Just a thought... if in the event loop we would use weakrefs for sockets. Do we ever have to handle CancelledError? Would the socket automatically be closed by the garbage collector if there is no reference anymore to the transport, and because of that also no reference to the socket anymore?
I notice that cancellation can be hard to get right in my own projects, but it's sometimes too easy to forget about it, leading to weird bugs. Le samedi 5 avril 2014 18:49:39 UTC+2, Guido van Rossum a écrit : > > I believe Victor already put a patch up for review, although I cannot find > the issue right now. IIRC I even reviewed it -- but no search keywords I > try show it, only the (AFAIK derived) http://bugs.python.org/issue21155 > > > On Sat, Apr 5, 2014 at 1:24 AM, Jonathan Slenders > <[email protected]<javascript:> > > wrote: > >> How much time do we have for 3.4.1? >> >> I think in this case I think it should be a catch of CancelledError, not >> a finally. On success, we don't want to close the socket when the coroutine >> terminates. Do we? >> >> >> >> Le jeudi 3 avril 2014 19:33:24 UTC+2, Guido van Rossum a écrit : >> >>> On Thu, Apr 3, 2014 at 4:16 AM, Jonathan Slenders <[email protected] >>> > wrote: >>> >>>> By blocking, I meant "not progressing anymore". Not that is blocks the >>>> thread. >>>> >>>> >>>> Should we catch CancelledError here at this line? >>>> https://code.google.com/p/tulip/source/browse/asyncio/ >>>> base_events.py#410 >>>> >>>> >>> Or use a finally block. There are probably some other places where we >>> don't clean up when unexpected exceptions strike. This was a good catch, >>> Jonathan! We should make sure this is fixed in as many places as possible >>> before 3.4.1 goes out. >>> >>> >>>> >>>> Le jeudi 3 avril 2014 12:10:56 UTC+2, Victor Stinner a écrit : >>>>> >>>>> Hi, >>>>> >>>>> 2014-04-03 11:34 GMT+02:00 Jonathan Slenders <[email protected]>: >>>>> > First he proposed to wrap create_connection into wait_for with a >>>>> timeout=1 >>>>> > parameter, but that would leave the file descriptor open that was >>>>> created in >>>>> > BaseEventLoop.create_connection(). Therefore he proposes to create >>>>> the >>>>> > socket ourself, using sock_connect , and if that fails due to a >>>>> timeout, >>>>> > call sock.close manually. >>>>> >>>>> When you use wait_for(), the coroutine gets a CancelledError and so >>>>> can cleanup its data (close the socket). >>>>> >>>>> If create_connection() doesn't close the socket on error, it's a bug. >>>>> Please open an issue. >>>>> >>>>> > How is it possible that loop.create_connection blocks? >>>>> >>>>> The slowest part is probably the DNS resolution. By default, >>>>> getaddrinfo() is called in a thread. >>>>> >>>>> > Doesn't it just fail after a while if the host is not reachable? >>>>> >>>>> Network operations can be slow, but create_connection() should not >>>>> "block", the coroutine should be running in background, as any "async" >>>>> task. >>>>> >>>>> It would help to identify exactly which Python line hangs. >>>>> >>>>> Victor >>>>> >>>> >>> >>> >>> -- >>> --Guido van Rossum (python.org/~guido) >>> >> > > > -- > --Guido van Rossum (python.org/~guido) >
