Re: [python-tulip] Event loop error handler
Oops, it was pointed out to me that this message isn't super clear. To put it more simply: Please go ahead and do this, using the loop API everywhere would be a lot better for my uses :-). -glyph On Feb 10, 2014, at 5:12 PM, Glyph gl...@twistedmatrix.com wrote: I have a use-case, which is to direct Twisted's logging to Tulip's (or vice versa) in a way which behaves consistently. Using the logger directly is, as I said in a message a long time ago, a really wide interface for alternate logging systems to try to use. (And stdlib logging.py is slow enough that it can seriously start to impact your frame rate if you're trying to do logging from a GUI, so there's still a pretty substantial use-case for Twisted's logging client-side. This is probably less true than it was 10 years ago when I really cared deeply about this for my day job, but nevertheless, I still dream of making some cool games in Python ;-).) On Feb 7, 2014, at 4:18 PM, Guido van Rossum gu...@python.org wrote: OK, go for it unless Glyph pipes up. :-) On Fri, Feb 7, 2014 at 2:03 PM, Yury Selivanov yselivanov...@gmail.com wrote: On 2/7/2014, 4:47 PM, Guido van Rossum wrote: On Fri, Feb 7, 2014 at 1:37 PM, Yury Selivanovyselivanov...@gmail.comwrote: On 2/7/2014, 3:52 PM, Guido van Rossum wrote: Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. Sure. Another question: logger.exception is also used in: - selector_events.py: in _accept_connection, in case of errors in pause_writing/resume_writing and _fatal_error - proactor_events.py: in case of failed accept, _fatal_error and errors in pause/resume writing - unix_events.py: In pipe transport's _fatal_error, in case of exception in SIGCHLD handler - windows_events.py: pipe accept failed All of the above sites are logging exceptions (typically OSErrors). Should we use the loop exception API there, or you want to keep using loggers directly? I've got a hunch saying that every place where we log an exception we should use the new handler, but TBH I don't have much of a use case -- I've been very happy with the default logging and I would probably just be configuring the logger rather than overriding the exception handler. I think it's time to ask Glyph for a use case that can't be dealt with by overriding the logger. I'm +1 on using the loop API everywhere. It just gives us more control and flexibility in improving/working with error reporting later down the road. Using logger directly feels inconsistent. Yury -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On 2/6/2014, 11:02 PM, Guido van Rossum wrote: On Thu, Feb 6, 2014 at 2:54 PM, Yury Selivanovyselivanov...@gmail.comwrote: A question to you and Guido: should the callback receive three arguments: loop, exception, and context, where context is a namedtuple (so we can add new fields to it in future releases). For now, 'context' can have two fields: 'message' and 'callback'. The former is the log message, and the latter is the failed callback function (or None). Hm... I don't like to have this much structure, but I do know that it's a pain to extend a callback API with new information, so I think it's a good idea. But maybe apart from the loop everything should be passed this way (i.e. also the exception)? And may I suggest to make it a dict instead of a named tuple? It's easier to check a dict for the presence of something that may or may not exist. While working on the patch I encountered the following problem: event loop may not always be available, if we decide to use the new error handling API for futures._TracebackLogger.__del__ and 'futures.Future.__del__. And I think that those errors should go through the new API. Should we move the 'loop' callback parameter to the 'context' dict? And make it optional too? Yury
Re: [python-tulip] Event loop error handler
Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. On Fri, Feb 7, 2014 at 11:17 AM, Yury Selivanov yselivanov...@gmail.comwrote: On 2/6/2014, 11:02 PM, Guido van Rossum wrote: On Thu, Feb 6, 2014 at 2:54 PM, Yury Selivanovyselivanov...@gmail.com wrote: A question to you and Guido: should the callback receive three arguments: loop, exception, and context, where context is a namedtuple (so we can add new fields to it in future releases). For now, 'context' can have two fields: 'message' and 'callback'. The former is the log message, and the latter is the failed callback function (or None). Hm... I don't like to have this much structure, but I do know that it's a pain to extend a callback API with new information, so I think it's a good idea. But maybe apart from the loop everything should be passed this way (i.e. also the exception)? And may I suggest to make it a dict instead of a named tuple? It's easier to check a dict for the presence of something that may or may not exist. While working on the patch I encountered the following problem: event loop may not always be available, if we decide to use the new error handling API for futures._TracebackLogger.__del__ and 'futures.Future.__del__. And I think that those errors should go through the new API. Should we move the 'loop' callback parameter to the 'context' dict? And make it optional too? Yury -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On 2/7/2014, 3:52 PM, Guido van Rossum wrote: Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. Sure. Another question: logger.exception is also used in: - selector_events.py: in _accept_connection, in case of errors in pause_writing/resume_writing and _fatal_error - proactor_events.py: in case of failed accept, _fatal_error and errors in pause/resume writing - unix_events.py: In pipe transport's _fatal_error, in case of exception in SIGCHLD handler - windows_events.py: pipe accept failed All of the above sites are logging exceptions (typically OSErrors). Should we use the loop exception API there, or you want to keep using loggers directly? And one more, aesthetic question: the currently agreed on signature of exception handlers is '(loop, context)'. I have a method BaseEventLoop.default_exception_handler(self, context), but when the method is bound, its signature is just '(context)', hence, this bound method cannot be passed to 'set_exception_handler'. Should the signature be BaseEventLoop.default_exception_handler(self, loop, context), or can I just make it a staticmethod with '(loop, context)'? Yury
Re: [python-tulip] Event loop error handler
On Fri, Feb 7, 2014 at 1:37 PM, Yury Selivanov yselivanov...@gmail.comwrote: On 2/7/2014, 3:52 PM, Guido van Rossum wrote: Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. Sure. Another question: logger.exception is also used in: - selector_events.py: in _accept_connection, in case of errors in pause_writing/resume_writing and _fatal_error - proactor_events.py: in case of failed accept, _fatal_error and errors in pause/resume writing - unix_events.py: In pipe transport's _fatal_error, in case of exception in SIGCHLD handler - windows_events.py: pipe accept failed All of the above sites are logging exceptions (typically OSErrors). Should we use the loop exception API there, or you want to keep using loggers directly? I've got a hunch saying that every place where we log an exception we should use the new handler, but TBH I don't have much of a use case -- I've been very happy with the default logging and I would probably just be configuring the logger rather than overriding the exception handler. I think it's time to ask Glyph for a use case that can't be dealt with by overriding the logger. And one more, aesthetic question: the currently agreed on signature of exception handlers is '(loop, context)'. I have a method BaseEventLoop.default_exception_handler(self, context), but when the method is bound, its signature is just '(context)', hence, this bound method cannot be passed to 'set_exception_handler'. Should the signature be BaseEventLoop.default_exception_handler(self, loop, context), or can I just make it a staticmethod with '(loop, context)'? But why would you want to pass the default handler to set_exception_handler()? I had expected that there would be something like this: def call_exception_handler(self, exception, context): if self._exception_handler is not None: try: self._exception_handler(self, exception, context) return except Exception as exc: self.default_exception_handler(exc, {'message': 'custom exception handler failed'}) self.default_exception_handler(exception, context) -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On 2/7/2014, 4:47 PM, Guido van Rossum wrote: And one more, aesthetic question: the currently agreed on signature of exception handlers is '(loop, context)'. I have a method BaseEventLoop.default_exception_handler(self, context), but when the method is bound, its signature is just '(context)', hence, this bound method cannot be passed to 'set_exception_handler'. Should the signature be BaseEventLoop.default_exception_handler(self, loop, context), or can I just make it a staticmethod with '(loop, context)'? But why would you want to pass the default handler to set_exception_handler()? I had expected that there would be something like this: def call_exception_handler(self, exception, context): if self._exception_handler is not None: try: self._exception_handler(self, exception, context) return except Exception as exc: self.default_exception_handler(exc, {'message': 'custom exception handler failed'}) self.default_exception_handler(exception, context) That's almost exactly the code I currently have. And the question is rather rhetorical: i.e. if it's OK to pass loop.default_exception_handler to loop.set_exception_handler [1] or it shouldn't be used that way, always use loop.set_exception_handler(None) [2]. I like the [2], and that's how I have it implemented now, but wanted to confirm it with the list. Yury
Re: [python-tulip] Event loop error handler
OK, don't change it. :-) On Fri, Feb 7, 2014 at 1:53 PM, Yury Selivanov yselivanov...@gmail.comwrote: On 2/7/2014, 4:47 PM, Guido van Rossum wrote: And one more, aesthetic question: the currently agreed on signature of exception handlers is '(loop, context)'. I have a method BaseEventLoop.default_exception_handler(self, context), but when the method is bound, its signature is just '(context)', hence, this bound method cannot be passed to 'set_exception_handler'. Should the signature be BaseEventLoop.default_exception_handler(self, loop, context), or can I just make it a staticmethod with '(loop, context)'? But why would you want to pass the default handler to set_exception_handler()? I had expected that there would be something like this: def call_exception_handler(self, exception, context): if self._exception_handler is not None: try: self._exception_handler(self, exception, context) return except Exception as exc: self.default_exception_handler(exc, {'message': 'custom exception handler failed'}) self.default_exception_handler(exception, context) That's almost exactly the code I currently have. And the question is rather rhetorical: i.e. if it's OK to pass loop.default_exception_handler to loop.set_exception_handler [1] or it shouldn't be used that way, always use loop.set_exception_handler(None) [2]. I like the [2], and that's how I have it implemented now, but wanted to confirm it with the list. Yury -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On 2/7/2014, 4:47 PM, Guido van Rossum wrote: On Fri, Feb 7, 2014 at 1:37 PM, Yury Selivanovyselivanov...@gmail.comwrote: On 2/7/2014, 3:52 PM, Guido van Rossum wrote: Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. Sure. Another question: logger.exception is also used in: - selector_events.py: in _accept_connection, in case of errors in pause_writing/resume_writing and _fatal_error - proactor_events.py: in case of failed accept, _fatal_error and errors in pause/resume writing - unix_events.py: In pipe transport's _fatal_error, in case of exception in SIGCHLD handler - windows_events.py: pipe accept failed All of the above sites are logging exceptions (typically OSErrors). Should we use the loop exception API there, or you want to keep using loggers directly? I've got a hunch saying that every place where we log an exception we should use the new handler, but TBH I don't have much of a use case -- I've been very happy with the default logging and I would probably just be configuring the logger rather than overriding the exception handler. I think it's time to ask Glyph for a use case that can't be dealt with by overriding the logger. I'm +1 on using the loop API everywhere. It just gives us more control and flexibility in improving/working with error reporting later down the road. Using logger directly feels inconsistent. Yury
Re: [python-tulip] Event loop error handler
OK, go for it unless Glyph pipes up. :-) On Fri, Feb 7, 2014 at 2:03 PM, Yury Selivanov yselivanov...@gmail.comwrote: On 2/7/2014, 4:47 PM, Guido van Rossum wrote: On Fri, Feb 7, 2014 at 1:37 PM, Yury Selivanovyselivanov...@gmail.com wrote: On 2/7/2014, 3:52 PM, Guido van Rossum wrote: Can't you add a reference to the loop to the tb logger object? The loop should outlive any futures anyway (since the future has a reference to the loop) and it shouldn't be a ref cycle. Sure. Another question: logger.exception is also used in: - selector_events.py: in _accept_connection, in case of errors in pause_writing/resume_writing and _fatal_error - proactor_events.py: in case of failed accept, _fatal_error and errors in pause/resume writing - unix_events.py: In pipe transport's _fatal_error, in case of exception in SIGCHLD handler - windows_events.py: pipe accept failed All of the above sites are logging exceptions (typically OSErrors). Should we use the loop exception API there, or you want to keep using loggers directly? I've got a hunch saying that every place where we log an exception we should use the new handler, but TBH I don't have much of a use case -- I've been very happy with the default logging and I would probably just be configuring the logger rather than overriding the exception handler. I think it's time to ask Glyph for a use case that can't be dealt with by overriding the logger. I'm +1 on using the loop API everywhere. It just gives us more control and flexibility in improving/working with error reporting later down the road. Using logger directly feels inconsistent. Yury -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On Thu, Feb 6, 2014 at 8:53 AM, Yury Selivanov yselivanov...@gmail.comwrote: As discussed with Guido in issue #80 (https://code.google.com/p/tulip/issues/detail?id=80), I'm proposing a strawman design for having a user definable event loop handler for unhandled exceptions. Two new API points: - loop.set_exception_handler(callback) Sets 'callback' as a new unhandled error handler for the event loop 'loop'. The signature of callback should be '(loop, exception, context)' where: 'loop' parameter is the current event loop running; 'exception' is the unhandled exception object; 'context' is a string (str), that will be the basic context information, like unhandled exception in add_reader callback (the kind of information that usually goes to the log title). I'd rename the context arg to 'message', then you don't have to explain what it is. :-) Also I'd steer clear from calling the handler 'callback' -- maybe just name it 'exception_handler'? It's OK to call 'loop.stop()' or 'loop.call_*' methods from the handler. Cool. If a user set handler raises an exception, that exception will be handled by the default event loop exception handler (that will simply log it) Nice. - loop.restore_default_exception_handler() Restores exception handler to the default one. Default exception handler just logs the error with 'loop.logger.exception()' Maybe instead of having a separate API for this rarely-used action, this could be spelled as loop.set_exception_handler(None)? A few questions: - Do we need an API to ask the loop for its current exception handler? If so, what should it return if no handler is set explicitly? - Should we allow multiple exception handlers, or just one? (I'd prefer just one.) - Should we have a way to invoke the default handler explicitly? E.g. if the user's handler doesn't want to do anything special, they could pass it to the default handler instead of having to figure out how the default handler logs things. - Should we have a way to invoke the current handler explicitly? E.g. if there's user code that has encountered an exception and it wants that exception to be treated the same as the loop handles other exceptions. -- --Guido van Rossum (python.org/~guido)
Re: [python-tulip] Event loop error handler
On 2/6/2014, 1:24 PM, Guido van Rossum wrote: On Thu, Feb 6, 2014 at 8:53 AM, Yury Selivanov yselivanov...@gmail.comwrote: As discussed with Guido in issue #80 (https://code.google.com/p/tulip/issues/detail?id=80), I'm proposing a strawman design for having a user definable event loop handler for unhandled exceptions. Two new API points: - loop.set_exception_handler(callback) Sets 'callback' as a new unhandled error handler for the event loop 'loop'. The signature of callback should be '(loop, exception, context)' where: 'loop' parameter is the current event loop running; 'exception' is the unhandled exception object; 'context' is a string (str), that will be the basic context information, like unhandled exception in add_reader callback (the kind of information that usually goes to the log title). I'd rename the context arg to 'message', then you don't have to explain what it is. :-) Agree. Need to make it clear, though, that 'message' is not necessarily the same as 'exception.args[0]' Also I'd steer clear from calling the handler 'callback' -- maybe just name it 'exception_handler'? It's OK to call 'loop.stop()' or 'loop.call_*' methods from the handler. Cool. If a user set handler raises an exception, that exception will be handled by the default event loop exception handler (that will simply log it) Nice. - loop.restore_default_exception_handler() Restores exception handler to the default one. Default exception handler just logs the error with 'loop.logger.exception()' Maybe instead of having a separate API for this rarely-used action, this could be spelled as loop.set_exception_handler(None)? Good. Since we always need to have at least default exception handler, using None in this context seems very logical. A few questions: - Do we need an API to ask the loop for its current exception handler? I thought about this, and my reasoning for not proposing 'loop.get_exception_handler()' was to keep the API simple. Besides, let's pretend we have this API in place, and you can get the current handler. The getter will return the handler to the caller, and that handler, or callback object will make sense only to the application/code that set it in the first place. And I think in this case, it's better to force the client code to carefully use and track the calls to 'set_exception_handler()'. If so, what should it return if no handler is set explicitly? If we make the default handler a public API, then we can return a bound method. If we decide to hide it -- then 'None'. - Should we allow multiple exception handlers, or just one? (I'd prefer just one.) The API will be more complex then, as we'll need functions to remove exception handlers, at least. Applications can always set one handler that will call other functions (as many as they need). - Should we have a way to invoke the default handler explicitly? E.g. if the user's handler doesn't want to do anything special, they could pass it to the default handler instead of having to figure out how the default handler logs things. I think it's a good idea to let users trigger the default handler somehow. One option is to add it to the public API. But it may cause some confusion about when you should use it. Basically, the only place it should be used in, is a custom exception handler. Another option would be to let the custom handler simply re-raise the exception, but this way we'll mix exceptions originated *in* the handler with other exceptions, so I'd be -1 on this approach. And another option is to establish a protocol: if a custom handler returns an exception object, that object is passed to the default handler. - Should we have a way to invoke the current handler explicitly? E.g. if there's user code that has encountered an exception and it wants that exception to be treated the same as the loop handles other exceptions. -1 on this. Non-sophisticated code should just stick to the logging module. The kind of code that uses its custom error handlers, can develop its own facilities for error reporting. Yury
Re: [python-tulip] Event loop error handler
On Thu, Feb 6, 2014 at 12:09 PM, Yury Selivanov yselivanov...@gmail.comwrote: On 2/6/2014, 1:24 PM, Guido van Rossum wrote: On Thu, Feb 6, 2014 at 8:53 AM, Yury Selivanov yselivanov...@gmail.com wrote: As discussed with Guido in issue #80 (https://code.google.com/p/tulip/issues/detail?id=80), I'm proposing a strawman design for having a user definable event loop handler for unhandled exceptions. Two new API points: - loop.set_exception_handler(callback) Sets 'callback' as a new unhandled error handler for the event loop 'loop'. The signature of callback should be '(loop, exception, context)' where: 'loop' parameter is the current event loop running; 'exception' is the unhandled exception object; 'context' is a string (str), that will be the basic context information, like unhandled exception in add_reader callback (the kind of information that usually goes to the log title). I'd rename the context arg to 'message', then you don't have to explain what it is. :-) Agree. Need to make it clear, though, that 'message' is not necessarily the same as 'exception.args[0]' If it was we wouldn't need an extra argument for it. :-) Also I'd steer clear from calling the handler 'callback' -- maybe just name it 'exception_handler'? It's OK to call 'loop.stop()' or 'loop.call_*' methods from the handler. Cool. If a user set handler raises an exception, that exception will be handled by the default event loop exception handler (that will simply log it) Nice. - loop.restore_default_exception_handler() Restores exception handler to the default one. Default exception handler just logs the error with 'loop.logger.exception()' Maybe instead of having a separate API for this rarely-used action, this could be spelled as loop.set_exception_handler(None)? Good. Since we always need to have at least default exception handler, using None in this context seems very logical. A few questions: - Do we need an API to ask the loop for its current exception handler? I thought about this, and my reasoning for not proposing 'loop.get_exception_handler()' was to keep the API simple. Sure. Besides, let's pretend we have this API in place, and you can get the current handler. The getter will return the handler to the caller, and that handler, or callback object will make sense only to the application/code that set it in the first place. Not necessarily, you know its signature. And I think in this case, it's better to force the client code to carefully use and track the calls to 'set_exception_handler()'. Not necessarily. But I agree to leave this out unless and until an actual real use case presents itself. If so, what should it return if no handler is set explicitly? If we make the default handler a public API, then we can return a bound method. If we decide to hide it -- then 'None'. - Should we allow multiple exception handlers, or just one? (I'd prefer just one.) The API will be more complex then, as we'll need functions to remove exception handlers, at least. Applications can always set one handler that will call other functions (as many as they need). Right. - Should we have a way to invoke the default handler explicitly? E.g. if the user's handler doesn't want to do anything special, they could pass it to the default handler instead of having to figure out how the default handler logs things. I think it's a good idea to let users trigger the default handler somehow. One option is to add it to the public API. But it may cause some confusion about when you should use it. Basically, the only place it should be used in, is a custom exception handler. I suppose so. But this doesn't bother me much. Another option would be to let the custom handler simply re-raise the exception, but this way we'll mix exceptions originated *in* the handler with other exceptions, so I'd be -1 on this approach. Agreed. And another option is to establish a protocol: if a custom handler returns an exception object, that object is passed to the default handler. Eew. Special return values are very fragile in my experience. - Should we have a way to invoke the current handler explicitly? E.g. if there's user code that has encountered an exception and it wants that exception to be treated the same as the loop handles other exceptions. -1 on this. Non-sophisticated code should just stick to the logging module. The kind of code that uses its custom error handlers, can develop its own facilities for error reporting. I'm not sure. The default exception handler may, over time, develop some useful custom behavior, and it's not like its signature is very complicated or using it is fraught with problems. So I don't see much downside to letting users call loop.default_exception_handler(). My suggestion would be two extra event loop methods:
Re: [python-tulip] Event loop error handler
On 2/6/2014, 3:44 PM, Guido van Rossum wrote: [snip] I'm not sure. The default exception handler may, over time, develop some useful custom behavior, and it's not like its signature is very complicated or using it is fraught with problems. So I don't see much downside to letting users call loop.default_exception_handler(). My suggestion would be two extra event loop methods: loop.default_exception_handler(exception, message) loop.call_exception_handler(exception, message) The former is the default handler. The latter calls the handler you set, if you set one, otherwise the default handler. So let me reiterate on the proposed design: loop.set_exception_handler(handler) Sets 'handler' as a new unhandled exceptions handler. If 'handler' is a callable object, then it should have the following signature (or compatible one): (loop, exception, message). If 'callback' is None, default event loop exceptions handler will be set. loop.default_exception_handler(exception, message) Triggers default exceptions handler of the event loop. loop.call_exception_handler(exception, message) Triggers the exceptions handler set with 'set_exception_handler'. If there was no custom exceptions handler set, the default one is triggered. Yury
Re: [python-tulip] Event loop error handler
Hi, 2014-02-06 17:53 GMT+01:00 Yury Selivanov yselivanov...@gmail.com: I'm proposing a strawman design for having a user definable event loop handler for unhandled exceptions. With my short experience of asyncio, I can say that it's a pain to reconnect an unhandled exception to the future and the original source code. - loop.set_exception_handler(callback) Sets 'callback' as a new unhandled error handler for the event loop 'loop'. The signature of callback should be '(loop, exception, context)' ... so it would be nice to give something to retrieve the source of the exception. The minimum would be the future, task, or handle object. The new parameter may be None if it doesn't apply. I don't know why what's the best option to recover the original source code from a task. An option is to use tracemalloc to get the traceback where a task has been created. Another option is to always save the traceback where a future has been created, directly in the future. That's why I'm in favor of a debug mode for asyncio, to record the traceback in debug mode, but don't record it in production to not add any overhead. The same problem applies from callbacks (handles, call_soon/call_at/etc.). Victor
Re: [python-tulip] Event loop error handler
However, I'm not sure how passing the future/handle to the handler may help. You have the name of the failed callback in the traceback of the exception. And, in many places all you have is just the failed callback function, and no related future or task. I'm talking about Future destructor which logs the unhandled exception. The future is just self. Do you know the new function tracemalloc.get_object_traceback()? Victor
Re: [python-tulip] Event loop error handler
On 2/6/2014, 5:34 PM, Victor Stinner wrote: Install an handler for unhandled exceptions is not something new: there is already sys.displayhook and sys.excepthook. I'm not sure that it's useful to get the default handler. Just give access to the current handler. So when you setup a new handler, just call the previous if you don't know what to do with it. It is the design chosen for the new malloc API, PEP 445. So you just need two function: get and set. For sys.displayhook/excepthook, the default is stored in sys.__displayhook/excepthook__. Victor You're right, it's nothing new. However, I always hated the need to save old hook, attach new one, then at some point restore the old one. This explicit management of the old hook is almost never gives you anything good. Yes, you have an option of stacking the hooks on top of each other, but this is also rarely used. To me, having an option to reset to or get the default hook looks simpler. And, if needed, we can always add 'get_exception_handler' to the spec later, right? Yury
Re: [python-tulip] Event loop error handler
On Thu, Feb 6, 2014 at 2:54 PM, Yury Selivanov yselivanov...@gmail.comwrote: A question to you and Guido: should the callback receive three arguments: loop, exception, and context, where context is a namedtuple (so we can add new fields to it in future releases). For now, 'context' can have two fields: 'message' and 'callback'. The former is the log message, and the latter is the failed callback function (or None). Hm... I don't like to have this much structure, but I do know that it's a pain to extend a callback API with new information, so I think it's a good idea. But maybe apart from the loop everything should be passed this way (i.e. also the exception)? And may I suggest to make it a dict instead of a named tuple? It's easier to check a dict for the presence of something that may or may not exist. Another option is to have a loop method to get the current/last executed callback. Much less flexible. BTW I don't like the excepthook/displayhook API much -- the sys module has lots of objects that you are allowed to assign to, but that's a bit of a historical accident (see e.g. the recent request to make sys.std{in,out,err} thread-local). I would like to go ahead with the proposed API (set_exception_handler, default_exception_handler, call_exception_handler). I don't see much of a use case for saving/restoring these outside unittests (where the better approach is to just create a new event loop). The expected usage pattern is to set a custom handler in main() right after creating the loop. A framework that wants some kind of pass the buck behavior can build that on top of the basic API combined with a convention. -- --Guido van Rossum (python.org/~guido)