// Comprehensive capture/audio/hook stress test against coop_mock_game.
//
// Launches the mock game (an animated, frame-numbered A/V source), injects coop_hook.dll,
// and drives the real shared-memory paths the host uses -- no Steam, no host UI. For each
// graphics backend (DX11, DX12) it opens the hook's shared video texture, decodes the
// frame number out of the captured pixels, and asserts the mirror sees a *monotonic,
// advancing* sequence (this is what the DX12 rotating-backbuffer bug broke -- it showed
// stale/repeated frames). A final A/V test injects with audio + video, checks both stream,
// cycles the audio subsystem off/on a few times (hook/unhook stress), and confirms the
// game never freezes/crashes. Skips cleanly without a D3D11 device.
#include <algorithm>
#include <cmath>
#include <cstdint>
#include <cstdio>
#include <string>
#include <vector>

#include <windows.h>

#include <tlhelp32.h>

#include <d3d11.h>

#include "capture/shared_texture.hpp"
#include "coop/audio_ring.hpp"
#include "coop/log_ring.hpp"
#include "coop/protocol.hpp"
#include "coop/shared_memory.hpp"
#include "coop/tool_paths.hpp"
#include "render_backend.hpp" // coop::mock::rgb_to_frame / kFrameBlock

using namespace coop;

namespace
{
int g_failures = 0;
void check(bool ok, const char* what)
{
	std::printf("%s %s\n", ok ? "  ok:" : "FAIL:", what);
	if (!ok)
	{
		++g_failures;
	}
}

std::wstring tool_path(const wchar_t* name)
{
	return exe_directory() + name; // coop_mock_game.exe is staged next to this test
}

// Kill any leftover mock games from a previous (crashed/interrupted) run, so a stray one
// holding coop_hook.dll can't perturb this run.
void kill_stray_mock_games()
{
	HANDLE snap = CreateToolhelp32Snapshot(TH32CS_SNAPPROCESS, 0);
	if (snap == INVALID_HANDLE_VALUE)
	{
		return;
	}
	PROCESSENTRY32W pe{};
	pe.dwSize = sizeof(pe);
	for (BOOL ok = Process32FirstW(snap, &pe); ok; ok = Process32NextW(snap, &pe))
	{
		if (_wcsicmp(pe.szExeFile, L"coop_mock_game.exe") == 0)
		{
			HANDLE h = OpenProcess(PROCESS_TERMINATE, FALSE, pe.th32ProcessID);
			if (h != nullptr)
			{
				TerminateProcess(h, 0);
				CloseHandle(h);
			}
		}
	}
	CloseHandle(snap);
}

// Inject coop_hook.dll (x64 -> the mock game is x64) into `pid` via LoadLibrary remote thread.
bool inject(unsigned long pid)
{
	const std::wstring dll = deployed_artifact_path(L"coop_hook.dll"); // root is one dir up from tests/
	if (GetFileAttributesW(dll.c_str()) == INVALID_FILE_ATTRIBUTES)
	{
		return false;
	}
	const DWORD access = PROCESS_CREATE_THREAD | PROCESS_QUERY_INFORMATION | PROCESS_VM_OPERATION |
	                     PROCESS_VM_WRITE | PROCESS_VM_READ;
	HANDLE process = OpenProcess(access, FALSE, pid);
	if (process == nullptr)
	{
		return false;
	}
	const SIZE_T bytes = (dll.size() + 1) * sizeof(wchar_t);
	void* remote = VirtualAllocEx(process, nullptr, bytes, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
	bool ok = false;
	if (remote != nullptr && WriteProcessMemory(process, remote, dll.c_str(), bytes, nullptr))
	{
		auto load = reinterpret_cast<LPTHREAD_START_ROUTINE>(
		    GetProcAddress(GetModuleHandleW(L"kernel32.dll"), "LoadLibraryW"));
		HANDLE th = CreateRemoteThread(process, nullptr, 0, load, remote, 0, nullptr);
		if (th != nullptr)
		{
			WaitForSingleObject(th, INFINITE);
			DWORD code = 0;
			GetExitCodeThread(th, &code);
			CloseHandle(th);
			ok = code != 0;
		}
	}
	if (remote != nullptr)
	{
		VirtualFreeEx(process, remote, 0, MEM_RELEASE);
	}
	CloseHandle(process);
	return ok;
}

// Inject with a few retries: a freshly-launched process can briefly refuse a remote thread.
bool inject_retry(unsigned long pid)
{
	for (int i = 0; i < 4; ++i)
	{
		if (inject(pid))
		{
			return true;
		}
		Sleep(300);
	}
	return false;
}

struct MockGame
{
	PROCESS_INFORMATION pi{};
	bool ok = false;

	// Launch coop_mock_game.exe with the given args (e.g. L"dx12 30" or L"dx11 30 48000 2 32 float").
	static MockGame launch(const std::wstring& args)
	{
		MockGame g;
		const std::wstring exe = tool_path(L"coop_mock_game.exe");
		std::wstring cmd = L"\"" + exe + L"\" " + args;
		STARTUPINFOW si{};
		si.cb = sizeof(si);
		g.ok = CreateProcessW(exe.c_str(), cmd.data(), nullptr, nullptr, FALSE, 0, nullptr, nullptr, &si,
		                      &g.pi) != 0;
		return g;
	}
	unsigned long pid() const
	{
		return pi.dwProcessId;
	}
	bool alive() const
	{
		return pi.hProcess != nullptr && WaitForSingleObject(pi.hProcess, 0) == WAIT_TIMEOUT;
	}
	unsigned long exit_code() const
	{
		DWORD code = 0;
		if (pi.hProcess != nullptr)
		{
			GetExitCodeProcess(pi.hProcess, &code);
		}
		return code;
	}
	void kill()
	{
		if (pi.hProcess != nullptr)
		{
			TerminateProcess(pi.hProcess, 0);
			WaitForSingleObject(pi.hProcess, 2000);
			CloseHandle(pi.hThread);
			CloseHandle(pi.hProcess);
			pi = PROCESS_INFORMATION{};
		}
	}
};

// Create the input SharedBlock the hook needs, with the given subsystems disabled (bit per
// HookSubsystem). Keeps the mapping alive in `shm`.
SharedBlock* make_ipc(SharedMemory& shm, unsigned long pid, std::uint32_t disabled_mask)
{
	if (!shm.create(shared_memory_name(pid), sizeof(SharedBlock)))
	{
		return nullptr;
	}
	auto* block = shm.as<SharedBlock>();
	block->version = kProtocolVersion;
	block->pad_count = 0;
	block->sequence.store(0, std::memory_order_relaxed);
	for (std::uint32_t s = 0; s < HookSubsys_Count; ++s)
	{
		block->control.subsystem_disabled[s].store((disabled_mask >> s) & 1u, std::memory_order_release);
	}
	block->magic = kProtocolMagic;
	return block;
}

ID3D11Device* make_device()
{
	ID3D11Device* dev = nullptr;
	const D3D_FEATURE_LEVEL fl[] = {D3D_FEATURE_LEVEL_11_1, D3D_FEATURE_LEVEL_11_0};
	if (FAILED(D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, 0, fl,
	                             static_cast<UINT>(std::size(fl)), D3D11_SDK_VERSION, &dev, nullptr, nullptr)))
	{
		return nullptr;
	}
	return dev;
}

VideoShareView read_video_share(const SharedBlock* block)
{
	VideoShareView v;
	v.generation = block->video.generation.load(std::memory_order_acquire);
	v.width = block->video.width;
	v.height = block->video.height;
	v.format = block->video.format;
	v.present_calls = block->video.present_calls;
	return v;
}

// Capture from the mock game on `backend` and assert the decoded frame numbers form a
// monotonic, advancing sequence.
void test_video_capture(const char* backend, ID3D11Device* device)
{
	std::printf("== video capture: %s ==\n", backend);
	const std::wstring args = (std::string(backend) == "dx12" ? L"dx12 " : L"dx11 ") + std::wstring(L"30");
	MockGame game = MockGame::launch(args);
	if (!game.ok)
	{
		check(false, "launch coop_mock_game");
		return;
	}
	Sleep(800); // let the window + swap chain come up

	// Only the video subsystem (disable input/focus/audio/mkb to keep the test focused).
	SharedMemory shm;
	const std::uint32_t disabled = (1u << HookSubsys_Input) | (1u << HookSubsys_Focus) |
	                               (1u << HookSubsys_Audio) | (1u << HookSubsys_Mkb);
	SharedBlock* block = make_ipc(shm, game.pid(), disabled);
	if (block == nullptr || !inject_retry(game.pid()))
	{
		check(false, "inject into mock game");
		game.kill();
		return;
	}

	SharedTextureSource src;
	src.init(device);

	std::vector<std::uint32_t> seq;
	std::uint64_t backward = 0;
	std::uint32_t last = 0;
	bool have_last = false;
	for (int i = 0; i < 80 && game.alive(); ++i) // ~4 s at 50 ms
	{
		Sleep(50);
		const VideoShareView share = read_video_share(block);
		if (!src.update(share, game.pid()))
		{
			continue;
		}
		std::uint8_t px[4] = {};
		if (!src.read_pixel(coop::mock::kFrameBlock / 2, coop::mock::kFrameBlock / 2, px))
		{
			continue;
		}
		const std::uint32_t f = coop::mock::rgb_to_frame(px[0], px[1], px[2]);
		if (have_last && f < last)
		{
			++backward; // a stale / wrong (rotated) buffer -> frame number went backwards
		}
		last = f;
		have_last = true;
		seq.push_back(f);
	}

	const std::uint32_t present = static_cast<std::uint32_t>(block->video.present_calls);
	std::printf("   present_calls=%u copied=%llu samples=%zu backward=%llu\n", present,
	            static_cast<unsigned long long>(src.frames_copied()), seq.size(),
	            static_cast<unsigned long long>(backward));

	check(present > 0, "hook captured Present calls");
	check(src.frames_copied() >= 5, "host copied multiple shared frames");
	check(seq.size() >= 5, "decoded multiple frame numbers from the captured pixels");
	check(backward == 0, "captured frame numbers never go backwards (no stale/rotated buffer)");
	if (!seq.empty())
	{
		const std::uint32_t span = seq.back() - seq.front();
		std::printf("   frame# %u..%u (span %u)\n", seq.front(), seq.back(), span);
		check(span >= 20, "captured frame numbers advance (mirror gets fresh frames)");
		const bool all_same = std::all_of(seq.begin(), seq.end(), [&](std::uint32_t v) { return v == seq[0]; });
		check(!all_same, "captured frames are not stuck on one number");
	}

	game.kill();
}

// Launch the mock game rendering audio at `rate`/`channels`/`bits`/`fmt`, inject the audio
// hook (late, so it's the guessed path), and verify the hook MEASURES the right sample rate
// for this variant and captures non-silent audio. (Channels/bit-depth aren't recoverable for
// a guessed stream -- that's the documented limitation -- so the rate is the variant check.)
void test_audio_variant(unsigned rate, unsigned channels, unsigned bits, const wchar_t* fmt)
{
	wchar_t args[64];
	swprintf(args, static_cast<int>(std::size(args)), L"dx11 30 %u %u %u %ls", rate, channels, bits, fmt);
	std::printf("== audio variant: %u Hz %u ch %u-bit %ls ==\n", rate, channels, bits, fmt);
	MockGame game = MockGame::launch(args);
	if (!game.ok)
	{
		check(false, "launch coop_mock_game (audio variant)");
		return;
	}
	Sleep(800);

	// Audio subsystem only.
	SharedMemory shm;
	const std::uint32_t disabled = (1u << HookSubsys_Input) | (1u << HookSubsys_Focus) |
	                               (1u << HookSubsys_Video) | (1u << HookSubsys_Mkb);
	SharedBlock* block = make_ipc(shm, game.pid(), disabled);
	SharedMemory ring_shm;
	AudioRingHeader* ring = nullptr;
	if (ring_shm.create(audio_ring_name(game.pid()), audio_ring_total_size(kAudioRingCapacity)))
	{
		ring = ring_shm.as<AudioRingHeader>();
		audio_ring_init(*ring, kAudioRingCapacity);
		ring->capture_enabled.store(1, std::memory_order_release);
	}
	if (block == nullptr || ring == nullptr || !inject_retry(game.pid()))
	{
		check(false, "inject into mock game (audio variant)");
		game.kill();
		return;
	}

	// Wait for the rate to be measured + published (Measured / LowConfidence / Exact), then
	// drain to prove non-silent capture.
	std::vector<std::uint8_t> drain(kAudioRingCapacity);
	std::uint32_t measured = 0;
	std::uint32_t state = 0;
	double peak = 0.0;
	for (int i = 0; i < 120 && game.alive(); ++i) // up to ~6 s
	{
		Sleep(50);
		std::uint32_t got = 0;
		while ((got = audio_ring_pop(*ring, drain.data(), static_cast<std::uint32_t>(drain.size()))) > 0)
		{
			if (ring->bits == 32 && ring->format_tag == 3)
			{
				const auto* f = reinterpret_cast<const float*>(drain.data());
				for (std::uint32_t k = 0; k < got / 4; ++k)
				{
					peak = std::max(peak, static_cast<double>(std::fabs(f[k])));
				}
			}
			else if (ring->bits == 16)
			{
				const auto* s = reinterpret_cast<const std::int16_t*>(drain.data());
				for (std::uint32_t k = 0; k < got / 2; ++k)
				{
					peak = std::max(peak, std::abs(s[k]) / 32768.0);
				}
			}
			if (got < drain.size())
			{
				break;
			}
		}
		state = block->status.audio_streams[0].format_state;
		if (state == AudioFormat_Measured || state == AudioFormat_LowConfidence || state == AudioFormat_Exact)
		{
			measured = block->status.audio_streams[0].sample_rate;
			if (measured != 0 && peak > 0.01)
			{
				break;
			}
		}
	}
	std::printf("   measured rate=%u Hz (state=%u) peak=%.4f\n", measured, state, peak);
	check(measured == rate, "hook measured this variant's sample rate");
	check(peak > 0.01, "captured non-silent audio for this variant");

	game.kill();
}

// Inject with audio + video, verify both stream, then cycle the audio subsystem off/on a
// few times (hook/unhook stress) and confirm the game stays alive (heartbeat advances).
void test_av_and_hook_cycles(ID3D11Device* device)
{
	std::printf("== A/V + hook/unhook stress (dx11 + audio) ==\n");
	MockGame game = MockGame::launch(L"dx11 30 48000 2 32 float");
	if (!game.ok)
	{
		check(false, "launch coop_mock_game (A/V)");
		return;
	}
	Sleep(800);

	SharedMemory shm;
	SharedBlock* block = make_ipc(shm, game.pid(), /*disabled=*/0); // all subsystems on
	SharedMemory ring_shm;
	AudioRingHeader* ring = nullptr;
	if (ring_shm.create(audio_ring_name(game.pid()), audio_ring_total_size(kAudioRingCapacity)))
	{
		ring = ring_shm.as<AudioRingHeader>();
		audio_ring_init(*ring, kAudioRingCapacity);
		ring->capture_enabled.store(1, std::memory_order_release);
	}
	if (block == nullptr || ring == nullptr || !inject_retry(game.pid()))
	{
		check(false, "inject into mock game (A/V)");
		game.kill();
		return;
	}

	SharedTextureSource src;
	src.init(device);

	// Let it settle, then verify video advances and audio is non-silent.
	double peak = 0.0;
	std::uint32_t first_frame = 0, last_frame = 0;
	std::vector<std::uint8_t> drain(kAudioRingCapacity);
	for (int i = 0; i < 60 && game.alive(); ++i) // ~3 s
	{
		Sleep(50);
		const VideoShareView share = read_video_share(block);
		if (src.update(share, game.pid()))
		{
			std::uint8_t px[4] = {};
			if (src.read_pixel(coop::mock::kFrameBlock / 2, coop::mock::kFrameBlock / 2, px))
			{
				const std::uint32_t f = coop::mock::rgb_to_frame(px[0], px[1], px[2]);
				if (first_frame == 0)
				{
					first_frame = f;
				}
				last_frame = f;
			}
		}
		std::uint32_t got = 0;
		while ((got = audio_ring_pop(*ring, drain.data(), static_cast<std::uint32_t>(drain.size()))) > 0)
		{
			if (ring->bits == 32 && ring->format_tag == 3)
			{
				const auto* f = reinterpret_cast<const float*>(drain.data());
				for (std::uint32_t k = 0; k < got / 4; ++k)
				{
					peak = std::max(peak, static_cast<double>(std::fabs(f[k])));
				}
			}
			if (got < drain.size())
			{
				break;
			}
		}
	}
	check(last_frame > first_frame, "A/V: video frames advance");
	check(peak > 0.01, "A/V: audio captured non-silent");

	// Hook/unhook stress: toggle the audio subsystem off->on a few times. Each phase is >
	// the worker's ~250 ms reconcile tick, so the install/remove fully completes each time
	// (this is the realistic cadence -- an operator toggling a checkbox, not thrashing it).
	const std::uint32_t hb_start = block->status.heartbeat.load(std::memory_order_relaxed);
	for (int c = 0; c < 3 && game.alive(); ++c)
	{
		block->control.subsystem_disabled[HookSubsys_Audio].store(1, std::memory_order_release);
		Sleep(400);
		block->control.subsystem_disabled[HookSubsys_Audio].store(0, std::memory_order_release);
		Sleep(400);
	}
	const std::uint32_t hb_end = block->status.heartbeat.load(std::memory_order_relaxed);
	if (!game.alive())
	{
		std::printf("   game exit code = 0x%08lX\n", game.exit_code());
	}
	check(game.alive(), "game survived hook/unhook cycles (no crash)");
	check(hb_end > hb_start, "hook heartbeat kept advancing through the cycles (no freeze)");

	// After the final re-enable, capture must resume: the audio must come back (proving the
	// rehook works end-to-end, not just that nothing crashed).
	std::uint64_t produced_before = ring->frames_produced.load(std::memory_order_relaxed);
	double peak2 = 0.0;
	for (int i = 0; i < 30 && game.alive(); ++i) // ~1.5 s
	{
		Sleep(50);
		std::uint32_t got = 0;
		while ((got = audio_ring_pop(*ring, drain.data(), static_cast<std::uint32_t>(drain.size()))) > 0)
		{
			if (ring->bits == 32 && ring->format_tag == 3)
			{
				const auto* f = reinterpret_cast<const float*>(drain.data());
				for (std::uint32_t k = 0; k < got / 4; ++k)
				{
					peak2 = std::max(peak2, static_cast<double>(std::fabs(f[k])));
				}
			}
			if (got < drain.size())
			{
				break;
			}
		}
	}
	const std::uint64_t produced_after = ring->frames_produced.load(std::memory_order_relaxed);
	check(produced_after > produced_before && peak2 > 0.01, "audio capture resumed after re-enable");

	game.kill();
}
} // namespace

int main()
{
	kill_stray_mock_games(); // clean slate: no leftover game holding coop_hook.dll

	ID3D11Device* device = make_device();
	if (device == nullptr)
	{
		std::printf("No D3D11 device -- skipping mock_game_test.\n");
		return 0;
	}

	test_video_capture("dx11", device);
	test_video_capture("dx12", device);

	// Audio variants: the hook must measure each variant's rate through the full inject path.
	test_audio_variant(44100, 2, 16, L"pcm");
	test_audio_variant(48000, 2, 32, L"float");
	test_audio_variant(96000, 2, 32, L"float");

	test_av_and_hook_cycles(device);

	device->Release();
	kill_stray_mock_games(); // belt-and-suspenders: ensure nothing is left running

	if (g_failures == 0)
	{
		std::printf("PASS mock_game_test\n");
		return 0;
	}
	std::printf("FAILED mock_game_test (%d)\n", g_failures);
	return 1;
}