[libcamera-devel] [PATCH v11] test: gstreamer: Add test for gstreamer single stream

Nicolas Dufresne nicolas at ndufresne.ca
Tue Aug 17 18:24:05 CEST 2021


Le samedi 14 août 2021 à 02:46 +0300, Laurent Pinchart a écrit :
> From: Vedant Paranjape <vedantparanjape160201 at gmail.com>
> 
> This patch adds a test to test if single stream using libcamera's
> gstreamer element works.
> 
> We need to work around two distinct issues with ASan when enabled in the
> build:
> 
> - glib has a known leak at initialization time. This is covered by the
>   suppression file shipped with glib, but it's not clear how to use it
>   automatically. For now, disable leak detection to avoid test failures.

Are Valgrind suppression usable for Asan ? Glib installs it into:

  /usr/share/glib-2.0/valgrind/glib.supp

For GStreamer suppressions, we don't install them (yet, under discussion). They
are located in each repos. I would enable GStreamer leak tracer though, it does
not have such a high run-time overhead, and will detect any GObject or
GstMiniObject leak, as long as the test calls gst_deinit() at the end (all tests
should).

> 
> - GStreamer spawns a child process to scan plugins. If GStreamer is
>   compiled without ASan (which is likely) but libcamera is, dlopen()ing
>   the libcamera plugin will cause an ASan link order verification
>   failure. Disable the verification child processes to work around the
>   problem. This requires gcc 8 or newer.

Have you considered simply disabling forks ? See gst_registry_fork_set_enabled()

> 
> Signed-off-by: Vedant Paranjape <vedantparanjape160201 at gmail.com>
> Reviewed-by: Paul Elder <paul.elder at ideasonboard.com>
> Reviewed-by: Kieran Bingham <kieran.bingham at ideasonboard.com>
> Tested-by: Kieran Bingham <kieran.bingham@@ideasonboard.com>
> Reviewed-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
> Signed-off-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
> ---
> This version incorporates changes coming from my review of v10, and
> fixes for ASan issues. ASan is a bit of a pain with GStreamer, for two
> independent reasons as explained in the commit message. I'd like to find
> a way to use leak suppression files to handle the glib initialization
> leak, but that's a big rabbit hole. The workaround for the second issue
> is acceptable in my opinion.
> 
> The biggest trouble with these workarounds is that they don't work with
> gcc version older than 8. As the stable version of the most common Linux
> distributions ship gcc 8 or newer, skipping this test for gcc 7 (which
> is the oldest version that libcamera supports) is also acceptable in my
> opinion.
> 
> Changes since v10:
> 
> - Disable ASan leak detection
> - Disable ASan link order verification for child processes
> - Include source_path.h
> - Remove unneeded explicit std::string construction
> - Declare variables at usage site
> - Make constants constexpr
> - Add a variable for the message type
> - Blank space fixes
> ---
>  .../gstreamer_single_stream_test.cpp          | 188 ++++++++++++++++++
>  test/gstreamer/meson.build                    |  19 ++
>  test/meson.build                              |   1 +
>  3 files changed, 208 insertions(+)
>  create mode 100644 test/gstreamer/gstreamer_single_stream_test.cpp
>  create mode 100644 test/gstreamer/meson.build
> 
> diff --git a/test/gstreamer/gstreamer_single_stream_test.cpp b/test/gstreamer/gstreamer_single_stream_test.cpp
> new file mode 100644
> index 000000000000..e26673b3471a
> --- /dev/null
> +++ b/test/gstreamer/gstreamer_single_stream_test.cpp
> @@ -0,0 +1,188 @@
> +/* SPDX-License-Identifier: GPL-2.0-or-later */
> +/*
> + * Copyright (C) 2021, Vedant Paranjape
> + *
> + * ipa_interface_test.cpp - Test the IPA interface
> + */
> +
> +#include <iostream>
> +#include <unistd.h>
> +
> +#include <libcamera/base/utils.h>
> +
> +#include "libcamera/internal/source_paths.h"
> +
> +#include <gst/gst.h>
> +
> +#include "test.h"
> +
> +using namespace std;
> +
> +extern "C" {
> +const char *__asan_default_options()
> +{
> +	/*
> +	 * Disable leak detection due to a known global variable initialization
> +	 * leak in glib's g_quark_init(). This should ideally be handled by
> +	 * using a suppression file instead of disabling leak detection.
> +	 */
> +	return "detect_leaks=false";

Have you consider GST_TRACERS=leaks ? It will heck for refcounted object leaks
on gstreamer/glib side, with very low overhead, and will abort on gst_deinit()
if it found some leaks.


> +}
> +}
> +
> +class GstreamerSingleStreamTest : public Test
> +{
> +protected:
> +	int init() override
> +	{
> +		/*
> +		 * GStreamer spawns a process to run the gst-plugin-scanner
> +		 * helper. If libcamera is compiled with ASan enabled, and as
> +		 * GStreamer is most likely not, this will cause the ASan link
> +		 * order check to fail when gst-plugin-scanner dlopen()s the
> +		 * plugin as many libraries will have already been loaded by
> +		 * then. Work around this issue by disabling the link order
> +		 * check. This will only affect child processes, as ASan is
> +		 * already loaded for this process by the time this code is
> +		 * executed, and should thus hopefully be safe.
> +		 *
> +		 * This option is not available in gcc older than 8, the only
> +		 * option in that case is to skip the test.
> +		 */
> +#if defined(__SANITIZE_ADDRESS__) && !defined(__clang__) && __GNUC__ < 8
> +		return TestSkip;
> +#endif
> +		setenv("ASAN_OPTIONS", "verify_asan_link_order=0", 1);

I think gst_registry_fork_set_enabled(FALSE) would be much cleaner, what do you
think ?

> +
> +		/* Initialize GStreamer */
> +		GError *errInit = nullptr;
> +		if (!gst_init_check(nullptr, nullptr, &errInit)) {
> +			g_printerr("Could not initialize GStreamer: %s\n",
> +				   errInit ? errInit->message : "unknown error");
> +			if (errInit)
> +				g_error_free(errInit);
> +
> +			return TestFail;
> +		}
> +
> +		/*
> +		 * Remove the system libcamera plugin, if any, and add the
> +		 * plugin from the build directory.
> +		 */
> +		GstRegistry *registry = gst_registry_get();
> +		GstPlugin *plugin = gst_registry_lookup(registry, "libgstlibcamera.so");
> +		if (plugin) {
> +			gst_registry_remove_plugin(registry, plugin);
> +			gst_object_unref(plugin);
> +		}
> +
> +		std::string path = libcamera::utils::libcameraBuildPath()
> +				 + "src/gstreamer";
> +		if (!gst_registry_scan_path(registry, path.c_str())) {
> +			g_printerr("Failed to add plugin to registry\n");
> +			gst_deinit();
> +			return TestFail;
> +		}
> +
> +		/* Create the elements */
> +		libcameraSrc_ = gst_element_factory_make("libcamerasrc", "libcamera");
> +		convert0_ = gst_element_factory_make("videoconvert", "convert0");
> +		sink0_ = gst_element_factory_make("fakesink", "sink0");

Please use fakevideosink instead.

> +
> +		/* Create the empty pipeline_ */
> +		pipeline_ = gst_pipeline_new("test-pipeline");
> +
> +		if (!pipeline_ || !convert0_ || !sink0_ || !libcameraSrc_) {

pipeline_ cannot be NULL, glib will abort if malloc failed.

> +			g_printerr("Not all elements could be created. %p.%p.%p.%p\n",
> +				   pipeline_, convert0_, sink0_, libcameraSrc_);
> +			if (pipeline_)
> +				gst_object_unref(pipeline_);
> +			if (convert0_)
> +				gst_object_unref(convert0_);
> +			if (sink0_)
> +				gst_object_unref(sink0_);
> +			if (libcameraSrc_)
> +				gst_object_unref(libcameraSrc_);

Use local g_autoptr() and g_steal_pointer() on success perhaps ? This will
reduce a lot the about of error prone code in failure case. Note that in other
software I've seen, an abort is used instead, as clean exit is just coding
overhead for a test.

An alternative, add them immediately to the pipeline, they are not owned by your
class anyway.

> +			gst_deinit();
> +
> +			return TestFail;
> +		}
> +
> +		return TestPass;
> +	}
> +
> +	void cleanup() override
> +	{
> +		gst_object_unref(pipeline_);
> +		gst_deinit();
> +	}
> +
> +	int run() override
> +	{
> +		GstStateChangeReturn ret;
> +
> +		/* Build the pipeline */
> +		gst_bin_add_many(GST_BIN(pipeline_), libcameraSrc_, convert0_, sink0_, NULL);

This can fail (e.g. duplicate name).

> +		if (gst_element_link_many(libcameraSrc_, convert0_, sink0_, NULL) != TRUE) {
> +			g_printerr("Elements could not be linked.\n");
> +			return TestFail;
> +		}
> +
> +		/* Start playing */
> +		ret = gst_element_set_state(pipeline_, GST_STATE_PLAYING);
> +		if (ret == GST_STATE_CHANGE_FAILURE) {
> +			g_printerr("Unable to set the pipeline to the playing state.\n");
> +			return TestFail;
> +		}
> +
> +		/* Wait until error or EOS or timeout after 2 seconds */
> +		constexpr GstMessageType msgType =
> +			static_cast<GstMessageType>(GST_MESSAGE_ERROR | GST_MESSAGE_EOS);
> +		constexpr GstClockTime timeout = 2000000000;
Perhaps 2 * GST_SECOND.
> +
> +		g_autoptr(GstBus) bus = gst_element_get_bus(pipeline_);
> +		g_autoptr(GstMessage) msg = gst_bus_timed_pop_filtered(bus, timeout, msgType);
> +
> +		gst_element_set_state(pipeline_, GST_STATE_NULL);
> +
> +		/* Parse error message */
> +		if (msg == NULL)
> +			return TestPass;

I would like some minimal validation. I would expect that after 2s some frames
got "rendered" properly. You can read the GstStructure property "stats" from
fakevideosink / fakesink, and read the "rendered" field. Make sure this not zero
perhaps ?

> +
> +		switch (GST_MESSAGE_TYPE(msg)) {
> +		case GST_MESSAGE_ERROR:
> +			gstreamer_print_error(msg);
> +			break;
> +		case GST_MESSAGE_EOS:
> +			g_print("End-Of-Stream reached.\n");
> +			break;
> +		default:
> +			g_printerr("Unexpected message received.\n");
> +			break;
> +		}
> +
> +		return TestFail;
> +	}
> +
> +private:
> +	void gstreamer_print_error(GstMessage *msg)
> +	{
> +		GError *err;
> +		gchar *debug_info;
> +
> +		gst_message_parse_error(msg, &err, &debug_info);
> +		g_printerr("Error received from element %s: %s\n",
> +			   GST_OBJECT_NAME(msg->src), err->message);
> +		g_printerr("Debugging information: %s\n",
> +			   debug_info ? debug_info : "none");
> +		g_clear_error(&err);
> +		g_free(debug_info);
> +	}
> +
> +	GstElement *pipeline_;
> +	GstElement *libcameraSrc_;
> +	GstElement *convert0_;
> +	GstElement *sink0_;
> +};
> +
> +TEST_REGISTER(GstreamerSingleStreamTest)
> diff --git a/test/gstreamer/meson.build b/test/gstreamer/meson.build
> new file mode 100644
> index 000000000000..b99aa0da0ba3
> --- /dev/null
> +++ b/test/gstreamer/meson.build
> @@ -0,0 +1,19 @@
> +# SPDX-License-Identifier: CC0-1.0
> +
> +if not gst_enabled
> +    subdir_done()
> +endif
> +
> +gstreamer_tests = [
> +    ['single_stream_test',   'gstreamer_single_stream_test.cpp'],
> +]
> +gstreamer_dep = dependency('gstreamer-1.0', required: true)
> +
> +foreach t : gstreamer_tests
> +    exe = executable(t[0], t[1],
> +                     dependencies : [libcamera_private, gstreamer_dep],
> +                     link_with : test_libraries,
> +                     include_directories : test_includes_internal)
> +
> +    test(t[0], exe, suite : 'gstreamer', is_parallel : false)
> +endforeach
> diff --git a/test/meson.build b/test/meson.build
> index 3bceb5df586f..d0466f17d7b6 100644
> --- a/test/meson.build
> +++ b/test/meson.build
> @@ -11,6 +11,7 @@ subdir('libtest')
>  
>  subdir('camera')
>  subdir('controls')
> +subdir('gstreamer')
>  subdir('ipa')
>  subdir('ipc')
>  subdir('log')




More information about the libcamera-devel mailing list