CI-Build 2022-08-19 v2

.github/workflows/ci.yml

@@ -58,7 +58,7 @@ jobs:
             name: macOS Release,
             os: macos-latest,
             buildtype: release,
-            args: -Ddefault_library=static -Dbuild_osx_bundle=true -Dlocal_boost=true -Davisynth=enabled -Dvapoursynth=enabled
+            args: -Ddefault_library=static -Dbuild_osx_bundle=true -Dlocal_boost=true
           }
 
     steps:

README.md

@@ -19,6 +19,7 @@ Being a collection of different feature additions, this repository consists of a
 The `cibuilds` branch makes some CI builds of snapshots of `feature` at relevant points in time.
 
 ### Branch/Feature list
+This list is for navigating the repository. Go to the [release page](https://github.com/arch1t3cht/Aegisub/releases) for a more structured changelog.
 - [`folding`](https://github.com/arch1t3cht/Aegisub/tree/folding): Add the ability to visually group and collapse lines in the subtitle grid
 - [`lua_api`](https://github.com/arch1t3cht/Aegisub/tree/lua_api): Add new functions to the Lua automation API, like controlling the selection or cursor in the text edit box
 - [`vector_clip_actions`](https://github.com/arch1t3cht/Aegisub/tree/vector_clip_actions): Make the different modes of the vector clip tool (lines, bezier curves, adding points, etc) bindable to hotkeys
@@ -28,13 +29,20 @@ The `cibuilds` branch makes some CI builds of snapshots of `feature` at relevant
 - [`vapoursynth`](https://github.com/arch1t3cht/Aegisub/tree/vapoursynth): Add Vapoursynth audio and video source
 - [`bugfixes`](https://github.com/arch1t3cht/Aegisub/tree/bugfixes): Various fixes necessary for compilation. Most branches are based on this.
 - [`fixes`](https://github.com/arch1t3cht/Aegisub/tree/fixes): Miscellaneous bugfixes
+- [`misc`](https://github.com/arch1t3cht/Aegisub/tree/misc): Other miscellaneous additions
 - [`misc_dc`](https://github.com/arch1t3cht/Aegisub/tree/misc_dc): Miscellaneous changes taken from AegisubDC
 - [`xa-ds2`](https://github.com/arch1t3cht/Aegisub/tree/xa-ds2): Add XAudio2 backend and allow stereo playback for some other backends, by wangqr and Shinon.
+- [`stereo`](https://github.com/arch1t3cht/Aegisub/tree/stereo): Add multi-channel support for the other audio backends where possible.
 - [`video_panning_feature`](https://github.com/arch1t3cht/Aegisub/tree/video_panning_feature): Merge [moex3's video zoom and panning](https://github.com/TypesettingTools/Aegisub/pull/150), with an OSX fix and more options to control zoom behavior
 - [`spectrum-frequency-mapping`](https://github.com/arch1t3cht/Aegisub/tree/spectrum-frequency-mapping): Merge EleonoreMizo's [spectrum display improvements](https://github.com/TypesettingTools/Aegisub/pull/94), and also make Shift+Scroll vertically zoom the audio display
 - [`wangqr_time_video`](https://github.com/arch1t3cht/Aegisub/tree/wangqr_time_video): Merge wangqr's feature adding a tool for timing subtitles to changes in the video
 
 ### Troubleshooting
+I'll gladly take any bug reports, but if you encounter an issue, please check first if it occurs only on my fork, or also on [earlier TSTools builds](https://github.com/TypesettingTools/Aegisub/actions).
+If it wasn't introduced by my fork, I can still take a look, but I can't promise anything.
+
+You can find me for support on various servers, including the cave and the TSTools server linked below.
+
 #### Building fails with a "CMake sandbox violation"
 This is an upstream bug in meson. For now, you need to downgrade meson using `pip install meson==0.62.2`.
 
@@ -55,6 +63,37 @@ If it's not because of this particular bug, you can also try an alternative vide
 If you're compiling yourself, try adding `--force-fallback-for=zlib` to the meson options.
 
 
+### Compilation
+For compilation on Windows, see the TSTools documentation below. Also check the [GitHub workflow](https://github.com/arch1t3cht/Aegisub/blob/cibuilds/.github/workflows/ci.yml) for the project arguments.
+
+On Linux, you can use the [TSTools PKGBUILD](https://aur.archlinux.org/packages/aegisub-ttools-meson-git) as a base, in particular for installing the necessary dependencies if you don't want to compile them yourself.
+To compile manually,
+- Install Meson (at the moment, you'll need to downgrade Meson below 0.63.0: `pip install meson==0.62.2`)
+- Clone the repository
+- In the repository, run `meson setup build` for the default configuration. See below for further options.
+- `cd` to the `build` directory and run `ninja`
+- You'll get an `aegisub` binary in the `build` folder. To install it to a system-wide location, run `ninja install`. To install to `/usr` instead of `/usr/local`, pass `--prefix=/usr` when configuring or reconfiguring meson.
+- When recompiling after pulling new commits, skip the `meson setup` setup and just immediately run `ninja` from the build directory - even when the build configuration changed.
+
+#### Compilation flags
+Some features are not enabled by default. To enable them, pass `-D<feature>=enabled` with the `meson setup` command:
+
+- `-Davisynth=enabled`: Avisynth support
+- `-Dbestsource=enabled`: BestSource
+- `-Dvapoursynth=enabled`: Vapoursynth support
+
+You can also disable options that are active by default in the same way. Check the file `meson_options.txt` for all options.
+
+To change the options of an existing build directory, run `meson setup --reconfigure <new arguments>` from inside the `build` directory.
+
+### Dependencies
+Apart from the dependencies for the TSTools version, there are some additional dependencies. These are cloned and compiled from scratch if not found, but you might want to install binaries instead:
+- `jansson`: For BestSource
+- `ffmpeg`: Becomes a direct dependency when compiling with BestSource
+- `avisynth` (or `avisynthplus`): Optional run-time dependency for the Avisynth source
+- `vapoursynth`: Optional run-time dependency for the VapourSynth source
+
+
 # Aegisub
 
 For binaries and general information [see the homepage](http://www.aegisub.org).

libaegisub/audio/provider.cpp

@@ -119,6 +119,47 @@ void AudioProvider::FillBufferInt16Mono(int16_t* buf, int64_t start, int64_t cou
 	free(buff);
 }
 
+// This entire file has turned into a mess. For now I'm just following the pattern of the wangqr code, but
+// this should really be restructured entirely again. The original type constructor-based system worked very well - it could
+// just give downmix/conversion control to the players instead.
+void AudioProvider::GetAudioWithVolume(void *buf, int64_t start, int64_t count, double volume) const {
+	GetAudio(buf, start, count);
+	if (volume == 1.0) return;
+
+	int64_t n = count * GetChannels();
+
+	if (float_samples) {
+		if (bytes_per_sample == sizeof(float)) {
+			float *buff = reinterpret_cast<float *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = static_cast<float>(buff[i] * volume);
+		} else if (bytes_per_sample == sizeof(double)) {
+			double *buff = reinterpret_cast<double *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = buff[i] * volume;
+		}
+	}
+	else {
+		if (bytes_per_sample == sizeof(uint8_t)) {
+			uint8_t *buff = reinterpret_cast<uint8_t *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = util::mid(0, static_cast<int>(((int) buff[i] - 128) * volume + 128), 0xFF);
+		} else if (bytes_per_sample == sizeof(int16_t)) {
+			int16_t *buff = reinterpret_cast<int16_t *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = util::mid(-0x8000, static_cast<int>(buff[i] * volume), 0x7FFF);
+		} else if (bytes_per_sample == sizeof(int32_t)) {
+			int32_t *buff = reinterpret_cast<int32_t *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = static_cast<int32_t>(buff[i] * volume);
+		} else if (bytes_per_sample == sizeof(int64_t)) {
+			int64_t *buff = reinterpret_cast<int64_t *>(buf);
+			for (int64_t i = 0; i < n; ++i)
+				buff[i] = static_cast<int64_t>(buff[i] * volume);
+		}
+	}
+}
+
 void AudioProvider::GetInt16MonoAudioWithVolume(int16_t *buf, int64_t start, int64_t count, double volume) const {
 	GetInt16MonoAudio(buf, start, count);
 	if (volume == 1.0) return;
@@ -261,4 +302,4 @@ void SaveAudioClip(AudioProvider const& provider, fs::path const& path, int star
 		out.write(buf);
 	}
 }
-}
+}

libaegisub/include/libaegisub/audio/provider.h

@@ -45,6 +45,7 @@ class AudioProvider {
 	virtual ~AudioProvider() = default;
 
 	void GetAudio(void *buf, int64_t start, int64_t count) const;
+	void GetAudioWithVolume(void *buf, int64_t start, int64_t count, double volume) const;
 	void GetInt16MonoAudio(int16_t* buf, int64_t start, int64_t count) const;
 	void GetInt16MonoAudioWithVolume(int16_t *buf, int64_t start, int64_t count, double volume) const;
 

src/async_video_provider.cpp

@@ -25,6 +25,12 @@
 
 #include <libaegisub/dispatch.h>
 
+#if BOOST_VERSION >= 106900
+#include <boost/gil.hpp>
+#else
+#include <boost/gil/gil_all.hpp>
+#endif
+
 enum {
 	NEW_SUBS_FILE = -1,
 	SUBS_FILE_ALREADY_LOADED = -2
@@ -81,6 +87,55 @@ std::shared_ptr<VideoFrame> AsyncVideoProvider::ProcFrame(int frame_number, doub
 	return frame;
 }
 
+VideoFrame AsyncVideoProvider::GetBlankFrame(bool white) {
+	VideoFrame result;
+	result.width = GetWidth();
+	result.height = GetHeight();
+	result.pitch = result.width * 4;
+	result.flipped = false;
+	result.data.resize(result.pitch * result.height, white ? 255 : 0);
+	return result;
+}
+
+VideoFrame AsyncVideoProvider::GetSubtitles(double time) {
+	// We want to combine all transparent subtitle layers onto one layer.
+	// Instead of alpha blending them all together, which can be messy and cause
+	// rounding errors, we draw them once on a black frame and once on a white frame,
+	// and solve for the color and alpha. This has the benefit of being independent
+	// of the subtitle provider, as long as the provider works by alpha blending.
+	VideoFrame frame_black = GetBlankFrame(false);
+	if (!subs) return frame_black;
+	VideoFrame frame_white = GetBlankFrame(true);
+
+	subs_provider->LoadSubtitles(subs.get());
+	subs_provider->DrawSubtitles(frame_black, time / 1000.);
+	subs_provider->DrawSubtitles(frame_white, time / 1000.);
+
+	using namespace boost::gil;
+	auto blackview = interleaved_view(frame_black.width, frame_black.height, (bgra8_pixel_t*) frame_black.data.data(), frame_black.width * 4);
+	auto whiteview = interleaved_view(frame_white.width, frame_white.height, (bgra8_pixel_t*) frame_white.data.data(), frame_white.width * 4);
+
+	transform_pixels(blackview, whiteview, blackview, [=](const bgra8_pixel_t black, const bgra8_pixel_t white) -> bgra8_pixel_t {
+		int a = 255 - (white[0] - black[0]);
+
+		bgra8_pixel_t ret;
+		if (a == 0) {
+			ret[0] = 0;
+			ret[1] = 0;
+			ret[2] = 0;
+			ret[3] = 0;
+		} else {
+			ret[0] = black[0] / (a / 255.);
+			ret[1] = black[1] / (a / 255.);
+			ret[2] = black[2] / (a / 255.);
+			ret[3] = a;
+		}
+		return ret;
+	});
+
+	return frame_black;
+}
+
 static std::unique_ptr<SubtitlesProvider> get_subs_provider(wxEvtHandler *evt_handler, agi::BackgroundRunner *br) {
 	try {
 		return SubtitlesProviderFactory::GetProvider(br);

src/async_video_provider.h

@@ -78,6 +78,9 @@ class AsyncVideoProvider {
 
 	std::vector<std::shared_ptr<VideoFrame>> buffers;
 
+	// Returns a monochromatic frame with the current dimensions
+	VideoFrame GetBlankFrame(bool white);
+
 public:
 	/// @brief Load the passed subtitle file
 	/// @param subs File to load
@@ -108,6 +111,15 @@ class AsyncVideoProvider {
 	/// @brief raw   Get raw frame without subtitles
 	std::shared_ptr<VideoFrame> GetFrame(int frame, double time, bool raw = false);
 
+	/// @brief Synchronously get the subtitles with transparent background
+	/// @brief time  Exact start time of the frame in seconds
+	///
+	/// This function is not used for drawing the subtitles on the screen and is not
+	/// guaranteed that drawing the resulting image on the current raw frame exactly
+	/// results in the current rendered frame with subtitles. This function is for
+	/// purposes like copying the current subtitles to the clipboard.
+	VideoFrame GetSubtitles(double time);
+
 	/// Ask the video provider to change YCbCr matricies
 	void SetColorSpace(std::string const& matrix);
 

src/audio_player_alsa.cpp

@@ -79,6 +79,7 @@ class AlsaPlayer final : public AudioPlayer {
 	std::atomic<double> volume{1.0};
 	int64_t start_position = 0;
 	std::atomic<int64_t> end_position{0};
+	bool fallback_mono16 = false;	// whether to convert to 16 bit mono. FIXME: more flexible conversion
 
 	std::mutex position_mutex;
 	int64_t last_position = 0;
@@ -88,6 +89,8 @@ class AlsaPlayer final : public AudioPlayer {
 
 	std::thread thread;
 
+	snd_pcm_format_t GetPCMFormat(const agi::AudioProvider *provider);
+
 	void PlaybackThread();
 
 	void UpdatePlaybackPosition(snd_pcm_t *pcm, int64_t position)
@@ -115,6 +118,36 @@ class AlsaPlayer final : public AudioPlayer {
 	void SetEndPosition(int64_t pos) override;
 };
 
+snd_pcm_format_t AlsaPlayer::GetPCMFormat(const agi::AudioProvider *provider) {
+	if (provider->AreSamplesFloat()) {
+		switch (provider->GetBytesPerSample()) {
+			case 4:
+				return SND_PCM_FORMAT_FLOAT_LE;
+			case 8:
+				return SND_PCM_FORMAT_FLOAT64_LE;
+			default:
+				fallback_mono16 = true;
+				return SND_PCM_FORMAT_S16_LE;
+		}
+	} else {
+		switch (provider->GetBytesPerSample()) {
+			case 1:
+				return SND_PCM_FORMAT_U8;
+			case 2:
+				return SND_PCM_FORMAT_S16_LE;
+			case 3:
+				return SND_PCM_FORMAT_S24_LE;
+			case 4:
+				return SND_PCM_FORMAT_S32_LE;
+			case 8:
+				return SND_PCM_FORMAT_S32_LE;
+			default:
+				fallback_mono16 = true;
+				return SND_PCM_FORMAT_S16_LE;
+		}
+	}
+}
+
 void AlsaPlayer::PlaybackThread()
 {
 	std::unique_lock<std::mutex> lock(mutex);
@@ -126,33 +159,19 @@ void AlsaPlayer::PlaybackThread()
 	BOOST_SCOPE_EXIT_ALL(&) { snd_pcm_close(pcm); };
 
 do_setup:
-	snd_pcm_format_t pcm_format;
-	switch (/*provider->GetBytesPerSample()*/ sizeof(int16_t))
-	{
-	case 1:
-		LOG_D("audio/player/alsa") << "format U8";
-		pcm_format = SND_PCM_FORMAT_U8;
-		break;
-	case 2:
-		LOG_D("audio/player/alsa") << "format S16_LE";
-		pcm_format = SND_PCM_FORMAT_S16_LE;
-		break;
-	default:
-		return;
-	}
+	snd_pcm_format_t pcm_format = GetPCMFormat(provider);
 	if (snd_pcm_set_params(pcm,
 	                       pcm_format,
 	                       SND_PCM_ACCESS_RW_INTERLEAVED,
-	                       /*provider->GetChannels()*/ 1,
+	                       fallback_mono16 ? 1 : provider->GetChannels(),
 	                       provider->GetSampleRate(),
 	                       1, // allow resample
 	                       100*1000 // 100 milliseconds latency
 	                      ) != 0)
 		return;
 	LOG_D("audio/player/alsa") << "set pcm params";
 
-	//size_t framesize = provider->GetChannels() * provider->GetBytesPerSample();
-	size_t framesize = sizeof(int16_t);
+	size_t framesize = fallback_mono16 ? sizeof(int16_t) : provider->GetChannels() * provider->GetBytesPerSample();
 
 	while (true)
 	{
@@ -176,7 +195,11 @@ void AlsaPlayer::PlaybackThread()
 		{
 			auto avail = std::min(snd_pcm_avail(pcm), (snd_pcm_sframes_t)(end_position-position));
 			decode_buffer.resize(avail * framesize);
-			provider->GetInt16MonoAudioWithVolume(reinterpret_cast<int16_t*>(decode_buffer.data()), position, avail, volume);
+			if (fallback_mono16) {
+				provider->GetInt16MonoAudioWithVolume(reinterpret_cast<int16_t*>(decode_buffer.data()), position, avail, volume);
+			} else {
+				provider->GetAudioWithVolume(decode_buffer.data(), position, avail, volume);
+			}
 
 			snd_pcm_sframes_t written = 0;
 			while (written <= 0)
@@ -236,7 +259,11 @@ void AlsaPlayer::PlaybackThread()
 
 			{
 				decode_buffer.resize(avail * framesize);
-				provider->GetInt16MonoAudioWithVolume(reinterpret_cast<int16_t*>(decode_buffer.data()), position, avail, volume);
+				if (fallback_mono16) {
+					provider->GetInt16MonoAudioWithVolume(reinterpret_cast<int16_t*>(decode_buffer.data()), position, avail, volume);
+				} else {
+					provider->GetAudioWithVolume(decode_buffer.data(), position, avail, volume);
+				}
 				snd_pcm_sframes_t written = 0;
 				while (written <= 0)
 				{
@@ -353,4 +380,4 @@ std::unique_ptr<AudioPlayer> CreateAlsaPlayer(agi::AudioProvider *provider, wxWi
 	return agi::make_unique<AlsaPlayer>(provider);
 }
 
-#endif // WITH_ALSA
+#endif // WITH_ALSA