so eio_eet didnt free fiels the same as the rest of eio. it thought it
was special. it thought it could just go free() it's objects (even
though they inherit the core Eio_File object type in their structs).
everyone else ended up calling eio_file_free() EXCEPT eio_eet. so the
eio_(long)_file_set() funcs ewnded up adding this eio file to the
tracking list via eio_file_register() but never unregistered because
they really liked to just do their own thing anyway...
BAD CEDRIC! SPANKING TIME
http://33.media.tumblr.com/3422c76c33c3b9b045f623ff73e0bf8d/tumblr_mhvu61N9br1rbavngo1_500.gi
so this unifies all allocation to now use a single allocator (that
also ensures all eio async io objects are zero'd out), and a SINGLE
free path, and then it all works. no more valgrind complaints on e
shutdown/restart whilst eio "things" were used before or ... maybe
still around.
this should fix T2129
so try again "git master" after this commit.
@fix
This patch check that when a monitor is removed the FSEvent service, it is
not restarted if there is no more paths to monitor. This was generating an
error message from FSEvent.
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>
This should avoid potential crash during shutdown while some Eio thread
were still running. We are still not blocking for more than 30s, so if
an IO is blocked on a dead device, you should be fine.
When using eio_file_stat_ls() (or any other exposing Eio_File *handler),
it will not be possible to use it from the main callback due to eio
freeing data too fast from the thread.
Moved frees from _eio_file_eina_ls_heavy() to eio_async_end() to be
sure that all the calls to main callback are over.
This implied adding the Eina_Iterator from eina_file_stat_ls to
the _Eio_File_Ls structure.
Eio does need to allocate memory from the thread, transmit it to the
main loop where it is processed and later on freed. If the main loop
is to slow to handle the stream of incoming data, a huge peak of
memory can happen. Some people may want to limit the amount of memory
used at peak time and so this new infrastructure make that possible.
@feature
Fix race condition when touching/changing a (theme) file often.
An Eio_Monitor was marked as "delete_me" but the rename callback
was still called, leading to memory access to already freed
objects.
Test protocol was:
ELM_THEME=~/default.edj elementary_test &
watch touch ~/default.edj
this MIGHT fix T45 but i can't reproduce to confirm, but reading a
backtrace indicates this could have been the issue. it looks like
there is room for a dangling pointer anyway, so fix.
stable release - cherry-pick me!
Being annoyed by different types of eina critical macros - CRI, CRIT,
CRITICAL -, I concluded to unify them to one. Discussed on IRC and
finally, CRI was chosen to meet the consistency with other macros -
ERR, WRN, INF, DBG - in terms of the number of characters.
If there is any missing bits, please let me know.
- fchmod() was isolated by HAVE_CHMOD, which was always present
before... then fchmod() is also present as no errors were reported
since its introduction.
- fchmod() is POSIX for a while now.
- lstat() is POSIX for a while now.
- setxattr is supported by EFL_CHECK_FUNCS() as is used by eina.
- splice() check added to EFL_CHECK_FUNCS()
SVN revision: 81938
some functions were being checked but their IFDEF not being handled,
others (getpwnam and getgrnam) were isolated in a code that is now
Windows, then no need to check them as well as these are POSIX.
SVN revision: 81934
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