Merge pull request godotengine#140 from jss2a98aj/cherry-pick-4.4-ass…

…orted

[4.4 backport] Assorted fixes

SConstruct

@@ -807,6 +807,7 @@ if env.msvc and not methods.using_clang(env):  # MSVC
                 "/wd4245",
                 "/wd4267",
                 "/wd4305",  # C4305 (truncation): double to float or real_t, too hard to avoid.
+                "/wd4324",  # C4820 (structure was padded due to alignment specifier)
                 "/wd4514",  # C4514 (unreferenced inline function has been removed)
                 "/wd4714",  # C4714 (function marked as __forceinline not inlined)
                 "/wd4820",  # C4820 (padding added after construct)

core/os/spin_lock.h

@@ -33,6 +33,10 @@
 
 #include "core/typedefs.h"
 
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
 #if defined(__APPLE__)
 
 #include <os/lock.h>
@@ -52,19 +56,52 @@ class SpinLock {
 
 #else
 
+#include "core/os/thread.h"
+
 #include <atomic>
 
-class SpinLock {
-	mutable std::atomic_flag locked = ATOMIC_FLAG_INIT;
+_ALWAYS_INLINE_ static void _cpu_pause() {
+#if defined(_MSC_VER)
+// ----- MSVC.
+#if defined(_M_ARM) || defined(_M_ARM64) // ARM.
+	__yield();
+#elif defined(_M_IX86) || defined(_M_X64) // x86.
+	_mm_pause();
+#endif
+#elif defined(__GNUC__) || defined(__clang__)
+// ----- GCC/Clang.
+#if defined(__i386__) || defined(__x86_64__) // x86.
+	__builtin_ia32_pause();
+#elif defined(__arm__) || defined(__aarch64__) // ARM.
+	asm volatile("yield");
+#elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) // PowerPC.
+	asm volatile("or 27,27,27");
+#elif defined(__riscv) // RISC-V.
+	asm volatile(".insn i 0x0F, 0, x0, x0, 0x010");
+#endif
+#endif
+}
+
+static_assert(std::atomic_bool::is_always_lock_free);
+
+class alignas(Thread::CACHE_LINE_BYTES) SpinLock {
+	mutable std::atomic<bool> locked = ATOMIC_VAR_INIT(false);
 
 public:
 	_ALWAYS_INLINE_ void lock() const {
-		while (locked.test_and_set(std::memory_order_acquire)) {
-			// Continue.
+		while (true) {
+			bool expected = false;
+			if (locked.compare_exchange_weak(expected, true, std::memory_order_acquire, std::memory_order_relaxed)) {
+				break;
+			}
+			do {
+				_cpu_pause();
+			} while (locked.load(std::memory_order_relaxed));
 		}
 	}
+
 	_ALWAYS_INLINE_ void unlock() const {
-		locked.clear(std::memory_order_release);
+		locked.store(false, std::memory_order_release);
 	}
 };
 

core/os/thread.h

@@ -42,6 +42,8 @@
 #include "core/templates/safe_refcount.h"
 #include "core/typedefs.h"
 
+#include <new>
+
 #ifdef MINGW_ENABLED
 #define MINGW_STDTHREAD_REDUNDANCY_WARNING
 #include "thirdparty/mingw-std-threads/mingw.thread.h"
@@ -85,6 +87,20 @@ class Thread {
 		void (*term)() = nullptr;
 	};
 
+#if defined(__cpp_lib_hardware_interference_size) && !defined(ANDROID_ENABLED) // This would be OK with NDK >= 26.
+#if defined(__GNUC__) && !defined(__clang__)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Winterference-size"
+#endif
+	static constexpr size_t CACHE_LINE_BYTES = std::hardware_destructive_interference_size;
+#if defined(__GNUC__) && !defined(__clang__)
+#pragma GCC diagnostic pop
+#endif
+#else
+	// At a negligible memory cost, we use a conservatively high value.
+	static constexpr size_t CACHE_LINE_BYTES = 128;
+#endif
+
 private:
 	friend class Main;
 
@@ -135,6 +151,8 @@ class Thread {
 
 	typedef uint64_t ID;
 
+	static constexpr size_t CACHE_LINE_BYTES = sizeof(void *);
+
 	enum : ID {
 		UNASSIGNED_ID = 0,
 		MAIN_ID = 1

core/string/ustring.cpp

@@ -221,18 +221,35 @@ void CharString::copy_from(const char *p_cstr) {
 /*  String                                                               */
 /*************************************************************************/
 
-Error String::parse_url(String &r_scheme, String &r_host, int &r_port, String &r_path) const {
-	// Splits the URL into scheme, host, port, path. Strip credentials when present.
+Error String::parse_url(String &r_scheme, String &r_host, int &r_port, String &r_path, String &r_fragment) const {
+	// Splits the URL into scheme, host, port, path, fragment. Strip credentials when present.
 	String base = *this;
 	r_scheme = "";
 	r_host = "";
 	r_port = 0;
 	r_path = "";
+	r_fragment = "";
+
 	int pos = base.find("://");
 	// Scheme
 	if (pos != -1) {
-		r_scheme = base.substr(0, pos + 3).to_lower();
-		base = base.substr(pos + 3, base.length() - pos - 3);
+		bool is_scheme_valid = true;
+		for (int i = 0; i < pos; i++) {
+			if (!is_ascii_alphanumeric_char(base[i]) && base[i] != '+' && base[i] != '-' && base[i] != '.') {
+				is_scheme_valid = false;
+				break;
+			}
+		}
+		if (is_scheme_valid) {
+			r_scheme = base.substr(0, pos + 3).to_lower();
+			base = base.substr(pos + 3, base.length() - pos - 3);
+		}
+	}
+	pos = base.find("#");
+	// Fragment
+	if (pos != -1) {
+		r_fragment = base.substr(pos + 1);
+		base = base.substr(0, pos);
 	}
 	pos = base.find("/");
 	// Path

core/string/ustring.h

@@ -452,7 +452,7 @@ class String {
 	String c_escape_multiline() const;
 	String c_unescape() const;
 	String json_escape() const;
-	Error parse_url(String &r_scheme, String &r_host, int &r_port, String &r_path) const;
+	Error parse_url(String &r_scheme, String &r_host, int &r_port, String &r_path, String &r_fragment) const;
 
 	String property_name_encode() const;
 

drivers/d3d12/rendering_device_driver_d3d12.cpp

@@ -2177,17 +2177,32 @@ void RenderingDeviceDriverD3D12::command_pipeline_barrier(CommandBufferID p_cmd_
 	}
 
 	// Define the barrier groups and execute.
+
 	D3D12_BARRIER_GROUP barrier_groups[3] = {};
-	barrier_groups[0].Type = D3D12_BARRIER_TYPE_GLOBAL;
-	barrier_groups[1].Type = D3D12_BARRIER_TYPE_BUFFER;
-	barrier_groups[2].Type = D3D12_BARRIER_TYPE_TEXTURE;
-	barrier_groups[0].NumBarriers = global_barriers.size();
-	barrier_groups[1].NumBarriers = buffer_barriers.size();
-	barrier_groups[2].NumBarriers = texture_barriers.size();
-	barrier_groups[0].pGlobalBarriers = global_barriers.ptr();
-	barrier_groups[1].pBufferBarriers = buffer_barriers.ptr();
-	barrier_groups[2].pTextureBarriers = texture_barriers.ptr();
-	cmd_list_7->Barrier(ARRAY_SIZE(barrier_groups), barrier_groups);
+	uint32_t barrier_groups_count = 0;
+
+	if (!global_barriers.is_empty()) {
+		D3D12_BARRIER_GROUP &barrier_group = barrier_groups[barrier_groups_count++];
+		barrier_group.Type = D3D12_BARRIER_TYPE_GLOBAL;
+		barrier_group.NumBarriers = global_barriers.size();
+		barrier_group.pGlobalBarriers = global_barriers.ptr();
+	}
+
+	if (!buffer_barriers.is_empty()) {
+		D3D12_BARRIER_GROUP &barrier_group = barrier_groups[barrier_groups_count++];
+		barrier_group.Type = D3D12_BARRIER_TYPE_BUFFER;
+		barrier_group.NumBarriers = buffer_barriers.size();
+		barrier_group.pBufferBarriers = buffer_barriers.ptr();
+	}
+
+	if (!texture_barriers.is_empty()) {
+		D3D12_BARRIER_GROUP &barrier_group = barrier_groups[barrier_groups_count++];
+		barrier_group.Type = D3D12_BARRIER_TYPE_TEXTURE;
+		barrier_group.NumBarriers = texture_barriers.size();
+		barrier_group.pTextureBarriers = texture_barriers.ptr();
+	}
+
+	cmd_list_7->Barrier(barrier_groups_count, barrier_groups);
 }
 
 /****************/

drivers/gles3/shaders/canvas.glsl

@@ -262,15 +262,15 @@ void main() {
 
 	color_interp = color;
 
+	vertex = (canvas_transform * vec4(vertex, 0.0, 1.0)).xy;
+
 	if (use_pixel_snap) {
 		vertex = floor(vertex + 0.5);
 		// precision issue on some hardware creates artifacts within texture
 		// offset uv by a small amount to avoid
 		uv += 1e-5;
 	}
 
-	vertex = (canvas_transform * vec4(vertex, 0.0, 1.0)).xy;
-
 	vertex_interp = vertex;
 	uv_interp = uv;
 

editor/debugger/editor_debugger_server.cpp

@@ -77,8 +77,8 @@ Error EditorDebuggerServerTCP::start(const String &p_uri) {
 
 	// Optionally override
 	if (!p_uri.is_empty() && p_uri != "tcp://") {
-		String scheme, path;
-		Error err = p_uri.parse_url(scheme, bind_host, bind_port, path);
+		String scheme, path, fragment;
+		Error err = p_uri.parse_url(scheme, bind_host, bind_port, path, fragment);
 		ERR_FAIL_COND_V(err != OK, ERR_INVALID_PARAMETER);
 		ERR_FAIL_COND_V(!bind_host.is_valid_ip_address() && bind_host != "*", ERR_INVALID_PARAMETER);
 	}

editor/plugins/asset_library_editor_plugin.cpp

@@ -991,7 +991,8 @@ void EditorAssetLibrary::_request_image(ObjectID p_for, int p_asset_id, String p
 		String url_host;
 		int url_port;
 		String url_path;
-		Error err = trimmed_url.parse_url(url_scheme, url_host, url_port, url_path);
+		String url_fragment;
+		Error err = trimmed_url.parse_url(url_scheme, url_host, url_port, url_path, url_fragment);
 		if (err != OK) {
 			if (is_print_verbose_enabled()) {
 				ERR_PRINT(vformat("Asset Library: Invalid image URL '%s' for asset # %d.", trimmed_url, p_asset_id));

modules/openxr/openxr_api.cpp

@@ -1980,8 +1980,9 @@ bool OpenXRAPI::poll_events() {
 				if (local_floor_emulation.enabled) {
 					local_floor_emulation.should_reset_floor_height = true;
 				}
-				if (event->poseValid && xr_interface) {
-					xr_interface->on_pose_recentered();
+
+				if (xr_interface) {
+					xr_interface->on_reference_space_change_pending();
 				}
 			} break;
 			case XR_TYPE_EVENT_DATA_INTERACTION_PROFILE_CHANGED: {

modules/openxr/openxr_interface.cpp

@@ -1137,6 +1137,12 @@ void OpenXRInterface::process() {
 	if (head.is_valid()) {
 		head->set_pose("default", head_transform, head_linear_velocity, head_angular_velocity, head_confidence);
 	}
+
+	if (reference_stage_changing) {
+		// Now that we have updated tracking information in our updated reference space, trigger our pose recentered signal.
+		emit_signal(SNAME("pose_recentered"));
+		reference_stage_changing = false;
+	}
 }
 
 void OpenXRInterface::pre_render() {
@@ -1318,8 +1324,8 @@ void OpenXRInterface::on_state_exiting() {
 	emit_signal(SNAME("instance_exiting"));
 }
 
-void OpenXRInterface::on_pose_recentered() {
-	emit_signal(SNAME("pose_recentered"));
+void OpenXRInterface::on_reference_space_change_pending() {
+	reference_stage_changing = true;
 }
 
 void OpenXRInterface::on_refresh_rate_changes(float p_new_rate) {

modules/openxr/openxr_interface.h

@@ -70,6 +70,7 @@ class OpenXRInterface : public XRInterface {
 private:
 	OpenXRAPI *openxr_api = nullptr;
 	bool initialized = false;
+	bool reference_stage_changing = false;
 	XRInterface::TrackingStatus tracking_state;
 
 	// At a minimum we need a tracker for our head
@@ -207,7 +208,7 @@ class OpenXRInterface : public XRInterface {
 	void on_state_stopping();
 	void on_state_loss_pending();
 	void on_state_exiting();
-	void on_pose_recentered();
+	void on_reference_space_change_pending();
 	void on_refresh_rate_changes(float p_new_rate);
 	void tracker_profile_changed(RID p_tracker, RID p_interaction_profile);
 

modules/websocket/editor/editor_debugger_server_websocket.cpp

@@ -77,8 +77,8 @@ Error EditorDebuggerServerWebSocket::start(const String &p_uri) {
 
 	// Optionally override
 	if (!p_uri.is_empty() && p_uri != "ws://") {
-		String scheme, path;
-		Error err = p_uri.parse_url(scheme, bind_host, bind_port, path);
+		String scheme, path, fragment;
+		Error err = p_uri.parse_url(scheme, bind_host, bind_port, path, fragment);
 		ERR_FAIL_COND_V(err != OK, ERR_INVALID_PARAMETER);
 		ERR_FAIL_COND_V(!bind_host.is_valid_ip_address() && bind_host != "*", ERR_INVALID_PARAMETER);
 	}

modules/websocket/emws_peer.cpp

@@ -68,8 +68,9 @@ Error EMWSPeer::connect_to_url(const String &p_url, Ref<TLSOptions> p_tls_option
 	String host;
 	String path;
 	String scheme;
+	String fragment;
 	int port = 0;
-	Error err = p_url.parse_url(scheme, host, port, path);
+	Error err = p_url.parse_url(scheme, host, port, path, fragment);
 	ERR_FAIL_COND_V_MSG(err != OK, err, "Invalid URL: " + p_url);
 
 	if (scheme.is_empty()) {

modules/websocket/wsl_peer.cpp

@@ -482,8 +482,9 @@ Error WSLPeer::connect_to_url(const String &p_url, Ref<TLSOptions> p_options) {
 	String host;
 	String path;
 	String scheme;
+	String fragment;
 	int port = 0;
-	Error err = p_url.parse_url(scheme, host, port, path);
+	Error err = p_url.parse_url(scheme, host, port, path, fragment);
 	ERR_FAIL_COND_V_MSG(err != OK, err, "Invalid URL: " + p_url);
 	if (scheme.is_empty()) {
 		scheme = "ws://";

scene/2d/gpu_particles_2d.cpp

@@ -688,6 +688,7 @@ void GPUParticles2D::_notification(int p_what) {
 				RS::get_singleton()->particles_set_speed_scale(particles, 0);
 			}
 			set_process_internal(true);
+			set_physics_process_internal(true);
 			previous_position = get_global_position();
 		} break;
 
@@ -711,15 +712,6 @@ void GPUParticles2D::_notification(int p_what) {
 		} break;
 
 		case NOTIFICATION_INTERNAL_PROCESS: {
-			const Vector3 velocity = Vector3((get_global_position() - previous_position).x, (get_global_position() - previous_position).y, 0.0) /
-					get_process_delta_time();
-
-			if (velocity != previous_velocity) {
-				RS::get_singleton()->particles_set_emitter_velocity(particles, velocity);
-				previous_velocity = velocity;
-			}
-			previous_position = get_global_position();
-
 			if (one_shot) {
 				time += get_process_delta_time();
 				if (time > emission_time) {
@@ -739,6 +731,19 @@ void GPUParticles2D::_notification(int p_what) {
 				}
 			}
 		} break;
+
+		case NOTIFICATION_INTERNAL_PHYSICS_PROCESS: {
+			// Update velocity in physics process, so that velocity calculations remain correct
+			// if the physics tick rate is lower than the rendered framerate (especially without physics interpolation).
+			const Vector3 velocity = Vector3((get_global_position() - previous_position).x, (get_global_position() - previous_position).y, 0.0) /
+					get_physics_process_delta_time();
+
+			if (velocity != previous_velocity) {
+				RS::get_singleton()->particles_set_emitter_velocity(particles, velocity);
+				previous_velocity = velocity;
+			}
+			previous_position = get_global_position();
+		} break;
 	}
 }
 

scene/2d/light_2d.cpp

@@ -417,7 +417,7 @@ Vector2 PointLight2D::get_texture_offset() const {
 }
 
 PackedStringArray PointLight2D::get_configuration_warnings() const {
-	PackedStringArray warnings = Node2D::get_configuration_warnings();
+	PackedStringArray warnings = Light2D::get_configuration_warnings();
 
 	if (!texture.is_valid()) {
 		warnings.push_back(RTR("A texture with the shape of the light must be supplied to the \"Texture\" property."));

scene/2d/physics/rigid_body_2d.cpp

@@ -641,7 +641,7 @@ void RigidBody2D::_notification(int p_what) {
 PackedStringArray RigidBody2D::get_configuration_warnings() const {
 	Transform2D t = get_transform();
 
-	PackedStringArray warnings = CollisionObject2D::get_configuration_warnings();
+	PackedStringArray warnings = PhysicsBody2D::get_configuration_warnings();
 
 	if (ABS(t.columns[0].length() - 1.0) > 0.05 || ABS(t.columns[1].length() - 1.0) > 0.05) {
 		warnings.push_back(RTR("Size changes to RigidBody2D will be overridden by the physics engine when running.\nChange the size in children collision shapes instead."));