Recently I have seen that eeze_suite times out more often on jenkins when
being under load. Looking in this revealed that we have been using the default
4s timeout for eeze while almost all others suites had the timeout disabled
already. In eeze there should be no test that would really block and if I find
one I can simply add a specific timeout like we did for eina_barrier.
Situation:
- Evas Object A has a clip C and a map M.
Problem:
- Clip C will be applied once inside the map surface S and
again when the surface S is drawn to the canvas.
Solution:
- Track whether the current object is the mapped object
or a child of the mapped object. In the first case,
discard the clipper when rendering to the map surface.
In the second case, the child's clipper is PROBABLY[*]
inside the map, so it should be applied when rendering the
map surface itself.
Note: This also applies to masks.
[*] This is clearly not the ultimate clipping fix.
Masking in GL assumes only one texture to sample from. This means
RGB+Alpha and YUV types are not supported. While it would
make sense for RGB+Alpha, it doesn't make any sense for YUV (because
masks are alpha planes and YUV is opaque...)
Added optional constructor methods for C++ Eolian wrappers.
Changed the interface of wrappers' main constructors.
If there are optional constructor methods they should be passed as variadic
template argument at the end of the constructor.
To support variadic template arguments, the optional "parent" parameter is
now the first parameter and there is another constructor without the
"parent" parameter.
Checking for @optinal and @nullable attributes instead of @nonull.
Now @nonull is the default, and eina::optional is only used when @optional
or @nullable attribute is specified.
The names of constructor methods no longer have the class name prefixed.
Added unit tests for checking the binding of optional constructors.
Added new .eo file to be used in the test.
Changed the generated documentation of constructors.
Changed the efl::eo::inherit accordingly, to address these new features.
Now the constructor methods should be explicit called in the
efl::eo::inherit constructor, which will receive them via variadic
template arguments.
Added another constructor to efl::eo::inherit for passing the parent
object.
Updated some tests and examples to follow the new interface.
Removed some code that is no longer necessary.
Also, fix Eolian C++ support for constructing properties. fix
assertion when parsing constructing properties.
Now if a property is a constructing property eolian_cxx will generate a
constructor method that have the property name (without the "_set" suffix).
Now an eina::optional that wraps a specific type can be constructed or
have content assigned to it using another eina::optional that wraps a
different type, provided that the two wrapped types are convertible
between each other.
Added "disengage" member function to be able to disengage a R-value
eina::optional of different contained type.
It also adds for increased convenience.
Fix constructing an engaged eina::optional from an disengaged one.
Fix small assertion problem of trying to construct an eina::optional
already flagged as engaged.
Fix incorrect use of std::move. Changed it to std::forward.
Added constexpr property for trivial constructors.
Added auxiliary function "make_optional".
Added unit test to check compatibility between eina::optionals of
convertible types.
Fixed crash when disconnecting event inside of its own event callback.
Instead of deleting the callback object immediately during disconnection
(which causes the callback to be freed), the deletion is now scheduled
for later (using ecore_main_loop_thread_safe_call_async).
Updated some Makefiles to proper include ecore now that it is used in
all event wrappers.
Added a unit test to verify crashes under these circumstances.
Using eina::string_view in eolian generated interfaces (instead of
std::string) to allow lightweight passing of both C strings and C++
std::string.
Also, No longer using eina::optional in generated headers for types
that already implements the concept of null state (like Eo wrappers
and eina_accessor).
Also fix allocating callback objects require by class methods
(i.e. static) in static vectors so the memory will be freed when the
programs exit.
Added a new test case for testing callbacks on class methods.
Moved method definitions and supplementary code from generated C++
wrappers to auxiliary header file (.eo.impl.hh) generated together
with the main ".eo.hh" file. Updated Makefiles to list such files in
the compilation and cleanup processes. Updated .gitignore to include
these new generated files.
Made general adjustments on the documentation of generated C++ wrappers
Added "PREDEFINED" preprocessor macro definition in the Doxyfile.in in
order to make some adjustments for better documentation in the C++
generated headers. Excluding generation of documentation for classes
in the "eo_cxx" namespace (the namespace for "abstract" eolian C++
wrappers). Now generating the documentation for the events too.
Hiding some auxiliary code from being documented. Some aesthetic
adjustments for generated white space. Generate documentation for the
main constructor of C++ wrappers and added auxiliary grammars to list
parameters names.
Add an implementation of string_view to Eina C++. It is a non owning
reference to a string that allows lightweight argument passing of both
C++ std::string and C strings (const char*) on interfaces.
Added declaration of all inherited events on the concrete type to
avoid unnecessary type conversion on event callbacks.
Note that when two events happens to have the same name, only the
event of the derived class is declared in the concrete class. Access to
such events should be made through the base class type specification.
Fix on forwarding callback object (changed std::move to std::forward).
Additional line break at the end of event handling methods in the
generated headers to improve readability.
Summary: Fix coverity CID1294212 (potential Null pointer dereference) as
the alloc for subdir_cache Could fail, however if it does we Were
still trying to use it. In the case that it fails, ERR msg and return.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
It seems like before the switch to Eo, all the object_*_add functions
had magic checks that only allowed non NULL Evas pointers to be passed.
Since we switched to Eo we apparently missed the early failure we used
to have, and we instead fail later. This patch should fix that.
Thanks to jaehwan for reporting.
@fix
This problem have been observed by Scimmia22 on the Arch builds as well as on
the jenkins wayland build job. While compiling works fine the relinking during
make install fails with ecore-drm linking does not find eeze or eldbus as its
deps. This only shows on systems with no efl installed, a build from scratch.
As far as I can see we have all dependencies set correctly in configure as
well as in the Makefile which are working fine even in highly parallel builds.
It was a bit surprising here to me that the include order is still important
with our correct dependencies. Autotools wisdom is welcome here to either
explain to me why this is needed or what the correct fix would be.
The includes all moved before Ecore_Evas because that would use ecore_drm if
enabled.
Fixes T2281
This should fix Dave's crash with the GL engine.
Indeed the pointer to the color array is passed directly
to the driver, without memcpy. Then, during glDrawArrays, the
driver will try to address it... but it could as well be NULL or
invalid. So, crashes would occur.
Also mark use_vertex as always true. We keep this field only for the
alloc() function.
Since the new shader selection system is able to check the
value of nomul, we can avoid pushing color when they are not
required (color is opaque white).
Hopefully this will fix Dave's crash (mayyyyybe...)
Summary: This fixes Coverity CID1270028, CID1270029, CID1270030 where
variables RGBA_Image and Native where dereferencing function input
paramaters Before null checking them.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: This fixes Coverity CID1293519 where einfo was being used
Before it was being null checked (which Could have caused a crash if
in fact einfo Was NULL).
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: Fix Coverity CID1287154. _alloc function can return null
however 'color' was never checked after call to _alloc and thus Could
be NULL.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: This fixes a reported resource leak (by coverity).
_pack_meshes_vertex_data allocates memory stored in 'vertices' however
vertices is never free'd unless count > 0
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: This fixes an issue of coverity reporting Explicit null
dereference as fetchfunc is defaulted to NULL, yet no check is made
later before calling it.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: This fixes a resource leak of eina_array in
eng_ector_renderer_draw which would return without freeing an
allocated array.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary: This option Was added originally so that software drm and
hardware drm could be done in the same 'engine'. Since we have drm and
gl_drm now as separate engines, this option is no longer needed.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
efreetd was losing the last char on every line of extra_icon.dirs and
icon.exts thus resulting in gradual degredation of these files as more
and more changes happen to have things like:
.jpeg
.jpe
.jp
.j
.
etc. for extension or dirs like
/home/raster/.e/e/icons
/home/raster/.e/e/icon
/home/raster/.e/e/ico
...
/home/ras
/home/ra
/home/r
...
/ho
/h
/
.. you get the idea. before long the list of extra icon dirs (and
extensions) was massigve and caued all sorts of filesystem rummaging.
this fixes that to no longer degrade these files. this also changes
their names to force new files to be used instead of the broken old
ones.
if you unlink then rename, there is a gap between the unlink and
rename - no file is there. the point of the rename is to be atomic.
either old file OR new file. nothing in between. this fixes that race.
this fixes major slowness in things like enlightenment login. when
enlightenment starts, efreetd is started. efreetd takes a long time to
start because it is scanning lots of directories recursively. it is
early in code freeze, so i think this can go in as a fix, even though
it is more of an optimization, but as such it has a fairly major speed
impact, so i consider this fixing a core performance problem for
things like logging in. The solution is a cache file containing all
the sub directories in a directory tree. we still have to stat every
directory, but this avoids a lot of stating of all files as well and
avoids any readirs etc. so it is much faster. on an ssd this comes out
to 4 times faster for efreetd to start up. on an hdd it's about 2x
faster to start uncached.
this should bea good fix for startup times - on my systems thats a 1
second speedup on ssd (out of about 8 seconds boot time) and 3 seconds
on hdd (out of about 39 seconds boot). so w save 1 and 3 seconds
respecively in boot + login.
now can we get this to 0 or close? that's a matter of designing
something like a deferred scan + monitor add so we wait until "startup
is done" then set up in the bg for a bit. that might shave another 3
seconds off boot time for hdd's but for ssd's wil barely blip (maybe
save 0.1 sec). so leave that till normal dev mode.
@fix
i found my efreetd was not just monitoring a small set of dirs but
literally $HOME - recursively. this explains a lot of performance
issues i have seen on spinning hdd's vs ssd's. i never knew if was in
efreetd though... until now. for whatever reason my efret cache caused
efreet to try monitro a list like:
/
/h
/ho
/hom
/home
/home/
/home/r
... etc.
when trying to monitor something like:
/home/raster/.local/share/icons
that means it ended up trying $HOME as wellas a vast range of dirs it
just shouldn't sensibly ever try. i have no idea how my cache ended up
this way, but deleting it and re-populating it fixed it. so this ads
code to detect such insanity (as wellas give actual complaints on deep
recursion so we can see things better later). not a bug fix as such,
but a major improvement to detecting bugs which is what we want for
stability.
I also noticed that one legacy pixmaps dir is not recursively
monitored, but another is - so be consistent and recurs both.
Also apply checks to desktop file monitoring too.