In the class `logging.handlers.QueueListener` the `_sentinel` is `None`. But >>> a = None >>> b = None >>> a is b True Because of this, the QueueListener stops if it receives a `None`. Were the sentinel a proper instance, like: _sentinel = {} This would not happen.
Why do you think None is a valid value to send for normal operation? Since the queue is only meant for sending logging events (records), it seems reasonable to send None as a sentinel - it should never be seen during operation. The QueueListener is *supposed* to stop if it sees a None, so don't send those over the queue - just send logging events - until you want to shut the listener down. You can of course set the sentinel to whatever you want, in your own code, and then send that. The default value of None seems OK to me. So I don't believe this is a valid issue - marking as such and as pending, and I will close this shortly unless you come back with some more information.
What mislead me is: The current code uses `is` and opposed to `==` which I assume is for the very specific reason to match identity and not value. The sentinel starts with a _, which to a casual reader (me) suggests that it's a private implementation detail that I should not have to touch. (am I right on this?) http://plumberjack.blogspot.co.uk/2010/09/improved-queuehandler-queuelistener.html In the introduction of this very same class you are also mentioning that: "... QueueListener is not even especially logging-specific: You can pass it as a handler any object that has a handle method which takes a single argument, and that method will be passed any non-sentinel object which appears on the queue." also "You should be able to paste QueueHandler and QueueListener into your own code..." So suddenly logging in not the only documented use case. And yes, I can override the sentinel in my subclass, yet it is named as `_sentinel`, which again suggests "do not touch" to me. It's a tiny inconsistency that may never come up again for anyone else, but I just used a copy of the class and it came up for me. My tests caught it, I fixed it up for my own use-case, case closed. So honestly it's not the end of the world, but neither is changing it to a safer default, like `{}` or even `self`. I won't be bothering you with this issue anymore, so feel free to close it if you want.
> The sentinel starts with a _, which to a casual reader (me) suggests > that it's a private implementation detail that I should not have to > touch. (am I right on this?) Python is a language for consenting adults, so nothing is off-limits, except that you need to know what you are doing when you make changes to internal attributes, and may not get support from the original author if something breaks. However, notice that I set it up as a class value which could be overridden at an instance level, rather than hard-coding None into the sentinel test. So, it *was* intended to be changed if needed, but it's more appropriate for a subclass to do that than a user of the class (not that a user is *forbidden* - that couldn't be enforced, anyway).
History
Date
User
Action
Args
2022-04-11 14:57:39
admin
set
github: 60875
2012-12-13 09:26:07
vinay.sajip
set
status: open -> closedmessages: +
2012-12-13 09:05:55
Andras.Szalai
set
status: pending -> openmessages: +
2012-12-13 00:33:58
vinay.sajip
set
status: open -> pendingversions: - Python 3.1messages: + assignee: vinay.sajipresolution: not a bug