This would have forced who ever used future,none signal to manually
filter out event triggered by all the future beeing fullfiled and
disconnecting once they receive a value or are marked failed.
As discussed in the mailing list, many people will use worker threads
to execute blocking syscalls and mandating ecore_thread_check() for
voluntary preemption reduces the ecore_thread usefulness a lot.
A clear example is ecore_con usage of connect() and getaddrinfo() in
threads. If the connect timeout expires, the thread will be cancelled,
but it was blocked on syscalls and they will hang around for long
time. If the application exits, ecore will print an error saying it
can SEGV.
Then enable access to pthread_setcancelstate(PTHREAD_CANCEL_ENABLE)
via eina_thread_cancellable_set(EINA_TRUE), to pthread_cancel() via
eina_thread_cancel(), to pthread_cleanup_push()/pthread_cleanup_pop()
via EINA_THREAD_CLEANUP_PUSH()/EINA_THREAD_CLEANUP_POP() and so on.
Ecore threads will enforce non-cancellable threads on its own code,
but the user may decide to enable that and allow cancellation, that's
not an issue since ecore_thread now plays well and use cleanup
functions.
Ecore con connect/resolve make use of that and enable cancellable
state, efl_net_dialer_tcp benefits a lot from that.
A good comparison of the benefit is to run:
./src/examples/ecore/efl_io_copier_example tcp://google.com:1234 :stdout:
before and after. It will timeout after 30s and with this patch the
thread is gone, no ecore error is printed about possible SEGV.
This moved all the eoid tables, eoid lookup caches, generation count
information ad eo_isa cache into a TLS segment of memory that is
thread private. There is also a shared domain for EO objects that all
threads can access, but it has an added cost of a lock. This means
objects accessed outside the thread they were created in cannot be
accessed by another thread unless they are adopted in temporarily, or
create4d with the shared domain active at the time of creation. child
objects will use their parent object domain if created with a parent
object passed in. If you were accessing EO (EFL) objects across threads
before then this will actually now cause your code to fail as it was
invalid before to do this as no actual objects were threadsafe in EFL,
so this will force things to "fail early".
ecore_thread_main_loop_begin() and end() still work as this uses the
eo domain adoption features to temporarily adopt a domain during this
section and then return it when done.
This returns speed back to eo brining the overhead in my tests of
lookup for the elm genlist autobounce test in elementary from about
5-7% down to 2.5-2.6%. A steep drop.
This does not mean everything is perfect. Still to do are:
1. Tests in the test suite
2. Some API's to help for sending objects from thread to thread
3. Make the eo call cache TLS data to make it also safe
4. Look at other locks in eo and probably move them to TLS data
5. Make eo resolve and call wrappers that call the real method func do
recursive mutex wrapping of the given object IF it is a shared object
to provide threadsafety transparently for shared objects (but adding
some overhead as a result)
6. Test test est, and that is why this commit is going in now for wider
testing
7. Decide how to make this work with sending IPC (between threads)
8. Deciding what makes an object sendable (a sendable property in base?)
9. Deciding what makes an object shareable (a sharable property in base?)
It has been discussed on the ML (thread: "[RFC] rename efl_self") and
IRC, and has been decided we should rename it to this in order to avoid
confusion with the already established meaning of self which is very
similar to what we were using it for, but didn't have complete overlap.
Kudos to Marcel Hollerbach for initiating the discussion and
fighting for it until he convinced a significant mass. :)
This commit breaks API, and depending on compiler potentially ABI.
@feature
while in a job we do not have the safety of eo holding us alive and
when we call back the user, he may have deleted the object, releasing
both the object and its private data that we're using.
then keep an extra reference, call the methods and release it.
Efl.Object.event_callback_call no longer calls legacy smart callbacks;
calling only event callbacks registered with the given event description
pointer.
Create the method Efl.Object.event_callback_legacy_call to inherit the old
behavior from Efl.Object.event_callback_call, calling both Efl.Object events
and legacy smart callbacks.
Update all other files accordingly in order to still supply legacy
callbacks while they are necessary.
These interfaces allows generic operations on objects that can store
or provide data, such as a file or a buffer.
With well defined interfaces and events we can create code such as
Efl.Io.Copier, that will link a source with a destination and
progressively copy data as they appear.
it was silently ignoring the parent, which led me to hours trying to
investigate why my code wasn't working just to realize my parent
wasn't being assigned and refcounts were screwed.
This places the following behind beta:
- efl_quicklaunch_fallback
- efl_build_version_set
I don't think EFL_MAIN would have worked without BETA API support,
so no need to expose those for now.
@fix
in theory another libc call could overwrite errno between select
exiting and errno being used for errors. be paranoid. i know of no
real bug that this causes though.
i've fixed almost all the eina init/shutdown pairs to do the right
thing now... except one (ecore_shutdown) with comment inline where
eo_shutdown is not called. if this is called we are in crash land.
this needs further inspection.
ecore_timer_del() checks a flag "inside_call" that can be
set before calling the timer cb... but it was never reset
to 0. So, all legacy timers would keep on ticking forever
and ever, until they return CANCEL.
Anyway, I find the distinction between eo_del and
ecore_timer_del very troubling. eo_del() should work
on a legacy timer. Ping @cedric. Maybe override eo_del()?
Fixes T3898
The original idea behind knowing the app's version of EFL is not
a great story. It comes from the fact that some bugs exist in
earlier versions of EFL, and some things need to be fixed. But
those fixes may break behaviour for older apps. This patch is
opening the way to the slippery slope of bug compatibility.
Unfortunately this is a requirement if we want to be able to move
forward and not break apps when we fix bugs (behaviour or ABI).
I hope we will not need to implement too many (if any) workaround
such issues. For now, this will only be used as debugging info.
EFL_MAIN() and ELM_MAIN() will both set the app's EFL version
automatically at startup time. Some internal helpers can be added
later to check how the app build-time and run-time version of
EFL differ.
@feature
Note: this is both @class and @property. Hope that's ok for
all bindings.
This returns same as ecore_main_loop_get() (which now uses the eo
api instead).
Ping @cedric (so he can check this patch).
evas 3d examples would always exit on a double free, since
EINA_INLIST_FREE was misused. Not surprising considering
it's different from EINA_LIST_FREE but has a similar name.
On Solaris, this header is necessary for finite(). Instead of including it
if the sun compiler is used, include it if it exists. This fixes a warning
if gcc is used on Solaris
Clockid_t should be used as an opaque type. Some platform might want
to (and even do, e.g. DragonFlyBSD) declare clockid_t as an unsigned.
On such platform, testing the sign of clockid_t is never false, and
assigning it a negative value is an UB, which makes this code unlikely to
work as intended. Fixes black window on dragonfly!
Thanks to gcc for spotting this.
CC lib/ecore/lib_ecore_libecore_la-ecore_time.lo
In file included from ../src/lib/eina/Eina.h:215:0,
from lib/ecore/Ecore.h:304,
from lib/ecore/ecore_time.c:18:
lib/ecore/ecore_time.c: In function 'ecore_time_get':
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Turns out there is no PRI?SIGATOMIC in the C99 standard. Work around
that by deducing the effective integer type by comparing the
SIG_ATOMIC_MAX with integers *MAX.
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
The bug came from the fact we need to handle the destruction of the
main loop which destroy the underlying timer. The event handler that
catch the destruction of the timer can not make the difference between
eo_del call from the timeout code and eo_del from the main loop
destruction. By removing the event handler, the double free is properly
avoided.
As we add more object in the main loop, they can't live in the top
namespace as they make little sense there (Efl.Fd !). For coherence,
everyone should in the loop namespace, so move timer there.
Now when dealing with pointer types, we will not get pointer to
pointer semantics in callbacks and eina_promise_owner_value_set
for Eina_Promise.
It will work as expected:
Eina_Promise_Owner* promise = eina_promise_add();
void* p = malloc(sizeof(T));
eina_promise_owner_value_set(promise, p, &free);
This lets me narrow down the remaining cases of pointers across the EFL.
The void pointers will later need to be reevaluated on per-case basis and
replaced appropriately where possible/feasible.
This reverts commit 546ff7bbba.
It seems that eo_del() is useful and removing it was creating bugs.
The issue is that the way we defined parents in eo, both the parent and
the programmer share a reference to the object. When we eo_unref() that
reference as the programmer, eo has no way to know it's this specific
reference we are freeing, and not a general one, so in some
circumstances, for example:
eo_ref(child);
eo_unref(child); // trying to delete here
eo_unref(container); // container is deleted here
eo_unref(child); // child already has 0 refs before this point.
We would have an issue with references and objects being freed too soon
and in general, issue with the references.
Having eo_del() solves that, because this one explicitly unparents if
there is a parent, meaning the reference ownership is explicitly taken
by the programmer.
eo_del() is essentially a convenience function around "check if has
parent, and if so unparent, otherwise, unref". Which should be used when
you want to delete an object although it has a parent, and is equivalent
to eo_unref() when it doesn't have one.
this is an args event. right now we don't use it, but this should be
done by some of the setup/init of an app and then produce an args
event. the idea would be that this can be used by single-instance apps
like web browsers, terminology to treat launch as an event.
Complex types (i.e. list, array, hash, accessor etc.) now do not require
pointers with them anymore (the pointer is implied) and the same goes for
class handles. Eolian now explicitly disallows creating pointers to these
as well. This is the first part of the work to remove pointers from Eolian
completely, with the goal of simplifying the DSL (higher level) and therefore
making it easier for bindings (as well as easier API usage).
@feature
Previously events used to use class name as a prefix and ignored eo_prefix
when specified. This is no longer the case. Events follow eo_prefix by default
now. In order to get around this for classes where this is undesirable, a new
field event_prefix was added which takes priority over eo_prefix. If neither
is specified, class name is used like previously.
@feature
We used to have eo_del() as the mirrored action to eo_add(). No longer,
now you just always eo_unref() to delete an object. This change makes it
so the reference of the parent is shared with the reference the
programmer has. So eo_parent_set(obj, NULL) can free an object, and so
does eo_unref() (even if there is a parent).
This means Eo no longer complains if you have a parent during deletion.
So ecore main loop does restart everything with an main loop shutdown
and init when it detect a bad fd. This can happen if you del a fd after
you have destroyed it. Something terminology is doing (and should be
legal), but that then ended up with a main loop with no event handler
registered and the process was looking like stuck with nothing happening.
This allow you to monitor fd and get notification using Eo events. I
have not implemented the buffered read as used by X. I think that if
this is useful, we should just do another class to handle bufferred fd.
This reverts commit a13570c17c.
This doesn't really fix the problem which is hidden by eo capability to not
crash on bad unref. With legacy API you are allowed to do a ecore_timer_del
and also return EINA_FALSE. In that case you have a double eo_del (which is
luckily protected) and a double free (that is not). It does crash on the
double free, but the issue is a lifecycle issue. Will bring a better patch
for this.