This results in wayland wl_output modes having their refresh rates
1000x too high, which has the magical effect of making some native
Qt apps fail with a GTK error when Xwayland crashes on 32-bit systems.
@fix
Fixes T3587
if an application is connected to the session bus, being disconnected
should indicate that the session has been terminated and all applications
in that session should also be terminated
this ensures that no zombie processes will remain after the session bus dies
Add a way for users of the promise owner to get notified when a
promise progress is registered. Also added a convenience composition
function that creates a promise which is fulfilled when another
promise has a progress notification.
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.
Due to dependencies from elementary codegen on the edj files triggered by the
generation of .h and .c, we actually need to provide the edj as it would
otherwise to build edje_cc prior to any make dist.
As the edje file was added after in the git repository, you may have some
timestamp issue to fix before doing a make dist. Try touch on all codegen_example
files.
T3588
The idea here is to provide a "real" fdo icon theme along the icons in the elm edje theme.
To minimize file redundancy the icons are profided directly in a fdo structure and picked
in icon_fdo.edc. Also I'm using all the icons in svg format so that we don't need to provide
all the creazy sizes as per fdo spec.
More icons will comes in the next commits
This commit breaks behaviour!
Re-parenting no longer detaches composite objects, so watch out.
Now you can have an object be a composite object of an object although
it's not its child. This allows widgets to do things like having an
object as the child of a child object while still making it a composite
object to the main object.
With this change, composite objects don't keep a reference to the child,
but instead composite "bonds" are implicitly removed when either the
parent or the child are destructed.
add support for reconnecting wayland applications if the compositor dies
disconnect -> destroy gl ctx + image textures -> block rendering ->
reconnect -> create gl ctx -> create image textures -> unblock rendering ->
sprinkle special seasoning on top -> just like ma used to make
#SamsungFeatures
@feature
in the event that the global gl context changes, all engine image data
must be destroyed and then re-created in order to continue rendering
successfully
this can be null if engine info has not yet been set or if engine info
has been unset, leading to engine crashes. instead of adding null checks in
every engine, better to add the checks to the originating functions
someone with time to kill should go through all the EAPI functions and
add more checks
This patch fixes a compiler warning where eo_del was being called on
an Ecore_Job.
ref c1141c7b0b
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
Small patch to add an API function that can be used to set the current
pointer position on a given seat name. This function (typically) will
be used inside an Ecore_Evas to "pointer warp".
@feature
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
Small patch to add an API function that can be used to retrieve
current pointer position on a given seat name.
@feature
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
Summary:
object_find is more generic, so other mechanisms can also reuse the
code.
The object itself has to support the function, so there is no need for
eo_isa which would have a negative performance impact.
The base class implementation calls interface_get on the parent, so a
override of the function can just call the super function to continue in
the recursion.
Test Plan: just run the eo test suite
Reviewers: raster, tasn, jpeg
Reviewed By: tasn, jpeg
Subscribers: felipealmeida, netstar, cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D3909
If the compositor crashes at just the right time this dispatch can fail
and result in a stuck app spinning in a tight loop. Fix it so we do a
proper shutdown in this case.
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
On shutdown we need to set wl_surface to NULL to let the engine know
there's nothing to write to - however we've been setting up an entirely
new outbuf instead of just changing the old one.
The outbuf code should free the surface structure, not the dmabuf
abstraction or a use after free occurs on fallback.
Re-organize some code to make sure we don't rely on anything that may
have already been freed.
Add a wl_surface_commit() to keep the animation timer alive through
the fallback process.
This include is leftover from the porting to Ecore_Wl2. Remove
unused include for old Ecore_Wayland library
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
This patch deprecates the Ecore_Wayland library with a warning to not
use it. This (imo) is a better method of deprecation rather than
adding EINA_DEPRECATED to every function.
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
Because we were forcing EWAPI in this macro, one couldn't create a class
that is "static" or even just private or the module. The symbol was
always exposed.
Since in C the attributes of a function are set based on the first
declaration, we don't need to specify any attributes in this macro and
we can just rely on them being specified in the declaration. So for
example, for class "foo":
foo.h:
EWAPI const Eo_Class *foo_class_get(...);
foo.c:
const Eo_Class *foo_class_get(...);
Would give the desired results, a class would be EWAPI. This is already
done automatically for all of the classes using Eolian. Because of the
lack of specifiers, the default visibility will now be the default
visibility based on compiler flags and settings.
Thanks to JP for reporting this issue.
This can potentially (but very unlikely) break things.
If we do drag & drop and then do copy & paste, both _wl_selection_receive
and _wl_dnd_receive are called for one action (dnd or cnp). It is reduntdant.
We only need one data received callback to handle two cases dnd and cnp.
We have copy & paste and drag & drop selection types, but we cannot
distinguish between these two types when requesters receive data
from data ready event.
This patch adds a new enum to help selection requesters distinguish
between two selection types and have suitable actions for each type.
If we do drag & drop and then do copy & paste, both _wl_selection_send
and _wl_dnd_send are called for one action (dnd or cnp). It is reduntdant.
We only need one callback to handle two cases dnd and cnp.
we support different types for DnD, but there is no converters.
This patch adds converters for types, so that we can send
different data for different types.
When these changes got it with 0c76f82a31
they used the non existing symbols ecore_promise_value_get and
ecore_promise_then (most likely due to an unfinished refactoring)
Make sure we actually use the correct eina_promise_ symbols and add NULL
for the error callback which is now needed.
partially reverts e4d815dc48
this caused efreetd to crash almost immediately due to non-stringshared
strings being used in a stringshare-only hash data descriptor
fixes the following warnings:
/usr/local/include/efl-1/Efl_Model_Common.h:14:30: warning: redefinition of typedef 'Eina_Promise' is a
C11 feature [-Wtypedef-redefinition]
typedef struct _Eina_Promise Eina_Promise;
^
/usr/local/include/eina-1/eina/eina_promise.h:10:30: note: previous definition is here
typedef struct _Eina_Promise Eina_Promise;
To remove the typedef i had to cleanup the includes of header in
evas_canvas3d_eet.c.
this really needs a better solution since it results in a blank genlist
item space, with the item itself typically located somewhere offscreen due
to caching
the parent/group item is located after its child items in the list of items,
so if an item is appended to the parent's list of items then it must be
prepended to the parent or else it will end up being in the wrong group
@fix
more fixes for the comically unreviewed revision from elm which continues
to cause bugs months after it was pushed
ref 4c86a66f28876b68e92a90c8f741eed1130dd034 (elm)
ref e88423e994
Efl - efl_model_base changed to use eina_promise
Eio - eio_model use efl_model_base with promise
Eldbus - elddbus models use promise now
Elementary - elm_view_list and elm_view_form use new models with promise
updated all related examples and tests
This code is an absolute mess. This is the first step towards fixing that.
This cleanup let me find a bug that would have printed errors when using
Eina_Value so I fixed that too.
It is possible situation when SPACER structure has colors values.
For example:
group { name: "abc";
parts {
part { name: "rect"; type: RECT;
description { "default" 0.0;
color: 7 7 7 255;
}
}
}
}
group { name:"abc_2";
inherit: "abc";
parts {
part { name: "rect"; type: SPACER;
}
}
}
To avoid failing compilation of generated source code, need avoid
generate color source code for a SPACER part.
This fixes the segfault reported by Jack.
The problem was that the object was being reparented and thus
removed from the composition and never added back.
The hicolor fallback requirement is handled by efreet
and the usage of fdo is user specified now not by code.
This means the only (theoretical) way this could be a
problem is if the user removes a theme.
This seems like a good tradeoff to remove the overhead
and enable the apps to switch icons based on config change.
Allow to delete set if it is not used by any part
Function to check if set is used by any part is:
edje_edit_set_usage_list_get
Since it uses same struct as image_used_list_get function, it
can be freed by edje_edit_image_usage_list_free.
get list of images of set (edje_edit_image_set_images_list_get)
add image to set (edje_edit_image_set_image_add)
delete image from set by it's place (edje_edit_image_set_image_del)
list of sets (edje_edit_image_set_list_get)
renaming set (edje_edit_image_set_rename)
add new set (edje_edit_image_set_add)
set's id (edje_edit_image_set_id_get)
This reverts commit b48d3eb04e.
This patch can be occurred performance issue.
update is needed for only non-homogeneous case.
I'll revert patch and update new patch in phabricator for check.
C++ doens't allow empty size structure, empty structure looks harmful.
I've thought about the solution and this patch is my conclusion.
Because I'm not a big fan of elm_prefs,
if you have any better idea then please fix it on your way.
NB: Minor cosmetic changes such as removing a commented out line of
code and fixing an error message
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
we will remove elm_icon in the future.(only remain it in legacy)
for this, I added elm_image_icon_set API to the elm_image.
it is same as elm_icon_standard_set.
Summary:
In tree mode if parent item is hidden all its subitems should also be hidden.
@fix
Reviewers: Hermet, raster, cedric
Reviewed By: cedric
Subscribers: jpeg, reutskiy.v.v
Differential Revision: https://phab.enlightenment.org/D3918
Signed-off-by: Cedric Bail <cedric@osg.samsung.com>
Summary:
*_visibility_apply() functions are called from each *_visibility_adjust()
function. And it is also called from _elm_scroll_scroll_bar_size_adjust().
So, calling *_visibility_apply() functions after calling
_elm_scroll_scroll_bar_size_adjust() is unnecessary.
@fix
Test Plan: N/A
Reviewers: raster, jaehwan, eagleeye, woohyun, cedric
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3914
Signed-off-by: Cedric Bail <cedric@osg.samsung.com>
This was done following a feature request by @raster. There was no real
reason for it not to be an eo function and this gives us more
flexibility.
The reason why this done was to provide a way for classes to do special
things when an object deletion was requested, for example in the case of
Evas, hide the object.
Split function _edje_generate_source_of_state into
separated smallest functions, that represent generation
of source code for different blocks.
List of added internal functions:
- _edje_generate_source_state_map
- _edje_generate_source_state_box
- _edje_generate_source_state_table
- _edje_generate_source_state_image
- _edje_generate_source_state_proxy
- _edje_generate_source_state_relative
Add support collapse blocks, in cases when
block contain only one attribute.
For example:
next EDC code
description { state: "default";
{
rel1 {
relative: 0.5 0;
}
}
now will be replaced by:
description { state: "default";
{
rel1.relative: 0.5 0;
}
drm_mode_fb_cmd2 has fields for a modifier to handle tiling,
compression, etc (per plane). Even tho we do not use these, we should
at least initialize them to zero else we end up with uninitialized
bytes in the cmd structure.
==11706== Syscall param ioctl(generic) points to uninitialised byte(s)
==11706== at 0x57E05D9: ioctl (in /usr/lib/libc-2.20.so)
==11706== by 0x4D30FA3: drmIoctl (in /usr/lib/libdrm.so.2.4.0)
==11706== by 0x4CDF66B: _ecore_drm_fb_create2 (ecore_drm_fb.c:63)
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
The eglfs, drm and gl_drm engines were all linking to eeze internal
libraries however none of those engines actually use eeze at all so we
can remove the linking there.
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
This renames some APIs from Efl.Pack in order to try and
make them more sensible. Feedback welcome.
Also, most APIs now return bool, where false is quite unlikely
to happen.
This moves the part_box and part_table APIs to a fake object
like was done in Edje Object.
This also adds support a few new APIs to those containers,
so they behave exactly like Edje.Object.
This is another implementation bit of "eo_part".
This fixes on top of 49a27688b1, which
assumed somehow that table items had names, although that
might not be the case. In my situation, name = NULL and there
was a crash everytime I clicked on the clock widget.
This function was returning only the text defined by part_text_set.
This is a behaviour change, as part_text_get() no longer returns NULL
if there is a string defined in EDC. But this now means that the actual
value displayed on screen can be retrieved by a simple API call.
@feature
This is similar to the previous patch for Box,
but for Table.
Those new EO API as well as the legacy ones still need to
be tested (no test case in make check...)
@feature
This removes Efl.Pack_Named which had a terrible name,
removes Elm.Container which should have been renamed
Efl.Ui.Container anyway, and introduces an interface
Efl.Container instead.
The hierarchy tree is now changed as objects don't inherit
from Efl.Container (it's an interface, not a regular class)
but only implement it. Obviously it is very easy to
reintroduce an Efl.Ui.Container parent class if we need it,
but I guess it should have some actual logic. It's basically
part of what Elm.Widget already does.
Some function names have been modified to look better in C
with the efl_content prefix.
@feature
As discussed on the mailing list we're taming the icon_standard code.
This means that icon themes are set by the user not the app.
Deprecate all the icon_lookup calls etc and remove the behaviour.
Add a config panel to elementary_config where the theme can be selected.
this is copied from weston, except all the code which is supposed to be of
struct type "offer" has been renamed "source" and then reused in the same way
that weston uses "offer" in order to maximize reader confusion
deleting the source here is invalid since the selection only changes in the
corresponding callback from the wayland interface
The definition of where to load icons is now up to the user
(through the configuration of the icon_theme config value)
rather than being defined in code per-app or even per-component
Summary:
add programs block to the sample code of LazEDC,
and add comment about possibility to omit "default"
when part description inherits from "default".
Reviewers: Hermet, zmike
Subscribers: cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D3915
Summary:
The toolbar's box was resized in _sizing_eval(), _resize_job().
In _sizing_eval(), the box was resized according to its minimum size.
And in _resize_job(), toolbar would recalculate it and resize the box again.
If _sizing_eval() was called after resizing the box properly from _resize_job(),
the box was shrank before calling the next job.
If the box's minimum size is needed for calculation in the job callback,
it shouldn't change box's size before the job callback.
@fix
Test Plan: N/A
Reviewers: jaehwan, eagleeye, woohyun, cedric
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3911
Summary: On Windows 64 bits, long is 32 bits lonG
Test Plan: compilatioN
Reviewers: cedric, jpeg, Hermet
Reviewed By: Hermet
Differential Revision: https://phab.enlightenment.org/D3910
If the table item has a name, posible case when item name length, with
index, will be 12. The 12 is predefined length for box index.
Quote Cedric
In a box, the index is one dimension, one int, thus the length
of it (from int to string) will always fit inside 12 bytes. That's
where this 12 comes from. That's also how the unique name of that item
is defined.
This commit separate the items name generation by part type, it will be
more correctly.
@fix
so geneirc data, wrefs, comments and id's are not that common so put
them all into their own memory segment that's allocated separately to
the core object so we only use this memory when needed. we already had
an extension section anyway so it's not new - just using it now for
more of the rarer bits of data. 2 more pointers gone from most objects
anyway and now extension code handling is a bit cleaner.
so memory for class id -> ptrs is mallocs. this means it likely will
be next to other memory malloced. which means overrunning memory
someone mallocs could walk into the class table and corrupt it. we put
eo ids in mmaped regions to avoid this if possible in case of buggy
code. let's do it for classes too.
this also now allocs in larger chunks. for mmap its in page chunks
(which can hold either 1024 or 512 classes depending on 32 or 64bit).
reallocs still work if mmap is not there and we do them in chunks of
128 classes (it seems that we start at about 70 or so classes atm when
elm_test starts and it grows to ~100 or let's do 128 as that's pretty
much our base as a power of 2 and we now dont realloc much).
if an object iot reffed or not hide on del. it should have been this
way before eo. eoifications i think messed a few things up.
this does bring up an issue... in eo we have no way to explicitly do
stuff on eo_del regardless of references at the time. this needs to be
solved.
@fix
lots of long paths for monitoring file paths for icons etc. are in
memory for efreetd. this reduces that memory by sharing them much more.
@optimization
Summary: moved mempools adding to separate method to avoid this problem in future
Reviewers: cedric, reutskiy.v.v
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3903
Signed-off-by: Cedric Bail <cedric@osg.samsung.com>
Summary:
For EDJE_RP_TYPE_SWALLOW objects (like GROUP, SWALLOW, EXTERNAL),
edje_object_part_object_get() will return NULL or transparent rectangle.
The note is added to have developers use edje_object_part_swallow_get()
in that case.
Reviewers: jpeg, Hermet, cedric
Reviewed By: cedric
Subscribers: Hermet
Differential Revision: https://phab.enlightenment.org/D3859
Signed-off-by: Cedric Bail <cedric@osg.samsung.com>
So it seems we are using Eina_Hash_Iterator quite a lot more than before.
This lead to a huge amount of alloc/free of Eina_Rbtree_Iterator that
was noticable in Enlightenment callgrind trace. This patch make it vanish
from the trace :-)
This patch ports the existing session recovery protocol from
Ecore_Wayland so that it is used inside Ecore_Wl2.
@feature
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
We use the summary line in various places where such a big block of text looks
ugly. In general we handle it a bit like our commit messages. One line summary
which should stand alone with more details and explanantions following after a
empty line.
Summary :
when mouse up happens, item must be focused, and after that,
select must be called.
current logic, focus is already called inside the select function
and again focus set is called in the end of mouse_up,
so if user new focused widget inside the select callback,
mouse_up forcely take the focus into the item again.
select callback should comes after the focus callback calls
to work properly.
@fix
this means that on loop_get on any obj as long as its a child of a
loop obj... it'll retunr that loop now. it will work. no more code
needed.
we can shortcut this with ui/gfx objects returning the mainloop
singletone.