With a maxsize to the Queue of threads we are introducing a blocking call while
adding new threads.
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The problem lies in the fact that offlineimap.folder.Base's method
syncmessagesto_copy() uses threaded code everytime it is suggested by
the derived class's suggeststhreads() (currently, only IMAP does this
suggestion), but offlineimap/init.py will not spawn the
exitnotifymonitorloop() from offlineimap.threadutil.
The root cause is that ExitNotifyThread-derived threads need
offlineimap.threadutil's exitnotifymonitorloop() to be running the
cleaner for the exitthreads Queue(), because it fills the queue via
the run() method from this class: it wants to put() itself to the
Queue on exit, so when no exitnotifymonitorloop() is running, the
queue will fill up. And if this thread is an instance of
InstanceLimitedThread that hits the limit on the number of threads,
then it will hold the instancelimitedsems[] semaphore will prevent
other InstanceLimitedThread()s of the same name to pass its start()
method.
The fix is to avoid using threaded code if we're running
single-threaded.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
Obtained-from: X-Ryl669 <boite.pour.spam@gmail.com>
Replace low-level thread.get_ident() with threading.currentThread().ident.
This works both in python2.6 and python3. (thread is renamed _thread and its
direct use is not recommended)
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The huge UI rework patch removed some obscure logic and special handling of
thread exit messages. It turns out that this was in fact still needed as a
specific exit message of the SyncRunner thread signified the threatmonitor
to quit.
We will want a nicer machinery for this in the future I guess, but fix the
eternal hang on exit by reintroducing a special exit message for the
SyncRunner thread, and return from the infinite monitor loop if SyncRunner
finishes.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
All ExitNotifyThreads and InstanceLimitThreads are setDaemon(True) in their
constructor, so there is no need to do that again in the code.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
It is basically unused by now. Rework to be able to make use of it
later, no functional changes.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Logging was flawed as the output was e.g. heavily buffered and people
complained about missing log entries. Fix this by making use of the
standard logging facilities that offlineimap offers.
This is one big ugly patch that does many things. It fixes the
Blinkenlights backend to work again with the logging facilities.
Resize windows and hotkeys are still not handled absolut correctly, this
is left for future fixing. THe rest of the backends should be working fine.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Rather than setting a global threadutil/profiledir variable, we make
set_profiledir a class function that sets the class variable profiledir.
While touching theprofiledir code, add warning to the user if the
profile directory existed before.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
apply() has been deprecated since Python 2.3, and won't be working in
python 3 anymore. Use the functional equivalent throughout.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
This is a regression introduced by commit d5493fe894
[threadutil: explicitly import get_ident from thread].
The threadid attribute was wrongly removed from the ExitNotifyThread class.
Restore it.
Tested-by: Mark Foxwell <fastfret79@archlinux.org.uk>
Reviewed-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
We are missing the import of 'os' python module since commit d839be3c61
(Sat Apr 16 20:33:35 2005 +0100) which was when John switched from SVN to Git.
Happily, it help us today: we still had no feedback for this missing import,
6 years later. So, we can remove the os.exit() call safely.
That beeing said, we still don't know if the above sys.exit() was ever touched.
My guess is that it never was. Keep this (hopefully) commented statement to
ensure the thread terminate and not play too much with the Murphy's law. :-)
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The semaphorewait()/waitforthread() logic is usefull for IMAP starting
connections. We actually use it in imapserver only.
This patch removes the over-engineered factorized methods. It tend to simplify
the code by cleaning out a chain of two direct calls with no other processes.
Reviewed-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The python module thread is the low-level module we should avoid to use in favor
of threading. We still need it to support old python because Thread.ident
doesn't exist before python 2.6:
http://docs.python.org/library/threading.html#threading.Thread.ident
Make it clear we should avoid it.
Reviewed-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The SigListener class was used to queue folders that we need to sync and
to receive "resync" and "abort" signals. It was undocumented and weird
and we had to pass "siglisteners" through the whole program.
Simply do away with it, and make 2 functions in the Account() class:
set_abort_event and get_abort_event which can be used to set and check
for such signals. This way we do not need to pass siglisteners all over
the place. Tested Blinkenlights and TTYUI uis to make sure that SIGUSR1
and SIGUSR2 actually still work.
Document those signals in MANUAL.rst. They were completly undocumented.
This simplifies the code and interdependencies by passing less stuff
around. Removes an undocumented and weirdly named class.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The only reason we used it here was to do a
traceback.print_exc(StringIO()) to get a string of our traceback. But we
can simply use traceback.format_exc() which exists since python 2.4.
One less module (and it is in the way to python 3 compatability too)
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Currently the Curses code is broken. Importing offlineimap.ui.Curses
will not succeed due to cyclic imports (threadutils imports ui, but ui
wants threadutils.MultiLock). So Curses cannot be chosen.
Incidentally, the only part in the code that uses "MultiLock" is the
Curses UI, to prevent concurrent access from several threads to the
ui-internal thread list and to IO resources such as the
screen. Fortunately for these purposes we don't need a MultiLock, so we
can do away with that implementation completely. A simple RLock that
allows us to have a thread "own" a lock and makes other threads wanting
access to the resource wait until the owning thread is finished.
The MultiLock implementation looked a bit weird, so simplifying code
here is a good thing, it might well be that we fix some "hangs" that
have been reported (and that would only ever occur with the Curses UI).
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The latter is shorter and looks nicer. UIBase was a very weird class
name for something that is "user visible". We don't need to use (or
see) it from higher level code for most of the code now.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
the cProfile/profile modules are great for performance debugging. The
pure-python profile module has much more overhead though and the
cProfile module is recommended if it exists. This changes to import to
first try the cProfile module and then fall back to the profile
module. The cProfile/profiles modules are API compatible for all that
its worth...
If that does not exist we continue to complain as before.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Rather than poll our exitthread Queue in a non-blocking fashion and
always sleep for 1 second inbetween, simply call it in a blocking
fashion which will return immediately when a thread has exited. This
is somewhat faster as we don't do unnecessary sleeps after a thread
exited.
Do note that we need to specify some timeout value here (the 60 chosen
is pretty arbitary, but what the value exactly is, is not that
important, it could be any positive value) in order to make the
Queue.get() call work with SIGINT (cf
http://bugs.python.org/issue1360).
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
From Jim Pryor
The first, "cleanup.patch", cleans up a few spots that tend to throw
exceptions for me as offlineimap is exiting from a KeyboardInterrupt.
module threading should be taken out back and shot.
Condition.wait() is a busywait loop that has negative implications for
battery consumption on laptops.
Queue.get() in blocking mode blocks SIGINT from being delivered.
Argh, argh, argh.
Closes: #493801.
Rodolfo García Peñas (kix)
Nicolas Sebrecht
X-Ryl669
Sebastian Spaeth
John Goerzen