`ClassDB::can_instantiate()` and other reflection methods deadlock if the type is an script global class, when such script indirectly uses a not-yet-registered class. The reason is the `ClassDB` read lock is still held when invoking the `ResourceLoader` to load the class script, which may in turn need to lock for writing (for the class registration).
In particular, this happens with some types related to animation tree, that aren't registered at engine startup, but can happen with others, especially ones from the user. Registration statements are also added for the animation-related types that were lacking them.
Fixes#81758
DisplayServerWeb::process_joypads handles buttons 6 and 7 of the
HTML5 Standard Gamepad as a special case by doing:
`input->joy_axis(idx, (JoyAxis)b, s_btns[b]);`
This doesn't work because there is no JoyAxis 6 or 7 in the enum
To fix this we use JoyAxis::TRIGGER_LEFT and TRIGGER_RIGHT for button 6
and 7
However since we are now lying to input->joy_axis we also need to lie in
the mappings for the standard gamepad in godotcontrollersdb.txt,
otherwise input->joy_axis will try to find a mapping to axis 4(LT) and
axis 5(RT) that's not defined.
Therefore we set lefttrigger to +a4 and righttrigger to +a5 in the
mapping, to match what we are actually sending.
A cleaner, and more involved fix to this would be modifying
input->joy_button so that it can handle analog buttons and map them to
axes preserving their value instead of converting to boolean
(cherry picked from commit 9dd372f316)
Before this change StringName used regular static field
definitions for its mutex, _table, configured and debug_stringname
fields.
Since in the general case the ordering of the static variable and field
initialization and destruction is undefined, it was possible that
the destruction of StringName's static fields happened prior to
the destruction of statically allocated StringName instances.
By changing the static field definitions to inline in string_name.h,
the C++17 standard guarantees the correct initialization and destruction
ordering.
(cherry picked from commit 723878bbab)
`GDExtension::open_library` has a check in it to see if the library was loaded
from a temp file, and if it was to restore the original name as that is the one
we actually care about. This check is breaking extension reloading on Mac when
the library path is to a framework folder, as the file inside the framework
will not generally be the same name as the folder.
This check also shouldn't be necessary even on Windows, which is the only
platform that uses `generate_temp_files`, since disposal of the created temp
file is handled within `OS_Windows::open_dynamic_library`, and
`GDExtension::open_library` (which is the only function to call
`open_dynamic_library` with a `p_data` argument) only cares about the original
library file path and has to do extra work to remove the name of the temp file.
Instead, I have removed that check and set `OS_Windows::open_dynamic_library`
to return the name of the original file and not the name of the copy.
This fixes GDExtension reloading on macOS. I do not have a Windows machine
available to test that it still works properly on Windows, so someone should
check that before merging this.
(cherry picked from commit f44d6a235f)
This fixes a rare but possible deadlock, maybe due to undefined behavior. The new implementation is safer, at the cost of some added boilerplate.
(cherry picked from commit f4d76853b9)
1. Make handling of user tokens atomic:
Loads started with the external-facing API used to perform a two-step setup of the user token. Between both, the mutex was unlocked without its reference count having been increased. A non-user-initiated load could therefore destroy the load task when it unreferenced the token.
Those stages now happen atomically so in the one hand, the described race condition can't happen so the load task life insurance doesn't have a gap anymore and, on the other hand, the ugliness that the call to load could return `ERR_BUSY` if happening while other thread was between both steps is gone.
The code has been refactored so the user token concerns are still outside the inner load start function, which is agnostic to that for a cleaner implementation.
2. Clear ambiguity between load operations running on `WorkerThreadPool`:
The two cases are: single-loaded thread directly started by a user pool task and a load started by the system as part of a multi-threaded load.
Since ensuring all the code dealing with this distinction would make it very complex, and error-prone, a different measure is applied instead: just take one of the cases out of the dicotomy. We now ensure every load happening on a pool thread has been initiated by the system.
The way of achieving that is that a single-threaded user-started load initiated from a pool thread, is run as another task.
(cherry picked from commit df23858488)
Benefits:
- Simpler code. The main load function is renamed so it's apparent that it's not just a thread entry point anymore.
- Cache and thread modes of the original task are honored. A beautiful consequence of this is that, unlike formerly, re-issued loads can use the resource cache, which makes this mechanism much more performant.
- The newly added getter for caller task id in WorkerThreadPool allows to remove the custom tracking of that in ResourceLoader.
- The check to replace a cached resource and the replacement itself happen atomically. That fixes deadlock prevention leading to multiple resource instances of the same one on disk. As a side effect, it also makes the regular check for replace load mode more robust.
(cherry picked from commit 28619e26cf)
This switches to 64-bit integers in select locations of the Image
class, so that image resolutions of 16384×16384 (used by
lightmap texture arrays) can be used properly. Values that are larger
should also work.
VRAM compression is also supported, although most VRAM-compressed
formats are limited to individual slices of 16384×16384. WebP
is limited to 16383×16383 due to format limitations.
This is about not letting the resource format loader set the error code directly on the task anymore. Instead, it's stored locally and assigned only when it is right to do so.
Otherwise, other tasks may see an error code in the current one before it's state having transitioned to errored. While this, besides the technically true data race, may not be a problem in practice, it causes surprising situations during debugging as it breaks assumptions.
