The allocations of commands in CommandQueueMT weren't aligned. This
commit aligns all accesses on 64bit boundaries regardless of target
platform. This ensures that all types are aligned.
Lock-wise the semaphores were maked as usable when the command had ran
but not when the synchronous stub had finished with it. This lead to a
race condition where sometimes the semaphore got reused before it was
waited on. We now mark the semaphore as free only once we're done
waiting on it.
-Changed Animation to have a special signal when tracks are changed, to avoid unnecesary track cache rebuilds in AnimationPlayer
-Added missing emit_changed whe modifying keys to Animation
-Changed AnimationPlayer to use the new refcounted connections instead of the previous hacky way to keep references
-Changed AnimationEditor to update the current track when keys are edited
-Fixed bug where undo/redo did not work with AnimationKeyEdit (was not being updated)
-Made sure UndoRedo does not mind deleted objects in undo/redo history, this would corrupt the history or clear it without need.
Only possible if the object class is a "native type". If the object class is a user class (that derives a "native type") then a script is needed.
Since CSharpLanguage does cleanup of script instance bindings when finished, cases like #25621 will no longer cause problems.
Fixed ~Object() trying to free script instance bindings after the language has already been removed, which would result in a NULL dereference.
Avoid CSharpInstance from accessing its state after self destructing (by deleting the Reference owner).
It's now safe to replace the script instance without leaking or crashing.
Also fixed godot_icall_Object_weakref return reference being freed before returning.
This allows most demos to run without any ubsan or asan errors. There
are still some things in thirdpart/ and some things in AudioServer that
needs a look but this fixes a lot of issues. This should help debug less
obvious issues, hopefully.
This fixes#25217 and fixes#25218
A GDScript call to:
`multiplayer.network_peer.some_prop = true`
seems to transalte to:
```
var temp = multiplayer.network_peer
temp.some_prop = true
multiplayer.network_peer = temp
```
Which caused the MultiplayerAPI to be resetted.
The call to set_network_peer is now ignored if the peer that's beeing
set is the same as the currently set one.
rpc_sender_id is now correctly initialized to 0 so get_rpc_sender_id()
work reliably even if called before receiving any RPC.
root_node is initialized to NULL (fix crashes when incorrectly using the
MultiplayerAPI).
clear function now resets the packet cache size to free more memory when
not running.
It has a big impact on 2D and text rendering performance (cf. #24466)
so the solution seems worse than the bug it aims to work around.
It's now opt-in via "rendering/quality/2d/gles2_use_nvidia_rect_flicker_workaround"
for those who need it and have a simple enough game for the performance
drop not to be an issue.
Fixes#24466.
Some used 'is_valid()' checks, others not. Validity is already checked in 'unref()',
and 'remove_resource_format_*()' has an ERR_FAIL condition on 'is_null()' already
(which shouldn't happen since we're only unregistering things that we previously
registered.
Also add missing GDCLASS statement in ResourceFormatLoaderVideoStreamGDNative,
missed in #20552 which was last amended before #19501 was merged.
During reloading in `GDScriptLanguage::reload_all_scripts` a placeholder instance that must remain so is replaced with a new placeholder instance. The state is then restored by calling `ScriptInstance::set` for each property. This does not work if the script is missing the properties due to build/parse failing.
The fix for such cases is to call `placeholder_set_fallback` instead of `set` on the script instance.
I took this chance to move the `build_failed` flag from `PlaceHolderScriptInstance` to `Script`. That improves the code a lot. I also renamed it to `placeholder_fallback_enabled` which is a much better name (`build_failed` could lead to misunderstandings).
Background: lstrip and rstrip were broken by changes to String in:
0e29f7974b
which removed it's access to Vector::find(CharType).
Moved Vector's find up into CowData so it can be shared by Vector and String.
Added String::find_char using CowData::find.
Implemented rstrip and lstrip using find_char.
Added a few tests for String rstrip and lstrip.
As per the C++ standard 21.3.4.1 for std::string:
Returns: If pos < size(), returns data()[pos]. Otherwise, if pos ==
size(), the const version returns charT(). Otherwise, the behavior is
undefined.
Since the behavior is undefined Godot now does the same thing for const
and non-const versions of operator[].
This fixes#21242 and fixes#22221.
Relying on various compiler primitives we can reduce the work done
in our memory allocators and CowData. For types with trivial ctors or
dtors we can skip looping over all elements when creating, resizing,
and destroying lists of objects.
These primitives are supported by clang, msvc, and GCC. However, once
we've moved to C++11 we can rely on several std:: primitives that do
the same thing and are standardized.
In my testing the extra conditionals introduced here get removed from
the generated program entirely as the results for these primitives is
known at compile time.
When opening projects for edition through the project manager, the
following checks are now done:
1. If the config_version is lower than the one used by the current
engine version, users are asked if they want to convert to the new
format or abort editing. Fixes#20626.
2. If the config_version is higher than the expected one (project
from a more recent and incompatible engine version), projects are
grayed out and can't be edited. Fixes#18758.
When editing from the command line, the behaviour is unchanged:
projects in situation (1) are automatically converted, while projects
in situation (2) show an error message (made more explicit).
The "Run" option from the project manager was not changed, so it will
still run (1) projects without converting them, and fail running (2)
projects.
Co-authored-by: groud <gilles.roudiere@gmail.com>
By introducing an intermediate proxy class for the array subscript
operator for String and CharString we can control better when CowData
will actually CoW.
This should improve performance of String usage for most cases.
By relying on the fact that a struct or class's first member has the
same address as the struct itself we can cast VectorWriteProxy<T> to
Vector<T> and access the CowData field.
This allows a Vector to be moved in memory without invalidating the
pointer to the cowdata field.
