Hi David, Since selectors.py is not technically part of asyncio, it is best to do this as a patch to the CPython standard library. (If it was asyncio only I would recommend writing the patch for the upstream asyncio project, i.e. tulip). The recommended way to contribute to CPython is to create an issue in the tracker at bugs.python.org with a reference to this email thread, and then to upload your patch to that issue. You can then send me an email with the link (don't worry, it won't get stuck in my spam filter and I won't ignore you). You should try to generate the patch for the Python 3.5 (== default) branch in the Python Mercurial repo. More info about contributing is at https://docs.python.org/devguide/. Looking forward to your patch!
--Guido On Thu, Sep 25, 2014 at 11:31 AM, David Jander <djan...@gmail.com> wrote: > > Dear Guido, > > Thanks for your fast reply. I have never contributed to the Python project > before. A quick look at the relevant modules seems to indicate that I'll > probably have to touch quite a few classes in both the selectors and > asyncio modules, but the changes should be quite simple. The case for > asyncore was (obviously) a lot simpler... > > Here's one of the examples where I've extended asyncore, to illustrate > what I mean: > > https://github.com/yope/grunner/blob/master/gpio.py > > The original code as contained in the ayncore-library would put any read- > or write file-descriptor also into the exception set, but I needed a way to > add it _only_ to exception and not any of the two others. The special file > /sys/class/gpio/<name>/value in Linux is _always_ ready for > reading/writing, but will generate an "exception" event that makes > select.select return with the FD added to the third list (called xlist) or > select.poll with POLLPRI/POLLERR if the content changes according to the > trigger set in the /sys/class/gpio/<name>/edge. For embedded systems (and > obviously for the raspberry-pi and similar kits) this is actually pretty > nifty functionality. > > This functionality should also be useful for certain TCP/IP socket code > and AFAIK also for some V4L devices, although I have not seen it used in > python-code. > > At my work, we are currently deciding about moving python-2.7 based code > that uses asyncore in a similar way to python-3.4 with asyncio, and knowing > that there are chances we can get this change into mainline Python with > blessings from the master himself, could actually make this transition > happen :-) > > Expect to see patches from either me or someone else once we start the > porting work. Any hint about the proper procedure in submitting for a > first-time contributor is appreciated. > > Best regards, > > > 2014-09-25 18:33 GMT+02:00 Guido van Rossum <gu...@python.org>: > >> It's totally fine to submit patches for this. I have never seen a use >> case for this in my life, but I am glad that you have one! >> >> On Thu, Sep 25, 2014 at 8:05 AM, David Jander <djan...@gmail.com> wrote: >> >>> >>> Hi all, >>> >>> I have been using asyncore until now, and am very excited about the new >>> asyncio library, but there is one thing I have always missed on asyncore >>> (although it was halfway there) which does not seem to be possible with >>> asyncio (please correct me if I'm wrong): >>> >>> On UNIX (Linux) the selectors module is specially powerful, since not >>> only sockets can be used, but this extra power is crippled by the fact that >>> there is no support (apparently) for select.select "exceptional condition" >>> lists, or for select.poll "POLLPRI or POLLERR" events. I can imagine that >>> the use-cases of these are a big mystery to many and probably few people >>> actually know what they're for. >>> >>> When using asyncore I usually extend the module by replacing the poll() >>> function with my own version that supports "exceptional condition"-only >>> dispatchers. These are very useful when dealing with direct character- and >>> sysfs-device access on (embedded-)Linux systems for example. >>> >>> One such example is the Linux GPIO sysfs interface. (see >>> https://www.kernel.org/doc/Documentation/gpio/sysfs.txt for more info). >>> Via this interface, one can manipulate digital input/output signals >>> directly from user-space. One exceptionally (no pun intended) cool feature >>> is the possibility of using inputs to trigger "software interrupts" in >>> user-space via select() or poll(). >>> >>> Probably this functionality would have to be implemented in the >>> selectors module first by means of support for EVENT_EXCEPTION in addition >>> to EVENT_READ and EVENT_WRITE. >>> >>> What about adding support for this to python selectors and asyncio? >>> Would patches for something like this be accepted or is this out of the >>> question? >>> >>> Best regards, >>> >>> >> >> >> -- >> --Guido van Rossum (python.org/~guido) >> > > > > -- > David Jander > -- --Guido van Rossum (python.org/~guido)
