Really bad code to critique

[Dave Mateer]

Thanks, Tim; VERY helpful.

About the syncX(), I'm not quite sure at this point *why* it is necessary. But if I don't include it, the video only shows up on the widget about 1/10 of the time. With that call, it always works. Again, a lot of this is a black box to me at this point ... I'm trying to get up to speed as quickly as I can. Your comments were very helpful.



-----Original Message-----
From: gstreamer-devel-bounces+dave_mateer=ntm.org at lists.freedesktop.org [mailto:gstreamer-devel-bounces+dave_mateer=ntm.org at lists.freedesktop.org] On Behalf Of Tim-Philipp Müller
Sent: Thursday, May 03, 2012 3:11 AM
To: gstreamer-devel at lists.freedesktop.org
Subject: Re: Really bad code to critique

On Tue, 2012-05-01 at 16:55 -0400, Dave Mateer wrote:

Hi Dave,

> Among other things, the fact that I'm testing for audio vs. video pad 
> by comparing a string prefix seems fragile and smelly. That can't be 
> right.

Well, in this context ([uri]decodebin[2]) it's quite okay. You know the output is going to be either "raw" audio or "raw" video.

If you want to do it "more" correctly, you could check for "audio/x-raw"
or "video/x-raw" prefixes, or you can check for the full media types explicitly, but then you need at least four checks: audio/x-raw-int or audio/x-raw-float for audio (in 0.10; in 0.11/1.0 this is unified as
audio/x-raw) and video/x-raw-rgb or video/x-raw-yuv or possibly even video/x-raw-gray for video (in 0.10; in 0.11/1.0 this is unified as video/x-raw).

Checking for media type strings is quite kosher. It's how it's supposed to be done in this case.

> Also, it seems like I'm leaking memory in the call to 
> gst_bin_get_by_name because the docs say that transfers ownership (not 
> just a reference count?). Not sure about that one. I'm sure there are 
> many other things as well.

My docs say "Caller owns returned reference", which isn't quite the same as "transfers ownership" IMHO, but it could be worded better I guess.

The refcount of the element/bin/object returned will be increased, and you are responsible for calling gst_object_unref() on it when you are done with it. This is how most element/bin/object related API works in GStreamer, for thread-safety reasons.


> static void on_pad_added(GstElement *element,
>                          GstPad *new_pad,
>                          gpointer user_data) {
>   GstCaps *caps = gst_pad_get_caps_reffed(new_pad);
>   gchar *name = gst_caps_to_string(caps);

This is fine, and easy to do. Something like this would be using the API more explicitly:

GstStructure *s = gst_caps_get_structure (caps, 0); const gchar *media_type = gst_structure_get_name (s);

or:

if (gst_structure_has_name (s, "audio/x-raw-int")
  || gst_structure_has_name (s, "audio/x-raw-float")) {
  g_print ("Audio!\n");
} else if (...) {
  ..
} else {
}

>   GstBin *pipeline = (GstBin*)user_data;
> 
>   if (g_str_has_prefix(name, "audio")) {
>     g_print("audio\n");
> 
>     // Leaking memory?? gst_bin_get_by_name transfers ownership?
>     GstElement *audio_sink = gst_bin_get_by_name(pipeline, "audio-sink");
>     if (!audio_sink) {
>       throw std::runtime_error("Could not extract audio sink from pipeline");
>     }

Yes, you'll need to gst_object_unref (audio_sink) later;

>     GstPad *sink_pad = gst_element_get_static_pad(audio_sink, "sink");
>     if (!sink_pad) {
>       throw std::runtime_error("Could not get sink pad from audio element.");
>     }
>     if (GST_PAD_LINK_FAILED(gst_pad_link(new_pad, sink_pad))) {
>       throw std::runtime_error("Failed to link audio src and sink");
>     }
> 
>     gst_object_unref(sink_pad);
>   } else {
>     g_print("video\n");
> 
>     // Leaking memory?? gst_bin_get_by_name transfers ownership?
>     GstElement *video_sink = gst_bin_get_by_name(pipeline, "video-sink");
>     if (!video_sink) {
>       throw std::runtime_error("Could not extract video sink from pipeline");
>     }

unref video_sink later too.

>     GstPad *sink_pad = gst_element_get_static_pad(video_sink, "sink");
>     if (!sink_pad) {
>       throw std::runtime_error("Could not get sink pad from video element.");
>     }
> 
>     if (GST_PAD_LINK_FAILED(gst_pad_link(new_pad, sink_pad))) {
>       throw std::runtime_error("Failed to link video src and sink");
>     }
> 
>     gst_object_unref(sink_pad);
>   }
> 
>   g_free(name);
>   gst_caps_unref(caps);
> }
> 
> MainWindow::MainWindow(QWidget *parent)
>   : QMainWindow(parent),
>     ui(new Ui::MainWindow) {
>   ui->setupUi(this);
> 
>   // Create the top-level pipeline.
>   pipeline_ = gst_pipeline_new(NULL);
>   if (!pipeline_) {
>     throw std::runtime_error("Could not create pipeline");
>   }
> 
>   // Create the decode bin for our video source.
>   GstElement *src = gst_element_factory_make("uridecodebin", NULL);
>   if (!src) {
>     throw std::runtime_error("Could not create uridecodebin");
>   }
>   g_object_set(src, "uri", "file:///path/to/media.ogg", NULL);
> 
>   // Add the decode bin to the pipeline.
>   if (!gst_bin_add(GST_BIN(pipeline_), src)) {
>     throw std::runtime_error("Could not add uridecodebin to pipeline");
>   }
> 
>   // Create the video sink. This will be linked in the pad-added signal handler.
>   GstElement *video_sink = gst_element_factory_make("xvimagesink",
>                                                     "video-sink");
>   if (!video_sink) {
>     throw std::runtime_error("Could not create xvimagesink");
>   }
>   if (!gst_bin_add(GST_BIN(pipeline_), video_sink)) {
>     throw std::runtime_error("Could not add video sink to pipeline");
>   }
>   WId id = ui->video_widget->winId();
>   gst_x_overlay_set_window_handle(GST_X_OVERLAY(video_sink), id);
>   qApp->syncX();

Out of curiosity, why is the syncX() call required? Just for comparison, in Gtk, you'd need to realize the video widget / application window before being able to get the xid from it, and you would also need to tell Gtk before that that you want it to create a "native" window for the video widget (instead of drawing on the parent widget's). I guess that's not required here. If you do run into problems, there are some Qt overlay examples in gst-plugins-base/tests/examples/.

>   // Create the audio sink. This will be linked in the pad-added signal handler.
>   GstElement *audio_sink = gst_element_factory_make("autoaudiosink",
>                                                     "audio-sink");
>   if (!audio_sink) {
>     throw std::runtime_error("Could not create autoaudiosink");
>   }
>   if (!gst_bin_add(GST_BIN(pipeline_), audio_sink)) {
>     throw std::runtime_error("Could not add audio sink to pipeline");
>   }
> 
>   // Register our interest in the pad-added signal so we can connect our sinks.
>   g_signal_connect(src, "pad-added", G_CALLBACK(on_pad_added), 
> pipeline_);
> 
>   // Start the playback.
>   gst_element_set_state(pipeline_, GST_STATE_PLAYING); }
> 
> MainWindow::~MainWindow() {
>   gst_element_set_state (pipeline_, GST_STATE_NULL);
>   gst_object_unref(pipeline_);
>   delete ui;
> }


Cheers
 -Tim

_______________________________________________
gstreamer-devel mailing list
gstreamer-devel at lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/gstreamer-devel