Fix management of safe-for-nodes flag in ResourceLoader and WorkerThreadPool

This commit is contained in:
Pedro J. Estébanez 2023-07-03 12:27:57 +02:00
parent 46424488ed
commit 30925fcfb0
2 changed files with 8 additions and 3 deletions

View File

@ -309,6 +309,10 @@ void ResourceLoader::_thread_load_function(void *p_userdata) {
}
// --
if (!Thread::is_main_thread()) {
set_current_thread_safe_for_nodes(true);
}
Ref<Resource> res = _load(load_task.remapped_path, load_task.remapped_path != load_task.local_path ? load_task.local_path : String(), load_task.type_hint, load_task.cache_mode, &load_task.error, load_task.use_sub_threads, &load_task.progress);
if (mq_override) {
mq_override->flush();

View File

@ -56,6 +56,8 @@ void WorkerThreadPool::_process_task(Task *p_task) {
Task *prev_low_prio_task = nullptr; // In case this is recursively called.
if (!use_native_low_priority_threads) {
// Tasks must start with this unset. They are free to set-and-forget otherwise.
set_current_thread_safe_for_nodes(false);
pool_thread_index = thread_ids[Thread::get_caller_id()];
ThreadData &curr_thread = threads[pool_thread_index];
task_mutex.lock();
@ -179,9 +181,6 @@ void WorkerThreadPool::_process_task(Task *p_task) {
if (post) {
task_available_semaphore.post();
}
// Engine/user tasks can set-and-forget, so we must be sure it's back to normal by the end of the task.
set_current_thread_safe_for_nodes(false);
}
}
@ -371,7 +370,9 @@ Error WorkerThreadPool::wait_for_task_completion(TaskID p_task_id) {
must_exit = true;
} else {
// Solve tasks while they are around.
bool safe_for_nodes_backup = is_current_thread_safe_for_nodes();
_process_task_queue();
set_current_thread_safe_for_nodes(safe_for_nodes_backup);
continue;
}
} else if (!use_native_low_priority_threads && task->low_priority) {