Summary:
This commit makes use of the `ownership,shared` and `ownership,unique`
events from Efl.Object in order to avoid the C# wrapper from being
collected while C code holds a reference to the object.
For example, creating a list of items in a for loop and attaching events to
them would fails without this commit, as the C# GC may collect the wrapper.
The basic idea is that we use a `WrapperSupervisor`, which is stored in
the Eo data storage, with a GCHandle allocated for the lifetime of the
underlying Eo object. This supervisor takes care of holding either a
weak C# reference (when in unique mode, allowing the wrapper to be GC'd)
or a hard C# reference (when in shared mode, making the wrapper
non-collectable while the Eo has extra references).
One limitation is that object graphs can leak if a shared object in the
graph - an Eo child for example - stores a hard reference to another
object in the graph as a C# field. In this example, this causes the
parent to always have a hard C# reference (from the child) as the child
is non-collectable due to the parent holding an Eo reference to it.
Depends on D8678
Test Plan: `ninja test` and `make test`
Reviewers: lauromoura, felipealmeida, woohyun, segfaultxavi
Reviewed By: lauromoura
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9014
Summary:
Rework general event handling to check individually each event call, if the
object is not alive then the event will not be propagated.
WeakReferences (and lambdas capturing those WeakRefs) are used to ensure this.
Dispose methods in object now take care of checking if efl libraries are still
initialized and thread-safely unregister each event before performing an
efl_unref on the Eo object.
Event handling in C# is now centered around a single dictionary inside the
object: `EoEvents`.
C# event triggers now properly trigger events on C too.
Standardize C# event-triggering methods names (remove underscores).
Some diminished use of static memory due events no longer requiring static key
objects to be registered/unregistered.
Some fixing of white space generation for generated events.
Depends on D8431
Reviewers: lauromoura, felipealmeida, segfaultxavi
Reviewed By: lauromoura
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D8564
Summary:
This commit mainly fixes errors caused by deallocating resources in the garbage
collector thread. Using `ecore_main_loop_thread_safe_call_async` to queue
resource deallocation in the main thread seems to solve it.
Also, some `efl_ref` calls are added in places they were missing, mainly
objects that unref in the destructor thus taking ownership if efl_ref is not
called.
Also fix improper resource deallocation in tests that were causing it to crash,
enabling it to call Efl.All.Shutdown again. This allocation and the deallocation
process was moved from the Eo class constructor to static class methods that are
called in the test 'set up' and 'tear down' methods.
Queuing resource deallocation in the main thread make it mandatory that tests
call `Efl.App.AppMain.Iterate()` if they want to check proper resource
deallocation (like TestFunctionPointers.set_callback_inherited_called_from_c).
Extras:
Remove duplicated declaration of 'eflcustomexportsmono' in meson in order to fix
some linking problems.
Remove some unused code around deallocation functions that had to be reworked.
Object allocation is now supplied with the call site information it expects
(file name and line for _efl_add_start).
Depends on D8550
Test Plan: meson test
Reviewers: felipealmeida, lauromoura, cedric, segfaultxavi
Reviewed By: lauromoura
Subscribers: segfaultxavi
Tags: #efl_language_bindings, #do_not_merge
Differential Revision: https://phab.enlightenment.org/D8431
Summary:
It was marshalling erroneously data into and out of arrays and lists.
Instead of passing data by value (or by address of correct size), it was
stuffing data into IntPtr and trying to parse out afterwards.
This commit changes the binding to use the same approach of plain
Get/Set, with proper overloads.
Reviewers: vitor.sousa, segfaultxavi, felipealmeida
Reviewed By: vitor.sousa
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D8057
Summary:
eina.Value.Empty now means that we have an zeroed (empty) eina value.
For optional values that are empty use eina.Value.OptionalEmpty.
This was required to support the empty values passed with
EINA_VALUE_EMPTY in some Ecore futures.
Also, returning an eina_value by value is not supported in eolian
for safety reasons, so we removed some tests that tried to use this
behavior.
Depends on D6171
Reviewers: felipealmeida
Reviewed By: felipealmeida
Subscribers: cedric, zmike
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D6172
Instead of messing around with varargs, create individual wrappers for
each type supported.
The va_list approach was getting problems with float/double on Windows.