what is happening is that a file gets announced through eio_model
listing code, at this point of time, the monitor does not yet know about
the file. If the file now gets deleted between the annoncing and the
learning of the file from the monitor, then the file got an ADD event,
but no DEL event. Which is a bug.
With this commit there is a new API which asks the monitor if the file
already has the knowledge about the files existance, or not. A few
monitors like win32 inotify or cocoa do not have context about the file
directly, if the OS is now having the same bug, then we are again in
trouble, however, we canot do anything about that. In the case of kevent
or poll, this asks the context of the monitor if the file is already
there.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10006
Summary:
when a monitor is destroyed, it unconditionally and immediately deletes
the monitor struct. this means that as soon as the monitor is dead, the
backend must never access the parent pointer again if its lifetime exceeds
the lifetime of the monitor (such as in threads)
the only member of the monitor data used by the fallback monitor is the
monitor path, so we can just copy it to the fallback data to avoid ever
needing to dereference this pointer
fixes reliability issues with efl sentry unit tests
@fix
Depends on D9708
Reviewers: cedric
Reviewed By: cedric
Subscribers: #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9709
Summary:
we need to to this here, otherwise we effectfily start the monitor
later, which means, we can miss a change in the filesystem. However,
this makes things a lot slower.
Reviewers: cedric
Reviewed By: cedric
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D8526
it seems that this should have been a pop call, not the function
pointer. This key eina_array_pop is never used. Nor would the array get
smaller, so i assume this should be a pop call.
Differential Revision: https://phab.enlightenment.org/D6814
emitting events if the delete_me flag is set may result in events being emitted
for an already-freed monitor, resulting in both invalid memory access and a
deadlock later on if eio_shutdown has been called at this point
this causes the monitoring thread to check the status of the backend during the
block where the main loop and thread are in sync, avoiding any data races which
could occur when checking the flag at another time, and also avoiding accessing
the internals of the Ecore_Thread which could also have been deallocated during
shutdown
fix T7086
Differential Revision: https://phab.enlightenment.org/D6449
the fallback method of calling stat() on the monitored paths does not allow
for various eio events to be emitted, meaning that any application which relies
on those events can never receive them
this provides a method for checking a monitor to determine which functionality
is available, and also provides more explicit documentation regarding events
that are not provided by fallback monitoring
this method is marked as beta
@feature
Differential Revision: https://phab.enlightenment.org/D6447
for some reason, the fallback thread would exit -> create timer ->
create idler -> create thread; the existence of the idler makes little
sense and introduces variability in the timer interval
@fix
Differential Revision: https://phab.enlightenment.org/D6441
threads should not be waited on here during shutdown since these same
threads may be waiting on the main loop anyway
instead, perform as much deallocation as possible,
mark the monitor as deleted, and then set the thread to canceled and
allow the thread to clean itself up during its cancel/end callback
@fix
Differential Revision: https://phab.enlightenment.org/D6440
Summary:
the default value for the fallback poll monitor timer interval is 60.0 seconds,
which is not useful for all cases, such as CI, where we don't care about cpu
usage and just want things to process as fast as possible at all times
this enables setting the interval to any value, ensuring that any existing
timers are modified to use that value immediately
@feature
Reviewers: stefan_schmidt, bu5hm4n, raster, devilhorns, ManMower
Reviewed By: bu5hm4n, ManMower
Subscribers: ManMower, raster, bu5hm4n, cedric, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D6247
Summary: The current poll backend code uses only filename when reporting changes, while eio tests (and presumably other backends) use absolute path. This makes poll backends behavior more consistent with other backends.
Test Plan: Ran eio test suite with poll backend.
Reviewers: cedric
Reviewed By: cedric
Subscribers: cedric
Projects: #efl
Differential Revision: https://phab.enlightenment.org/D3083
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Summary: Implement basic kqueue/kevent backend for eio. When it comes to tracking directory changes, this backend falls back to the polling one.
Test Plan: Ran Enlightenment for several days and some other EFL apps without any issue.
Reviewers: cedric
Reviewed By: cedric
Subscribers: cedric
Projects: #e_on_freebsd, #efl
Differential Revision: https://phab.enlightenment.org/D2983
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
this patch adds an implementation of eio_monitor based on FSEvent
for OSX. This implentation has some limitations compared to inotify
implementation. Folowing events are not detected:
- EIO_MONITOR_FILE_CLOSED
- EIO_MONITOR_SELF_RENAME
- EIO_MONITOR_SELF_DELETED
It should be noted that some events that happend before the call
to eio_monitor_add can be catched. This is why sleep timers have
been added in the test suite.
Tests have been added to check uncovered scenarios.
some things might still be improved:
- self_deleted events for files might be handled by checking the
file_name manually
- self_deleted events for directories might be handled by setting
kFSEventStreamCreateFlagWatchRoot. I've noticed by doing so that
a lot more unwanted event are raised
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Another try to make inotify checks more common.
This time uses AC_CHECK_HEADERS() as for others, that already define
HAVE_SYS_INOTIFY_H, then uses that.
I still kept AM_CONDITIONAL([HAVE_INOTIFY]) because I plan to convert
ecore_file to the same, smarter, method that is used in eio (compiling
the file depending on the backend.
SVN revision: 80358