api: get_streams (#886)

## Link to GitHub Issue or related Pull Request, if one exists
#0

## Description of change
MSE method of streaming video needs the dimensions up front. This new
api delivers that. Also provides an easier way to determine which
screens are available and a way to discover port number.

## Testing
See pending changes in substream project

src/spice2x/README.md

@@ -127,6 +127,22 @@ doesn't matter since the TCP protocol doesn't allow for out of order data,
 however this may change when/if support for UDP is being introduced. The only
 restriction is that the ID has to be a valid 64-bit unsigned integer.
 
+#### Capture
+- get_screens()
+  - returns the screen numbers the game has registered for capture
+- get_jpg(screen: uint, quality: uint, divide: uint)
+  - returns the timestamp, width, height and base64 encoded JPEG of one screen
+  - all parameters are optional and default to screen 0, quality 70, divide 1
+  - divide shrinks the image by that factor before encoding
+- get_streams()
+  - returns a dict describing the HTTP video stream, or no data at all when
+    `-apistream` is not enabled and there is nothing to describe
+  - `port` is the stream server port
+  - `formats` lists the wire formats this build serves, each with a `name`
+    (`h264` or `mjpeg`) and the `path` to request them on
+  - `screens` lists every capturable screen with its `width`, `height`, and
+    `busy`
+
 #### Card
 - insert(index: uint, card_id: hex)
   - inserts a card which gets read by the emulated card readers for the game
@@ -298,6 +314,10 @@ stream over plain HTTP. Enable it with `-apistream`. It listens on the API port
 plus two, in the same way the WebSocket server uses the API port plus one, so
 `-api 1337` puts the stream on 1339. This means `-api` has to be enabled too.
 
+Rather than working the port out, clients should ask the JSON API for it with
+`capture.get_streams()`, which also reports which of the formats below this
+build serves, the size of each screen and whether one is already taken.
+
 Two formats are served:
 
     http://host:1339/stream.mjpg    JPEG frames, multipart/x-mixed-replace
@@ -323,9 +343,10 @@ The stream is view only. Touch and other input still go through the JSON API,
 so a companion app needs both. There is no authentication on the stream port -
 anyone who can reach it can watch the screen.
 
-WinXP builds have no video stream. Neither encoder is compiled in, so every
-endpoint returns 404, and the JSON API's JPEG screen capture is unavailable for
-the same reason.
+WinXP builds have no video stream. Neither encoder is compiled in, so nothing
+listens on the stream port even with `-apistream`, `capture.get_streams()`
+returns no data, and the JSON API's JPEG screen capture is unavailable for the
+same reason.
 
 ## Native wrapper libraries
 Spicetools provides wrapper libraries in: Arduino, C++, Dart, and Python.

src/spice2x/api/modules/capture.cpp

@@ -1,8 +1,11 @@
 #include "capture.h"
+#include <atomic>
 #include <functional>
 #include <mutex>
 #include <unordered_map>
 #include "api/capture_pump.h"
+#include "api/stream_format.h"
+#include "api/stream_server.h"
 #include "external/rapidjson/document.h"
 #include "hooks/graphics/graphics.h"
 #include "hooks/graphics/jpeg_encoder.h"
@@ -71,6 +74,7 @@ namespace api::modules {
     Capture::Capture() : Module("capture") {
         functions["get_screens"] = std::bind(&Capture::get_screens, this, _1, _2);
         functions["get_jpg"] = std::bind(&Capture::get_jpg, this, _1, _2);
+        functions["get_streams"] = std::bind(&Capture::get_streams, this, _1, _2);
     }
 
     /**
@@ -141,4 +145,98 @@ namespace api::modules {
         CAPTURE_BUFFER.clear();
         try_cached_response(screen, res);
     }
+
+    /**
+     * get_streams()
+     */
+    void Capture::get_streams(Request &req, Response &res) {
+
+        auto &alloc = res.doc()->GetAllocator();
+
+        // nothing is listening without -apistream, so there is no stream to describe
+        const unsigned short port = stream_server_port();
+        if (port == 0) {
+            return;
+        }
+
+        Value formats(kArrayType);
+        for (const auto &[name, path] : stream_formats()) {
+            Value entry(kObjectType);
+            entry.AddMember("name", Value(name.c_str(), alloc), alloc);
+            entry.AddMember("path", Value(path.c_str(), alloc), alloc);
+            formats.PushBack(entry, alloc);
+        }
+
+        std::vector<int> screen_numbers;
+        graphics_screens_get(screen_numbers);
+
+        // measuring a screen nobody has captured yet waits for the game to present, which can
+        // take as long as the whole request is allowed, so only one screen is measured per
+        // call and the rest are reported null until a later one settles them. which screen
+        // gets the attempt rotates, otherwise one that never presents would take every
+        // request and the screens behind it would stay unmeasured forever
+        int probe_screen = -1;
+        {
+            std::vector<int> unmeasured;
+            for (const auto screen : screen_numbers) {
+                if (screen < static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)
+                        && !graphics_capture_last_size(screen, nullptr, nullptr)
+                        && !capture_pump::screen_claimed(screen)) {
+                    unmeasured.push_back(screen);
+                }
+            }
+
+            if (!unmeasured.empty()) {
+                static std::atomic<unsigned> probe_cursor { 0 };
+                probe_screen = unmeasured[probe_cursor.fetch_add(1) % unmeasured.size()];
+            }
+        }
+
+        Value screens(kArrayType);
+        for (const auto screen : screen_numbers) {
+            if (screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+                continue;
+            }
+
+            int width = 0;
+            int height = 0;
+            bool known = graphics_capture_last_size(screen, &width, &height);
+
+            // a probe holds the screen for as long as it waits, so a second caller arriving
+            // during one would queue behind it and then take a wait of its own; let it report
+            // the screen as unmeasured instead and pick the size up once the first is done
+            static std::atomic<bool> probe_running { false };
+            if (!known && screen == probe_screen && !probe_running.exchange(true)) {
+                std::shared_ptr<uint8_t[]> pixels;
+                known = capture_pump::capture_direct(
+                        screen, pixels, 1, nullptr, &width, &height);
+                probe_running = false;
+            }
+
+            // a screen of unknown size cannot be described, and a client told about it could
+            // not size its decoder anyway; leaving it out until it has been measured beats
+            // handing over an entry that has to be treated as absent
+            if (!known) {
+                continue;
+            }
+
+            Value entry(kObjectType);
+            entry.AddMember("screen", screen, alloc);
+            entry.AddMember("width", width, alloc);
+            entry.AddMember("height", height, alloc);
+
+            // a screen carries one viewer at a time, so this is what decides whether a client
+            // can connect at all; still racy by the time it does, only more honest than not
+            entry.AddMember("busy", capture_pump::screen_claimed(screen), alloc);
+
+            screens.PushBack(entry, alloc);
+        }
+
+        Value info(kObjectType);
+        info.AddMember("port", port, alloc);
+        info.AddMember("formats", formats, alloc);
+        info.AddMember("screens", screens, alloc);
+
+        res.add_data(info);
+    }
 }

src/spice2x/api/modules/capture.h

@@ -19,5 +19,6 @@ namespace api::modules {
         // function definitions
         void get_screens(Request &req, Response &res);
         void get_jpg(Request &req, Response &res);
+        void get_streams(Request &req, Response &res);
     };
 }

src/spice2x/api/resources/dart/spiceapi-websocket/spiceapi.dart

@@ -11,6 +11,7 @@ part "src/exceptions.dart";
 part "src/rc4.dart";
 part "src/wrappers/analogs.dart";
 part "src/wrappers/buttons.dart";
+part "src/wrappers/capture.dart";
 part "src/wrappers/card.dart";
 part "src/wrappers/coin.dart";
 part "src/wrappers/control.dart";

src/spice2x/api/resources/dart/spiceapi-websocket/src/connection.dart

@@ -4,8 +4,8 @@ part of spiceapi;
 class Connection {
 
   // settings
-  static const _TIMEOUT = Duration(seconds: 2);
-  static const _BUFFER_SIZE = 1024 * 8;
+  static const _TIMEOUT = Duration(seconds: 3);
+  static const _BUFFER_SIZE = 1024 * 1024 * 8;
 
   // state
   final String host, pass;

src/spice2x/api/resources/dart/spiceapi-websocket/src/wrappers/capture.dart

@@ -0,0 +1,47 @@
+part of spiceapi;
+
+class CaptureData {
+  int timestamp;
+  int width, height;
+  Uint8List data;
+}
+
+var _base64DecoderInstance = Base64Decoder();
+
+Future<List> captureGetScreens(Connection con) {
+  var req = Request("capture", "get_screens");
+  return con.request(req).then((res) {
+    return res.getData();
+  });
+}
+
+Future<CaptureData> captureGetJPG(Connection con, {
+  int screen = 0,
+  int quality = 70,
+  int divide = 1,
+}) {
+  var req = Request("capture", "get_jpg");
+  req.addParam(screen);
+  req.addParam(quality);
+  req.addParam(divide);
+  return con.request(req).then((res) {
+    var captureData = CaptureData();
+    var data = res.getData();
+    if (data.length > 0) captureData.timestamp = data[0];
+    if (data.length > 1) captureData.width = data[1];
+    if (data.length > 2) captureData.height = data[2];
+    if (data.length > 3) {
+      captureData.data = _base64DecoderInstance.convert(data[3]);
+    }
+    return captureData;
+  });
+}
+
+/// Describes the HTTP video stream, or null when this spice2x serves none.
+Future<Map> captureGetStreams(Connection con) {
+  var req = Request("capture", "get_streams");
+  return con.request(req).then((res) {
+    var data = res.getData();
+    return data.length > 0 ? data[0] : null;
+  });
+}

src/spice2x/api/resources/dart/spiceapi/src/wrappers/capture.dart

@@ -17,7 +17,7 @@ Future<List> captureGetScreens(Connection con) {
 
 Future<CaptureData> captureGetJPG(Connection con, {
   int screen = 0,
-  int quality = 60,
+  int quality = 70,
   int divide = 1,
 }) {
   var req = Request("capture", "get_jpg");
@@ -36,3 +36,12 @@ Future<CaptureData> captureGetJPG(Connection con, {
     return captureData;
   });
 }
+
+/// Describes the HTTP video stream, or null when this spice2x serves none.
+Future<Map> captureGetStreams(Connection con) {
+  var req = Request("capture", "get_streams");
+  return con.request(req).then((res) {
+    var data = res.getData();
+    return data.length > 0 ? data[0] : null;
+  });
+}

src/spice2x/api/resources/python/spiceapi/__init__.py

@@ -2,6 +2,7 @@ from .connection import Connection
 from .request import Request
 from .analogs import *
 from .buttons import *
+from .capture import *
 from .card import *
 from .coin import *
 from .control import *

src/spice2x/api/resources/python/spiceapi/capture.py

@@ -0,0 +1,33 @@
+import base64
+
+from .connection import Connection
+from .request import Request
+
+
+def capture_get_screens(con: Connection):
+    res = con.request(Request("capture", "get_screens"))
+    return res.get_data()
+
+
+def capture_get_jpg(con: Connection, screen: int = 0, quality: int = 70, divide: int = 1):
+    req = Request("capture", "get_jpg")
+    req.add_param(screen)
+    req.add_param(quality)
+    req.add_param(divide)
+    data = con.request(req).get_data()
+
+    if len(data) < 4:
+        return None
+
+    return {
+        "timestamp": data[0],
+        "width": data[1],
+        "height": data[2],
+        "data": base64.b64decode(data[3]),
+    }
+
+
+def capture_get_streams(con: Connection):
+    """Describes the HTTP video stream, or None when this spice2x serves none."""
+    data = con.request(Request("capture", "get_streams")).get_data()
+    return data[0] if data else None

src/spice2x/api/stream_format.cpp

@@ -67,4 +67,18 @@ namespace api {
 
         return nullptr;
     }
+
+    std::vector<std::pair<std::string, std::string>> stream_formats() {
+        std::vector<std::pair<std::string, std::string>> formats;
+
+#ifdef SPICE_JPEG
+        formats.emplace_back("mjpeg", "/stream.mjpg");
+#endif
+
+#ifdef SPICE_H264
+        formats.emplace_back("h264", "/stream.h264");
+#endif
+
+        return formats;
+    }
 }

src/spice2x/api/stream_format.h

@@ -4,6 +4,8 @@
 #include <functional>
 #include <memory>
 #include <string>
+#include <utility>
+#include <vector>
 
 #include "capture_pump.h"
 
@@ -35,4 +37,7 @@ namespace api {
     // null when the path does not name a format this build supports
     std::unique_ptr<StreamWriter> make_stream_writer(
             const std::string &path, int quality, int fps);
+
+    // name and path of every format compiled into this build, for clients to pick from
+    std::vector<std::pair<std::string, std::string>> stream_formats();
 }

src/spice2x/api/stream_server.cpp

@@ -195,11 +195,25 @@ namespace api {
                     "\r\n";
             send_all(socket, response);
         }
+
+        std::atomic<unsigned short> LISTENING_PORT { 0 };
+    }
+
+    unsigned short stream_server_port() {
+        return LISTENING_PORT.load();
     }
 
     StreamServer::StreamServer(unsigned short port)
         : port(port)
     {
+        // WinXP builds compile in neither encoder, so there would be nothing to serve and
+        // every request would 404; taking the port instead only invites confused clients
+        if (stream_formats().empty()) {
+            log_warning("api::stream",
+                    "this build has no video encoders, the video stream is unavailable");
+            return;
+        }
+
         if (!this->open_listener()) {
             // the stream was asked for explicitly, so say plainly that it is not there
             log_warning("api::stream",
@@ -212,6 +226,8 @@ namespace api {
             this->accept_worker();
         });
 
+        LISTENING_PORT = this->port;
+
         // deliberately not logging a full URL; local IPs would leak into shared logs
         log_info("api::stream", "video stream is listening on port: {}", this->port);
         log_warning("api::stream",
@@ -269,6 +285,7 @@ namespace api {
     StreamServer::~StreamServer() {
 
         this->running = false;
+        LISTENING_PORT = 0;
 
         if (this->listener != INVALID_SOCKET) {
             closesocket(this->listener);

src/spice2x/api/stream_server.h

@@ -11,6 +11,9 @@
 
 namespace api {
 
+    // 0 while no stream server is listening, so the API can tell clients not to look for one
+    unsigned short stream_server_port();
+
     class StreamServer {
     public:
 

src/spice2x/hooks/graphics/graphics.cpp

@@ -1481,6 +1481,27 @@ void graphics_capture_skip(int screen) {
     GRAPHICS_CAPTURE_CV[screen].notify_one();
 }
 
+bool graphics_capture_last_size(int screen, int *width, int *height) {
+    if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+        return false;
+    }
+
+    // consuming a frame clears the pixels but leaves the size, so this survives the read
+    std::lock_guard<std::mutex> lock(GRAPHICS_CAPTURE_BUFFER_M[screen]);
+    const auto &capture = GRAPHICS_CAPTURE_BUFFER[screen];
+    if (!capture.width || !capture.height) {
+        return false;
+    }
+
+    if (width != nullptr) {
+        *width = capture.width;
+    }
+    if (height != nullptr) {
+        *height = capture.height;
+    }
+    return true;
+}
+
 bool graphics_capture_receive_raw(int screen, std::shared_ptr<uint8_t[]> &out,
         int divide, uint64_t *timestamp,
         int *width, int *height) {

src/spice2x/hooks/graphics/graphics.h

@@ -146,6 +146,9 @@ void graphics_capture_trigger(int screen);
 bool graphics_capture_consume(int *screen);
 void graphics_capture_enqueue(int screen, uint8_t *data, size_t width, size_t height);
 void graphics_capture_skip(int screen);
+// size of the last frame captured off this screen, before any caller side downscale; false
+// until one has been captured, so it cannot report a size for a screen the game never drew
+bool graphics_capture_last_size(int screen, int *width, int *height);
 // on success `out` owns packed 24bpp RGB pixels, width * height * 3 bytes
 bool graphics_capture_receive_raw(int screen, std::shared_ptr<uint8_t[]> &out,
         int divide = 0,