In the specific case where you had "class A extends B composites C"
the correct composites branch was ignored and instead the implements
branch was used. This was entirely wrong/an oversight that did not
appear until now. Other combinations were handled correctly.
When a class composites an interface, we need to ignore all of
its extends (and extends of those) as well as the main interface
when doing API checks, as composites essentially provides a
guarantee that this *will* be implemented at runtime, which
further extends to the whole inheritance tree of that interface.
Fixes T8491.
Summary: This reverts commit 712266fb9a.
Test Plan:
run elementary_test -> genlist2
add more items by + and before/after button.
if you delete selected items by click - button,
issue cases, you can see item area become empty and not filled.
if you click U button, so update items,
issue cases, all item disappeared.
the issue is caused by removing calc_job,
so there is no job to performed,
and loof is idled states.
in Tizen,
to solving this problem,
we add dummy job in recalc cases.
Need to see how to solve the issue more simply,
but firstly raise the ticket to notify severe issue.
Reviewers: zmike, cedric, bu5hm4n
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10663
To mimic function behavior
Reviewed-by: Felipe Magno de Almeida <felipe@expertisesolutions.com.br>
Differential Revision: https://phab.enlightenment.org/D10381
Efl.Model can be nested and Efl.Ui.CollectionView use this ability. Listening
on the nested model instead of the nester model can lead to event being triggered
that shouldn't. Better listen on the top model.
Differential Revision: https://phab.enlightenment.org/D10621
This fix adding item not always showing up in the CollectionView right away.
The main issue comes down to the fact we can not chain a new request when one
is already resolved partially. The code now make sure to properly do this.
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D10614
ecore_file_cp can cause config data(eet file) invalid,
once multiple processes are trying to call elm_config_save.
this patch replaces it with ecore_file_mv to prevent the problem.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10578
It should takes true almost all times when it checks.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10606
efl_ui_position_manager_entity_entities_ready is expecting the end offset and not the
length of how many items have been updated. Sending the correct number fixes the problem.
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Reviewed-by: SangHyeon Jade Lee <sh10233.lee@samsung.com>
Differential Revision: https://phab.enlightenment.org/D10590
Summary:
in case of setting a state of check by calling a API,
the flags that used for changed cb aren't changed.
which causes that the callback not invoked when it should be.
this patch resets the flags to make it work properly.
Reviewers: bu5hm4n
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D10591
First - the big problem. This breaks enlightenment's bluez5 popup. it
does a sortyed inert using the item data and the item data for one of
the itmes to compare in _cb_insert_cmp() in e_mod_popup.c when it
calls elm_object_item_data_get(0 returns a junk ptr for the item data
after this patch. i haven't managed to figure out exactly why in my
last 30 mins of looking.
But a closer look... this disables immediate processing of:
1. the first block of items (32items) which was intended so for
short/small lists you have some content by the time you go to the
first frame, and at least the first block of itso you seem to have
visual contnt and not a blank list until idlers can process further
content. so the patch being reverted would have gotten rid of this
logic that gets you content as opposed to blank:
while ((sd->queue) && ((!sd->blocks) || (!sd->blocks->next)))
2. if it's a homogenous list, all items have the same size so we do
have to realize the first item of each class type but ONLY that one.
further items should not need realizing as we can ASSUME the height to
be the same as the first item... that's the point of homogenous +
compress lists - all items of the same class have the same height and
width thus shortcutting further need to calculate nd realize. if we
are reizing everything in a homogenous list then the issue lies with
something going wrong with this above logic. we shokuld be able to
handle such lists super fastif this logic was working.
that's the 2nd while:
while ((sd->queue) && (sd->blocks) &&
(sd->homogeneous) && (sd->mode == ELM_LIST_COMPRESS))
so overall, this should not have been realizing every item. either
just the first block of 32, OR just the first item of any class and
thus assume all further items are the same size without realizing
being needed. if these broke then the solution is not commenting this
out but finding out why this logic is breaking :)
and not to mention... this commenting out also caused segfaults in
existing applications which are doing the right thing. perhaps the
sorting logic also needed updating as well if this above is commented out...
but i didn't have time to chase it more than this.
---
This reverts commit 0777b74f07.
This avoid comparison with potentially uninitialized byte.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Differential Revision: https://phab.enlightenment.org/D10479
Summary:
ECORE_X_EVENT_FIXES_SELECTION_NOTIFY is only initialized when XFIXES
is available. If ecore_event_handler_add is called with type == 0, it
will trigger an abort and elementary would not initialize properly.
Depends on D10491
Reviewers: zmike, raster, bu5hm4n, Hermet
Reviewed By: zmike
Subscribers: #reviewers, #committers
Tags: #efl
Maniphest Tasks: T8321
Differential Revision: https://phab.enlightenment.org/D10492
Summary:
This reduce by 3 the amount of callback registered on the canvas. Another
potential improvement would be to only register those callback if someone
is listening for a 'seat,*' event or if the edje file define seat filters.
Depends on D10484
Reviewers: zmike, raster, bu5hm4n, Hermet
Reviewed By: zmike
Subscribers: #reviewers, #committers
Tags: #efl
Maniphest Tasks: T8321
Differential Revision: https://phab.enlightenment.org/D10486
if a legacy timer callback returns false, the timer is deleted. in the
case where the legacy timer is deleted inside the callback while the same
timer is ticking recursively, however, the deletion must be deferred until
the outer-most frame of the timer's callstack has returned from the callback
in order to avoid improperly handling the timer
@fix
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10545
no need to strcpy here when we can just pass the length to stringshare directly
CID 1382854
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10441
this cannot currently be used for anything and was not explicitly stabilized
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10540
this cannot be used for anything and serves no purpose
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10539
Summary: these should error so the user can detect that they screwed up
Reviewers: devilhorns
Reviewed By: devilhorns
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10563
if the block is realized, its items have been registered into the focus
manager and must be unregistered to avoid double-registering
@fix
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D10544
cached item contents should already be hidden by the edje clipper, so this simply
changes their visible state to break them out of the focus calcs
contents must also be explicitly marked as unfocusable during calc-only realize
operations in order to avoid triggering a full focus recalc which will error due
to missing focus adapter in the item block
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D10543
item move operations require pan recalc in order to process the item block
positioning updates
@fix
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D10533
this is only a smart callback and not an eo callback
@fix
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10527
the documentation says this should trigger when filtering is done, and if
no filtering is pending then it is done
@fix
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10526
passing the smart data here (which we always have) makes the function
a little clearer to read. similarly, we can check whether the filter_data
pointer is set inside the function to avoid having to check it everywhere else
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10525
this may be called successively during the same mainloop iteration,
so it's important to defer this as much as possible
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10524
now that we have a sane and consistent sizing calc mechanism, we no longer
need to be randomly creating jobs to do additional side calcs in addition
to our other multiple bespoke calculation methods
instead, we can now call the calc function directly during the group calc
to perform all the calcs at once and avoid the overhead of constantly triggering
calc jobs
a possible future improvement here could be to remove the _calc_job() call in
the internal pan object's group_calculate, as this is likely a duplicated op
but it will require additional testing to verify
unit test performance (e.g., tree) increases roughly 50-80% after this patch
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10523
this is a colossal waste of time. it ends up realizing every single item
immediately even if it won't be visible, which defeats the purpose of all
the batching and viewport calculations that genlist explicitly does to
avoid exactly this type of behavior
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10522
this was never implemented or documented anywhere and serves only to be
misleading for anyone reading the code
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10518
* highlight should only be triggered on mouse down, as that is the point of
interaction with an item which indicates it is "in use"; a mouse-up event
cannot occur on an item which has not previously received a mouse-down event,
so toggling this on mouse-up will be wrong/duplicated 100% of the time
* unhighlight should only be triggered during mouse-up events if the list is
in no-select mode, as it will otherwise be implicitly during selection if
necessary
this should ensure that these events are emitted exactly one time and correctly
for each click event
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10517
if the currently-processed timer is recursively deleted (efl_del) while
it is inside the timer tick event callback, we must correctly handle this
case:
* in the place where a timer's inlist is de-linked, we must check to see
if the timer is the current timer and then update that pointer with the next
timer in the list
* in the post-tick part of timer processing, we must NOT update the current timer
pointer if we detect that it has been updated recursively
this fixes processing of timers in the mainloop to trigger more than one legacy timer
per mainloop iteration and likely has some (positive) impact on mainloop throughput
@fix
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D10515
Summary:
It will be resized on window_resize_job. when the window got specific size.
I don't think this is needed here. It cause doing a calc in vain.
Test Plan: Launching Test applications.
Reviewers: raster, Hermet, zmike
Reviewed By: zmike
Subscribers: zmike, cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10550
Summary: this is identical to the conditional above and can be merged
to improve code clarity
Reviewers: raster, Hermet, zmike
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10549
Summary:
when the sportlight is hidden, the state should be applied to the
backclip and foreclip, in order to hide the subobjects correctly.
This fixes dangling showing widgets.
Reviewers: Jaehyun_Cho, zmike, segfaultxavi
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10468
if a path has a '/' character in it before the ':' character then it
should be treated as a local file which contains a ':' character in
its name
Reviewers: devilhorns, zmike
Reviewed By: zmike
Subscribers: zmike, cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D10409