Summary:
this is only meant to listen to data which is currently available,
not wait for new data
@fix
Depends on D5866
Reviewers: cedric
Reviewed By: cedric
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D5867
Reviewed-by: Cedric Bail <cedric@osg.samsung.com>
Summary:
this resolves a race condition when a thread join was pending during
shutdown but a pipe write was needed in order for the join to be
successfully executed before shutdown had occurred
@fix
Reviewers: cedric
Reviewed By: cedric
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D5866
Reviewed-by: Cedric Bail <cedric@osg.samsung.com>
Looks like a copy and paste left over from another test.
../../../src/tests/elementary/elm_test_genlist.c:58:16: warning: unused variable ‘type’ [-Wunused-variable]
if audio objects are around after ecore_audio has been shut down then
they may access the sndfile or pulse lib ptrs/funcs and thus dont
unload these. should fix crashes on shutdown.
By default all Pthread are cancellable. This leads to strange behavior that
happen during shutdown and hasn't been figured out. Most likely fix for T6622.
this has 2 bugs really. 1 - the format masks for plain text included
markup which ... is wrong - right? that should have a mime type of its
own... either way it leads to plain text pasts being handled wrong in
wayland... so this fixes that
also the string buffers in wayland just were nothandled right. buffers
were raw binary buffers plus size,not strings, but they were strdup'd
etc. assuming nul termination. oh so wrong to do that. this also fixes
that as i fpound it while fixing the above format bug.
this fixes T6284
@fix
[Dereference after null check]
(1) src/lib/ecore/ecore_main.c
- _efl_loop_handler_efl_object_finalize checks if pd->loop_data is NULL.
After that, _handler_reset > _handler_clear > _ecore_main_fd_handler_del >
_ecore_main_fdh_pool_del is directly dereferencing pd->pool_data.
- _efl_loop_handler_efl_object_parent_set checks if pd->loop_data as well.
Then it calls _handler_reset as well.
(2) src/lib/ecore_wayland/ecore_wl_dnd.c
- ecore_wl_dnd_selection_set checks if t - result of wl_array_add - is NULL.
And it is dereferecing t directly for wl_data_source_offer.
(3) src/lib/elementary/efl_ui_dnd.c
- Third parameter const char *data could be NULL.
In this case strlen dereferences NULL. The data should be non NULL value.
I have checked this with Mr. Thiep Ha.
(4) src/lib/evas/canvas/evas_object_inform.c
- _efl_canvas_object_efl_gfx_stack_stack_below checks if obj->layer is NULL.
So it could call evas_object_inform_call_call_restack which is dereferencing
obj->layer directly.
tell other windows they lost their selections correctly if you have
multiple windows within the same process. this fixes multi-window mode
in terminology.
@fix
well call sync returns a void * too.... instead of just void return
(nothing) but this makes it easier to write and maintain code due to
consistency of function callback signatures.
leave the exact numbers less defined because core count can change on
the fly and could be virtually limited by policy in future etc. so
keep things less exactly defined so people dont go depending on exact
results which was never really intended.
this works out to the same number of total tests as slstr_many but
now split across all the threads
fix T6846
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
these are supposed to be for testing the finding of values not in the
hash, not basic hash operations
fix T6839
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
there's no point in iterating this many times; the base step size is 32
and so each loop will only increase the buffer...not at all since it takes
3 loops for the buffer to increase by the step size
verify that the buffer increases and then stop instead of spinning uselessly
fix T6835
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
the timing does not matter for these tests, so there's no point in
randomly waiting longer and delaying build completion
fix T6832
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
this test is to try and match values in a tree which have not been
inserted, so remove irrelevant insert/remove operations as well as
insane looping
fix T6830
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
the constructor calls init for every object, so ensure a matching
shutdown call exists to balance the init count
ref 517471190f
ref T6816
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
for some reason, a few projects changed this from the expected value of
abs_top_builddir for their test suites, breaking the usage of this define
within those tests
fix T6812
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
this performs an eina init/shutdown check and loads available modules
before running remaining tests
note that this still does not work and may never have worked because
PACKAGE_BUILD_DIR is broken
ref T6812
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
individual tests should not need to explicitly call init/shutdown functions
in most cases, and many did not properly do this anyway
see followup commit which resolves some issues with eina tests
ref T6813
ref T6811
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
efl_check.h must be included and the EFL_START/END_TEST macros must be
used in place of normal START/END_TEST macros
timing is enabled when TIMING_ENABLED is set
https://phab.enlightenment.org/w/improve_tests/
Reviewed-by: Stefan Schmidt <stefan@osg.samsung.com>
When changing cursors under wayland sometimes we'll see the old cursor
moved to the new hotspot briefly before the cursor changes. This makes
that suck less often.
A proper fix would involve creating a new wayland surface for every
cursor change (actual change, not just a new frame of an animated
cursor). Given the current internals this is invasive. Do the
easy thing for now.
Summary:
we have to check whether all of descendant are none_logical.
ref T6800
Test Plan:
1. elementary_test -to ctxpopup
2. Click 'Ctxpopup with user content'
3. Try focus ctxpopup content(button) using arrow key.
4. Check that the content is focused (you can move scroll bar)
Reviewers: bu5hm4n
Subscribers: cedric
Maniphest Tasks: T6800
Differential Revision: https://phab.enlightenment.org/D5856
Summary:
model based list need to be under the namespace of 'Efl.Ui.View".
thus, I renamed 'Efl.Ui.List' to 'Efl.Ui.View.List' properly.
Test Plan: N/A
Reviewers: cedric, felipealmeida
Differential Revision: https://phab.enlightenment.org/D5855
Function pointers now go through the same argument marshalling pipeline
as normal functions.
This will enable interfaces like Efl.Ui.Format to work properly.
When we have an eina.Value_Native (representing an Eina_Value passed by
value) and assign it to an eina.Value (a class with an IntPtr to an
underlying Eina_Value) we copy it so the eina.Value can take ownership
and free the data normally.
A possibly better alternative would be adding an extra flag to
eina.Value (something like OwnsPointer) to check whether we should free
the struct we point to or not.
When the subchild where we request subchilds from was regular, the while
loop would have run infinitly. This is now fixed by at least calling
once _next, the check to not run outside the node is now done with
calculating the depth of the nodes.
Evas map supports anti-alias(aa) rendering on sw backened.
When aa is toggled on, map forcely turns alpha channel on while it draws on the surface.
Actually, it was intended to blend polygon edges with destination,
but it breaks one case if the original source image alpha channel were turned off.
Simply, it fixed the issue, new implmentation removes the alpha channel switching,
instead fill the alpha values with 255 when map + aa + alpha_off is drawing on it.
@fix T1975
Evas map anti-aliasing haven't worked at all if the smooth scaling were disabled.
evas map rendering has a lot of corner-cases, previous call-position was wrong,
(by mistake maybe) shouldn't be in a certian case.
Let aa post-processing function be performed in universally.
The highlight should start on the second genlist item,
instead nothing is highlighted on first run.
The highlight come back to normal if you press up/down,
or if you give/remove the focus to the window 2 times.
Ector renderer are created by a factory pattern which return a reference.
The possible logic improvement from here would be to rely on setting the
parent of the object to the Ector_Surface and manage a cache by intercepting
efl_noref to trigger an efl_del of the object or a temporary cache of it.
That's for later.
Summary:
This copyedits the descriptions for several _add and _free APIs.
Phrasing is made a bit more concise, without losing the meaning, and
made more consistent so the distinctive differences between the routines
are easier to spot. Some paragraph spacing is added for clarity, for
similar reasons.
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D5851
Reviewed-by: Cedric BAIL <cedric@osg.samsung.com>
127 is the "command not found" shell exeit code, 126 is "the command
file is found but is not executable" which i think i'd interpret not
just for execute permissions but that something is preventing it from
executing in general.
both exe and thread objects must (currently) stay around until the
child thread or exe (task) is done. if you don't do this "bad things
can happen". so produce an error to let the programmer know.
it wasn't looking at the copier queue "todo" at all thus skipping it
on flush. this will fix flushing of ecore ipc message buffers. this
happens to also now fix terminology's "one process" mode.
@fix
We should only have to actually force a tick if we can't set up
the tick job, as the evas changed callback or a following render
from the tick job's tick will cause a flip.
Should remove a potential for a missed tick near the start of
ticking.
Previously, when an error happened in Eolian, the state was left
in a presumably unusable and inconsistent condition. This work
aims to change that, as all changes are committed into a staging
area before being validated and merged back into main state.
This is not yet complete, as units and by-file lookups are not
currently involved in the rollback. This will change in the
subsequent commits.
@feature
Instead of
var bg = efl.ui.Background.static_cast(myobj.Part("background"));
Now do
var bg = myobj.Background;
Also a couple helper functions were added.
Now the generated DllImport call will have the actual library name
instead of placeholders on Linux. Before we relied on dlsym being able
to search for the symbols on all loaded so's while Windows needed you to
specify the correct library to call GetProcAddress.
This fixes the case when someone call an function from an efl interface
defined in efl.so (e.g. efl_part) and the library is not loaded yet.
Apparently when we initiate a client side move in ecore_wl2 we flag that
and send a mouse-up immediately on the next pointer enter.
Do the same for resize.
At some point this might need to be revisited, we should probably be
sending a "cancel" at the start of client initiated move/resize instead
of an up at the end?
Fix T6422
Unlike panic, this will be used to handle regular errors
such as parse errors. There will be no jumps and you
will be able to pass in a pointer to get the error
data into some local memory. That way you will be
able to override printing error messages.
In general, ptr(struct) parameters behavior depends whether the
parameter has the @owned modifier or not.
If there is no @owned parameter (meaning no transfer of ownership
happens) and it is a "complex" struct, with reference type fields
(like strings), the struct is converted to the respective
<Struct>Internal struct and passed with "ref" to the DllImport'd
function. For @in parameters, after the function it returns, this
intermediate struct is converted to the public struct type and
assigned to the original parameter, updating it to the external
world.
When we have ownership transfers, the structure is copied to unmanaged
memory and given to the callee. We can't send managed memory directly as
the callee may try to free it. On the managed side, the original struct
is left to be garbage collected normally.
When transferring the ownership of a ptr(struct) from Unamanaged to
managed, we should marshal the reference-typed fields or they can point
to Managed memory that would make no sense to access from C (for
example, strings would return garbage).
The downside is that it can cause potential leaks if the receiver of the
memory do not free it. In the current state of the EFL api this can
happen in two places:
- Efl.Ui.List.SegArray::remove
return -> ptr(Efl.Ui.List.LayoutItem) @owned
- Efl.Ui.Focus.Manager::fetch
return -> ptr(Efl.Ui.Focus.Relations) @owned
The resources copied by both structs may leak when those functions are
overriden in C# and the values returned to the C code.
Also hide some internal stuff instead of exporting it and generate
implicit conversion operators for struct external/internal.
Summary:
This will avoid infinite loops and errors when the parent tries
to orphan an invalidated child.
Fixes T6780
Test Plan: Run `make check`
Reviewers: cedric
Maniphest Tasks: T6780
Differential Revision: https://phab.enlightenment.org/D5839
Reviewed-by: Cedric BAIL <cedric@osg.samsung.com>
Similarly to commit 8b962d24ac prevent
extra flips in the drm engine.
These are potentially more harmful to DRM as they can cause frame
drops and missed render opportunities.
It's possible that we unregister then reregister an animator so quickly
that the ecore_evas idle enter/exiters haven't had a chance to run yet.
In this case a render will come shortly anyway and we shouldn't use a
'false' commit to kick off a frame callback.
Silences an ERR and unmeasurably improves protocol utilization.
settings the redirect might result in unfocus calls, that might unset
evas_object_focus on a object that might be registered in two managers
due to border linking. This fixes mysterical disapearing focus.
We dont delete the adapeter when we are still focused, to set anyway the
correct view to it, we need to set the view to the adapeter as often as
possible
somehow genlist leaks the view sometimes, thus the adapter is not
deleted when the item is deleted. This resulted in strange ghost focus
objects in the window.
after that someone probebly wants to check where the last focus was on,
if the set of boder elements has changed since the last border flush, then the
element might not be registered in the parent manager
ecore_wl2_window_commit() must be called during window size negotiation,
but this currently trips a warning when no frame callback has been
received for the first commit. We can't even have frame callbacks at
that point because no buffer is attached.
Don't set up the commit_pending logic until after we have a buffer.
This patch implements bug compatibility.
genlist internally uses both a tree structure with Eina_List and a flat
Eina_Inlist to track its items. ALL of the items are in the inlist while
subitems appear in their parent's list. As a consequence both lists must
be kept in sync pretty tightly. Obviously this is not done at all and
has led to countless bugs, as soon as tree or groups are used:
- Invalid order of items (visually)
- Invalid order of items with sorted_insert
- Glitches in the matrix
- Crashes with sorted_insert
- Odd/even styles not properly set
- Promote/demote functions broken by design
- Developers send to psychiatric hospitals
- Etc...
Legacy genlist (1.19 and before) used an inlist order that basically
didn't make sense, as it didn't follow the logical order of elements (as
they appear visually). Unfortunately this has "worked" (really, that's a
huge stretch to use this word here) for a long time this way. As a
consequence, some applications (*cough* empc *cough*) have relied on
this order to implement "next album" or "previous album" where the
album title is a group node.
By changing the order of items in the inlist, this has broken the
assumptions made above, and ends up in cases that return NULL, leading
to SEGV. Sure, the app should have checked NULL, but that's not really
the point here. The behavior has been changed.
This patch implements "fixes" for the following functions:
- elm_genlist_first_item_get(): Don't return a parent
- elm_genlist_last_item_get(): Return a parent
- elm_genlist_item_next_get(): return a parent upon reaching the last child
- elm_genlist_item_prev_get(): return a child when a parent is passed
Important notes:
- This does not cover 100% behavior compatibility here. The only way to
have it would be to simply revert the entire genlist code to its
original version and never touch it again, ever.
- An explicit API is required for an application to specify which API
level it targets, so that we can cherry-pick which bug compatibility
features we want to enable. We are already doing this for EDC,
unfortunately.
@fix
fix T5938
Signed-off-by: Mike Blumenkrantz <zmike@osg.samsung.com>
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Mike Blumenkrantz
YeongJong Lee
Alastair Poole
Carsten Haitzler
Stefan Schmidt
Cedric Bail
Xavi Artigas
Daniel Kolesa
Shinwoo Kim
Hermet Park
WooHyun Jung
Derek Foreman
Marcel Hollerbach
SangHyeon Lee
Lauro Moura
Felipe Magno de Almeida
Davide Andreoli
Andy Williams
Bryce Harrington
Arnaud Ferraris
Hosang Kim
Jihoon Kim
Woochan Lee
Jean-Philippe Andre