We can't blindly unload here - this loses swallowed content (in swallows,
boxes, tables etc.) ... this here along with an actual implementation of file
unload broke the pager in E for starters as shading then unshading (double
click titlebar) a window would lose the mini preview image obj swallowed in.
also fullscreening would do it. this also broke gadget bar, the xkb gadget in
it too and more... so this is a particularly bad thing break.
@fix
Summary:
Cluster consist of one glyph, expected to be removed on backspace or delete key.
Cluster consist of multible glyph, expectd to remove part of on backspace or delete key.
This is behaviour founded in Android. (our current way of handling similar to Qt)
**New Behaviour**
{F3750386}
**Old Behaviour**
{F3750387}
Test Plan:
Auto Testing is challenging because there are no easy way to emulate keyboard down on elm_entry
```
#include <Elementary.h>
EAPI_MAIN int
elm_main(int argc, char **argv)
{
Evas_Object *win,*box,*entry;
elm_policy_set(ELM_POLICY_QUIT, ELM_POLICY_QUIT_LAST_WINDOW_CLOSED);
win = elm_win_util_standard_add("", "");
elm_win_autodel_set(win, EINA_TRUE);
box = elm_box_add(win);
entry = elm_entry_add(box);
evas_object_size_hint_weight_set(box,EVAS_HINT_EXPAND,EVAS_HINT_EXPAND);
evas_object_size_hint_align_set(box,EVAS_HINT_FILL,EVAS_HINT_FILL);
elm_entry_entry_set(entry,"<font=NotoColorEmoji wrap=mixed >🇧🇬อั🇧🇬อั😀😀😀😀😀อั</font>");
evas_object_size_hint_weight_set(entry,EVAS_HINT_EXPAND,0.9);
evas_object_size_hint_align_set(entry,EVAS_HINT_FILL,EVAS_HINT_FILL);
evas_object_show(entry);
evas_object_show(box);
elm_box_pack_end(box,entry);
elm_win_resize_object_add(win,box);
evas_object_resize(win,320,480);
evas_object_size_hint_weight_set(entry,EVAS_HINT_EXPAND,0.1);
evas_object_size_hint_align_set(entry,EVAS_HINT_FILL,EVAS_HINT_FILL);
evas_object_show(win);
elm_run();
return 0;
}
ELM_MAIN()
```
Reviewers: tasn, woohyun, bowonryu
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9628
Summary:
During _elm_config_color_overlay_apply() application can call this api
100's of time depending on how many overlay it wants add . As each color_class
set triggers the global color class apply chain 3 times (for color , outline and shadow)
just club them once and emit single color_class,set event.
Reviewers: Hermet, ali.alzyod, zmike
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9784
Summary:
All those api's are working on the Edje_file level so by appending the file prefix
and changing the argument to Edje_File leads to easy maintainability.
Depends on D9776
Reviewers: Hermet, ali.alzyod, zmike
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9777
unload is a crucial method for classes that implement load, as this is
the method which is called during e.g., efl_file_simple_load() when
a new file is specified in order to destroy the previous object data
Reviewed-by: Marcel Hollerbach <mail@marcel-hollerbach.de>
Differential Revision: https://phab.enlightenment.org/D9786
Summary:
As user request for the evas_textblock_style through _edje_textblock_style_get()
api and that api implements lazy computation of the styles. By just marking the
style dirty (stl->cache=false) will be enough. no need to calculate the style
computaion at that time.
Reviewers: Hermet, ali.alzyod, kimcinoo, woohyun
Reviewed By: Hermet
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9776
Summary:
Now all textblock_style functionality are wrapped under api
for easy of optimization. so going forward no one should directly
acess the styles from the File . the styles should be acesses by
the internal api's provided by edje_textblock_style.
Reviewers: Hermet, ali.alzyod, woohyun, kimcinoo
Reviewed By: kimcinoo
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9774
Summary:
move the file related function to edje_cache so that code
is easy to read and maintainable and we don't have to do unnecessary
stuff for each edje object creation.
Reviewers: Hermet
Reviewed By: Hermet
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9761
Summary:
Why :
1. The function operates on the Edje_File level not on Edje object
level , so express it clearly in the function argument.
2. if its clear that this function works on file level then
this should be called when the theme file loaded for the first time
not on every Edje object creation.
Reviewers: Hermet, zmike, raster, cedric
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9753
Summary:
We were creating 3 eina_list and destroying inside a for loop.
replace that with creating a single eina_array at start and use
eina_array_clean() function ro reuse the array.
Reviewers: Hermet, zmike, cedric, raster
Reviewed By: zmike
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9745
Summary:
This function does lot of things
- can be called only to load the file (by passing coll as null)
- can be called to load both file and open the collection from the file.
- handles the file_cache logic
- handles fixing the collection after reading from file.
this patch is targeting to split the responsibility to
smaller function for easy maintenance and code readability.
future patch to follow for splitting the file opening and collection
opening to two different function.
Reviewers: Hermet, raster, cedric, zmike
Reviewed By: zmike
Subscribers: zmike, cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9715
Summary:
If the style is readonly then we know for sure it dosen't have any text_class/color_class.
If a style has text_class tag then call update only once not for each text_class tag it has.
Reviewers: ali.alzyod, Hermet, cedric, raster
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9641
i found some more cases where the hash may change, so del and add from
the hash when this happens...
and... i found a nasty. _edje_signal_match_key_cmp compared pointers
like:
int return = ptr_a - ptr_b;
what happens if .... ptr_a and ptr_b are more than 2^31 (2gb) apart?
overflow (or underflow) and we return the wrong thing. i suspect this
is part of the problem and why my has remove/adds have not been
working because ... i suspect that maybe the hash dels have not been
finding things. i can't be sure right now, but it is an obvious
problem that i fixed by just doing if's and returning -1 or 1. also i
found a double-add or overwrite int he hash - when we shuffled with
_edje_signal_callback_move_last the matches CAN match exactly
something already in the hash thus adding it in will conflict with
what is already there as keys match. handle this cvase now and i have
seen segv's go away for now.
@fix
Summary:
Instead of updating all the styles in an Edje_File just mark them
dirty so that subsequent call to style will recompute the new style before
returning the style.
Reviewers: ali.alzyod, Hermet, raster, cedric
Reviewed By: cedric
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9697
so i think there was also another bug as i saw this again now. this
should hopefully plug the signal/mtach memory bugs now by removing
and adding back to hash as the hash key relies on the memory addresses
of signal/src in the matches array which change on realloc. a bit
brute-forceey but... better than crashes.
@fix
Summary:
We already have a escaped string for the font_source so
this code is unnecessary.
Reviewers: Hermet
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9698
Summary:
If the style is not readonly we always compute the style again when
requested by edje. so this computation is unnecessary.
By avoiding the computation here we will avoid style computation of all the styles in the edje
that is not readonly hence saving memory.
Reviewers: ali.alzyod, Hermet, cedric, raster
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9693
in 1 situation at least we delete the eina file (close it) but keep
the ptr around (during destruction) which could cause issues with
callbaks and events on del and so on.... which may lead to multiple
closes where only one should happen ... which would explain my invalid
eina file ref problems i'm seeing. i carefully matched eina file
handle stores/opens/dups to closes in edje/evas and they seemed to all
match up so this audit with comments and fixes seems to have plugged
that now.
@fix
we rarely use tables in edje, so move it out of the hot path for
intruction prefetch/cache. also for calc single - move things to sub
funcs so things like mesh, light and so on code is always out of the
hot path as much as possible. we probably can merge our 2 switch
statements as well. this really just restructures the code to move
stuff into sub functions which also does make the calc funcs look
simpler and easier to read.
also a general cleanup of the code to make it easier to follow/read as
this seems to have problems but i cant reproduce them enough to find
them. i noted a realloc would have invalidated pre-stored pattern
ptrs that would cause segv's for sure. also code paths where reallocs may
fail and not handling those cases at all etc.
@fix
As edje mostly deals with style string. to get the style data each time
it linearly search through list to find out the style which is not very
cache friendly so keep a hash to do first lookup with less impact on cache.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D9547
Currently we do 2 pass updation. first we scan through all the styles
and check if they have text_style which matches the test_style we need to update
then we mark them dirty. then we call style_all_update() to go through the list
again and update those styles.
By combining them both in a single function we avoid scanning through the whole
list again.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D9639
Summary:
we should only add font_size tag if the new size is different.
we should only add font tag if there is a new font.
Reviewers: ali.alzyod, Hermet, cedric, raster
Reviewed By: Hermet
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9613
Obviously previous edje object file set would try reload file
even though the file is already loaded.
This brings different result if the file data has been modified or changed.
Just to check if the edje style has text_class tag we do
lot of pointer hopping by linearly scan through the tags in the
style which is not very cache efficient.
by keeping a readonly flag we can avoid those acess if the style dosen't
have any text_class tags. and if we have those tags then we can start
updating the style straight away.
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D9546
these function the same as the min size hint versions and enable
distinction between internally-set max size hints and user-set max size
hints
@feature
ref T8122
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Reviewed-by: Xavi Artigas <xavierartigas@yahoo.es>
Differential Revision: https://phab.enlightenment.org/D9553
Summary:
canvas objects do not need localization because they are not directly user-facing
this should only be inherited by objects which need to be localized
Depends on D9559
Reviewers: q66
Reviewed By: q66
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9560
Summary: these are not distributed apis so they should always remain beta
Reviewers: q66
Reviewed By: q66
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9559
Summary:
Both font and font_size are already added into the style text
in _edje_format_reparse() function and there we update the tag->font_size
as well as tag->font member. so I think it is unnecessary to
add again which has memory as well as parsing performance impact.
Note :
someone please update this cryptic comment
/* Add font name last to save evas from multiple loads */
how this is going to help saving multiple load.
Reviewers: ali.alzyod, Hermet, raster, cedric
Reviewed By: Hermet
Subscribers: segfaultxavi, cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9543
Evas_TextBlock_Style has no idea about the text_class tag its a garbage value to it.
So keep the text_class tag in the edje level and update the text style property in the
final style string when necessary.
Because text_class id tends to be unique by removing from the final style string
enable it to be shared (string shared string).
Reviewed-by: Cedric BAIL <cedric.bail@free.fr>
Differential Revision: https://phab.enlightenment.org/D9544
Summary:
If the color_class is not overridden by the object level the
hash will be empty but still we do the expensive call to _edje_hash_find_helper()
find the color_class in an empty hash. by checking if the hash is empty
and returning early we save lot of unnecessary hash computaion and lookup.
Reviewers: Hermet
Reviewed By: Hermet
Subscribers: cedric, #reviewers, #committers
Tags: #efl
Differential Revision: https://phab.enlightenment.org/D9532
do it once and remember the result from the first one. drops overhead
for sure by a chunk i actually could see in perf reports like about 1-2%
of cpu...
This reverts commit 895ffd93cc.
This commit broke E's widget toolbars that only had text - so many
config dialogs broke. too simple to fix - it's a wrong premise to
begin with it would seem.
Marcel Hollerbach
Daniel Kolesa
junsu choi
Carsten Haitzler
Ali Alzyod
subhransu mohanty
Mike Blumenkrantz
Hermet Park
Yeongjong Lee
Cedric BAIL