Created attachment 373895 [details] [review]
patch for this issue.

@Fei

Do you have a test that shows the problem?

What's the problem exactly?

(In reply to Víctor Manuel Jáquez Leal from comment #2)
> @Fei
> 
> Do you have a test that shows the problem?
> 
> What's the problem exactly?

Hi Victor, due to the licence we can't public the stream, and I put it into our public google drive(gstreamervaapi@gmail.com) gstreamer_bug_stream folder. If you forget the passwd, please ping me through fei.w.wang@intel.com. :)

The stream have 2 different resolution, when the resolution change, the context and surfaces will be destroyed and re-created. This patch makes sure context is created and available before use it to create surface. Otherwise the pipeline will be stuck with some error. I believe any streams with 2 or more resolutions can be reproduce this issue.

error log:
root@T:/media/# gst-launch-1.0 filesrc location=test.mpeg2 ! mpegvideoparse ! vaapimpeg2dec ! fakesink
Setting pipeline to PAUSED ...
Pipeline is PREROLLING ...
Got context from element 'vaapidecode_mpeg2-0': gst.gl.GLDisplay=context, gst.gl.GLDisplay=(GstGLDisplay)"\(GstGLDisplayX11\)\ gldisplayx11-0";
Got context from element 'vaapidecode_mpeg2-0': gst.vaapi.Display=context, gst.vaapi.Display=(GstVaapiDisplay)"\(GstVaapiDisplayDRM\)\ vaapidisplaydrm1";
Redistribute latency...
Pipeline is PREROLLED ...
Setting pipeline to PLAYING ...
New clock: GstSystemClock
Redistribute latency...
0:00:00.228859761 24841      0x12204f0 ERROR                  vaapi gstvaapidecoder_mpeg2.c:1509:gst_vaapi_decoder_mpeg2_start_frame: failed to reset context
0:00:00.228904080 24841      0x12204f0 ERROR            vaapidecode gstvaapidecode.c:755:gst_vaapidecode_handle_frame: decode error -1

@Victor，any other comments on this patch?

Just tested it. I already had the sample. And it is a regression. If you test with release 1.14 the works.

I would like to know on which commit this regression appeared.

It seems the commit that pop up the error is 

commit 82872f42344506c4f5e31c0a485a68a5f40f98cf (HEAD, refs/bisect/bad)
Author: Wangfei <fei.w.wang@intel.com>
Date:   Mon Oct 1 09:26:05 2018 +0800

    libs: context: query surface format before context to create surface.
    
    Before using context to create surface, the supported surface format
    should be checked first.
    
    https://bugzilla.gnome.org/show_bug.cgi?id=797222

This is logical since with commit 82872f4 now it check the formats available in the VAConfig (part of the GstVaapiContext), thus now we need to create first the context and then the surfaces.

@Fei, please add all this information in the commit log:

"""
In gst_vaapi_context_reset(), if the context has to be destroyed, make sure to create it first before allocating its associated surfaces.

This patch fixes a regression introduced in commit 82872f4 because the formats available in the current context now are ensured before creating the context's surfaces.
"

Created attachment 373928 [details] [review]
patch for this issue.

(In reply to Víctor Manuel Jáquez Leal from comment #7)
> This is logical since with commit 82872f4 now it check the formats available
> in the VAConfig (part of the GstVaapiContext), thus now we need to create
> first the context and then the surfaces.
> 
> @Fei, please add all this information in the commit log:
> 
> """
> In gst_vaapi_context_reset(), if the context has to be destroyed, make sure
> to create it first before allocating its associated surfaces.
> 
> This patch fixes a regression introduced in commit 82872f4 because the
> formats available in the current context now are ensured before creating the
> context's surfaces.
> "

Thanks Victor. Added.

attachment 373928 [details] [review] pushed as commit ecdfc623 - libs: context: create context first before using it to create surface.