with get_result(). Using a nornal Python destructor
- # does not work because it would be too easy to create a circular
- # reference with InProgress and yield_execution which would result
- # in a memory leak. Since we do not need the InProgress object to
- # log the unhandled exception, a weakref callback is used. So this
- # is __del__ without using __del__.
- trace =
''.join(traceback.format_exception(*self._exception)).strip()
+ if self.exception.count() == 0:
+ # There are no exception handlers, so we know we will end up
+ # queuing the traceback in the exception signal. Set it to None
+ # to prevent that.
+ tb = None
+
+ if self.exception.emit_when_handled(type, value, tb) == False:
+ # A handler has acknowledged handling this exception by returning
+ # False. So we won't log it.
+ self._unhandled_exception = None
+
+ if self._unhandled_exception:
+ # This exception was not handled synchronously, so we set up a
+ # weakref object with a finalize callback to a function that
+ # logs the exception. We could do this in __del__, except that
+ # the gc refuses to collect objects with a destructor. The weakref
+ # kludge lets us accomplish the same thing without actually using
+ # __del__.
+ #
+ # If the exception is passed back via get_result(), then it is
+ # considered handled, and it will not be logged.
cb = Callback(InProgress._log_exception, trace)
self._unhandled_exception = _weakref.ref(self, cb)
+ # Remove traceback from stored exception. If any waiting threads
+ # haven't gotten it by now, it's too late.
+ self._exception = type, value, None
+
# cleanup
self.disconnect_all()
self.exception.disconnect_all()
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Freevo-cvslog mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freevo-cvslog
