libvlc 1.1.0 not freeing resources

[AEtH]

Compiling myself is not a problem, I compiled 1.0.6 twice when I needed it. What I'm talking about is about log messages in win32, those obtained with libvlc_iterator_next(). I'm totally uncapable of obtaining any message under win32, but libvlc_log_count() returns me the total log messages everytime a call it, so I know that there are messages but I cannot obtain them.

[Jean-Baptiste Kempf]

The related code is not big, a few thousands line of code, in modules/video_output/msw/

Compiling VLC for windows is not that difficult nowadays.

[AEtH]

Yes, I found myself the modules involving win32 video output. I'll try to understand the changes involved in 1.1.0, comparing the new functions with those functions in 1.0.6. Should I ask questions regarding those modules to you or to someone else? Besides, it would be so helpfull knowing how the functions are called, because I suppose that the functions in 1.0.6 will not work in 1.1.0.

[Jean-Baptiste Kempf]

1.1.1 has huge changes in video output modules. You cannot just use the 1.0.6 code. You need to fix 1.1.0 code.

[AEtH]

Yes, I supossed that. Can I use the libvlc logging functions or they are broken in win32?? I tried some of them but they seem a little buggy.

[XilasZ]

Log functions works well in windows, i've used them since almost two years now. It's a little tricky to use, but it works.
You can take a look at my sources here (http://fritivi.git.sourceforge.net/git/ ... ;a=summary), and specifically the wrapper and the log window.
The code use v1.0.x, But it works the same way with 1.1, just remove the exceptions.

[AEtH]

Well... I made logging work, maybe with a debug output I can see anything. Thanks for the piece of code XilasZ, I had everything solved... but if I get a log iterator just before the creation of the libvlc instance... there is a really BIG chance that there are no logged messages, that's why I did not obtained anything

Thanks

[Jean-Baptiste Kempf]

Did you solve the resources issues?

[AEtH]

I'm on holiday, until monday I'll not start again. As my fiancee thinks that programming is "job", I had to slow down a bit. On monday I'll begin again, but now I can get log messages, so maybe this can help a lot. I'll post the results of the debugging around tuesday.

[Jean-Baptiste Kempf]

[AEtH]

Well, I have some news, I noticed some differences between 1.0.6 and 1.1.1 involving closing directX:

First, let's go with 1.0.6. This is the log involving video opening (just the directX part):

Code: Select all

[VIDEO OUTPUT] [VOUT_DIRECTX] --> creating DirectXEventThread [GENERIC] [MAIN] ---------------> thread (Vout Events Thread) created at priority 0 (directx.c:282) [GENERIC] [MAIN] ---------------> thread started [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCreateWindow [WINDOW] [MAIN] ----------------> looking for hwnd module: 1 candidate [WINDOW] [MAIN] ----------------> using hwnd module "drawable" [VIDEO OUTPUT] [VOUT_DIRECTX] --> created video sub-window [VIDEO OUTPUT] [VOUT_DIRECTX] --> Vout EventThread running [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXInitDDraw [VIDEO OUTPUT] [VOUT_DIRECTX] --> directx-device: [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXEnumCallback: Controlador de pantalla primaria, display [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXEnumCallback: NVIDIA Quadro NVS 285, \\.\DISPLAY1 [VIDEO OUTPUT] [VOUT_DIRECTX] --> selecting NVIDIA Quadro NVS 285, \\.\DISPLAY1 [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXEnumCallback: NVIDIA Quadro NVS 285, \\.\DISPLAY2 [VIDEO OUTPUT] [VOUT_DIRECTX] --> screen dimensions (0x0,1280x1024) [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectDraw Capabilities: overlay=1 yuvoverlay=1 can_deinterlace_overlay=1 colorkey=1 stretch=1 bltfourcc=1 [VIDEO OUTPUT] [VOUT_DIRECTX] --> align_boundary_src=0,0 align_boundary_dest=1,1 align_size_src=0,0 align_size_dest=0,0 [VIDEO OUTPUT] [VOUT_DIRECTX] --> End DirectXInitDDraw [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCreateDisplay [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCreateClipper [VIDEO OUTPUT] [VOUT_DIRECTX] --> disabling screen saver [VIDEO OUTPUT] [MAIN] ----------> using video output module "vout_directx" [VIDEO OUTPUT] [MAIN] ----------> Deinterlacing available [VIDEO OUTPUT] [VOUT_DIRECTX] --> NewPictureVec overlay:yes chroma:YV12 [VIDEO OUTPUT] [VOUT_DIRECTX] --> End NewPictureVec (failed) [VIDEO OUTPUT] [VOUT_DIRECTX] --> NewPictureVec overlay:yes chroma:YV12 [VIDEO OUTPUT] [VOUT_DIRECTX] --> End NewPictureVec (failed) [VIDEO OUTPUT] [VOUT_DIRECTX] --> NewPictureVec overlay:yes chroma:YUY2 [VIDEO OUTPUT] [VOUT_DIRECTX] --> End NewPictureVec (failed) [VIDEO OUTPUT] [VOUT_DIRECTX] --> Could not initialize directx overlay [VIDEO OUTPUT] [VOUT_DIRECTX] --> NewPictureVec overlay:no chroma:YV12 [VIDEO OUTPUT] [VOUT_DIRECTX] --> created plain surface of chroma:YV12 [VIDEO OUTPUT] [VOUT_DIRECTX] --> End NewPictureVec (succeeded) [VIDEO OUTPUT] [MAIN] ----------> got 1 direct buffer(s) [VIDEO OUTPUT] [MAIN] ----------> pic render sz 384x288, of (0,0), vsz 384x288, 4cc I420, ar 270857:216000, sar 1:1, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> pic in sz 384x288, of (0,0), vsz 384x288, 4cc I420, ar 270857:216000, sar 79:84, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> pic out sz 384x288, of (0,0), vsz 384x288, 4cc I420, ar 270857:216000, sar 79:84, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> direct render, mapping render pictures 0-14 to system pictures 1-15 [DECODER] [MAIN] ---------------> End of video preroll

And this is the log involving video closing:

Code: Select all

[VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseSurface [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseDisplay [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseDisplay clipper [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseDisplay display [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseDDraw [GENERIC] [VOUT_DIRECTX] -------> DirectXEventThread terminating [VIDEO OUTPUT] [VOUT_DIRECTX] --> DirectXCloseWindow [VIDEO OUTPUT] [VOUT_DIRECTX] --> WinProc WM_DESTROY [WINDOW] [MAIN] ----------------> removing module "drawable" [GENERIC] [MAIN] ---------------> thread ended [GENERIC] [MAIN] ---------------> thread times: real 0m8.218750s, kernel 0m0.000000s, user 0m0.000000s [VIDEO OUTPUT] [MAIN] ----------> removing module "vout_directx"

Now, let's go with 1.1.1. This is the log involving video opening (just the directX part):

Code: Select all

[GENERIC] [DIRECTX] ------------> creating Vout EventThread [GENERIC] [DIRECTX] ------------> DirectXCreateWindow [WINDOW] [MAIN] ----------------> looking for vout window hwnd module: 1 candidate [WINDOW] [MAIN] ----------------> using vout window hwnd module "drawable" [GENERIC] [DIRECTX] ------------> created video sub-window [GENERIC] [DIRECTX] ------------> Vout EventThread running [GENERIC] [DIRECTX] ------------> disabling screen saver [GENERIC] [DIRECTX] ------------> directx-device: [GENERIC] [DIRECTX] ------------> DirectXEnumCallback: NVIDIA Quadro NVS 285, \\.\DISPLAY1 [GENERIC] [DIRECTX] ------------> selecting NVIDIA Quadro NVS 285, \\.\DISPLAY1 [GENERIC] [DIRECTX] ------------> DirectXEnumCallback: NVIDIA Quadro NVS 285, \\.\DISPLAY2 [GENERIC] [DIRECTX] ------------> screen dimensions (0x0,1280x1024) [GENERIC] [DIRECTX] ------------> DirectDraw Capabilities: overlay=1 yuvoverlay=1 can_deinterlace_overlay=1 colorkey=1 stretch=1 bltfourcc=1 [GENERIC] [DIRECTX] ------------> align_boundary_src=0,0 align_boundary_dest=1,1 align_size_src=0,0 align_size_dest=0,0 [GENERIC] [MAIN] ---------------> VoutDisplayEvent 'resize' 392x322 window [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,784,1034,1024 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,236 [GENERIC] [DIRECTX] ------------> YUV plain surface created successfully [GENERIC] [MAIN] ---------------> using vout display module "directx" [VIDEO OUTPUT] [MAIN] ----------> option direct3d-desktop does not exist [VIDEO OUTPUT] [MAIN] ----------> using video output module "vout_wrapper" [VIDEO OUTPUT] [MAIN] ----------> got 1 direct buffer(s) [VIDEO OUTPUT] [MAIN] ----------> pic render sz 384x288, of (0,0), vsz 384x288, 4cc I420, sar 1:1, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> pic in sz 384x288, of (0,0), vsz 384x288, 4cc I420, sar 79:84, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> pic out sz 384x288, of (0,0), vsz 384x288, 4cc I420, sar 1:1, msk r0x0 g0x0 b0x0 [VIDEO OUTPUT] [MAIN] ----------> direct render, mapping render pictures 0-23 to system pictures 1-24 [GENERIC] [MAIN] ---------------> VoutDisplayEvent 'resize' 316x252 window [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,777,958,1014 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,770,958,1022 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,770,958,1022 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [DECODER] [AVCODEC] ------------> using direct rendering [DECODER] [MAIN] ---------------> End of video preroll

And this is the log involving video closing:

Code: Select all

[GENERIC] [DIRECTX] ------------> DirectXEventThread terminating [GENERIC] [DIRECTX] ------------> DirectXCloseWindow [GENERIC] [DIRECTX] ------------> WinProc WM_DESTROY [WINDOW] [MAIN] ----------------> removing module "drawable" [GENERIC] [MAIN] ---------------> removing module "directx" [VIDEO OUTPUT] [MAIN] ----------> removing module "vout_wrapper"

Am I missing something or there is a bunch of code referring closing surfaces in 1.0.6 that doesn't exists in 1.1.1? Anyway, the application is running, maybe the debug of the crash can tell us more.

[fenrir]

There has been some debug messages removed explaining the differences in the logs.
About the leak you experience, could you try the following patch ?

Code: Select all

diff --git a/modules/video_output/msw/directx.c b/modules/video_output/msw/directx.c index 6b0a6df..b15d61b 100644 --- a/modules/video_output/msw/directx.c +++ b/modules/video_output/msw/directx.c @@ -1201,6 +1201,8 @@ static void DirectXDestroyPictureResource(vout_display_t *vd) { vout_display_sys_t *sys = vd->sys; + if (sys->resource.p_sys->front_surface != sys->resource.p_sys->surface) + DirectXDestroySurface(sys->resource.p_sys->surface); DirectXDestroySurface(sys->resource.p_sys->front_surface); }

[AEtH]

What are you missing in the logs?? I have them just here, would you like me to post anything else? Anyway, I'm compiling now with the diff applied, let's see the results.

Thanks

[fenrir]

Nothing is missing from the logs you posted, I simply said that what you see (between 1.1 and 1.0) is expected. The code in 1.1 has been reworked, and some debug messages has been removed (in the code itself).
I will wait for you results.

[AEtH]

Well, these are the results from yesterday (without the patch). As expected, after 8 hours the player crashed (well, not crashed, just stopped showing anything)... but now I have more info. This is the log output of the crash:

Code: Select all

. . . [GENERIC] [MAIN] ---------------> VoutDisplayEvent 'resize' 316x252 window [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,777,958,1014 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,770,958,1022 [[GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_dst_clipped coords: 642,770,958,1022 [GENERIC] [DIRECTX] ------------> DirectXUpdateRects image_src_clipped coords: 0,0,384,288 [DECODER] [AVCODEC] ------------> using direct rendering [DECODER] [MAIN] ---------------> End of video preroll [VIDEO OUTPUT] [MAIN] ----------> Post-processing available [VIDEO OUTPUT] [MAIN] ----------> vlc_object_find_name(postproc) is not safe! [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) [GENERIC] [DIRECTX] ------------> could not blit surface (error -2005532467) . . .

Now I just started playing with the patch applied, maybe this work better.

[AEtH]

Right now the video is playing for 9 hours, I'll check again tomorrow, but I can say that the patch IS working. I changed your patch from

Code: Select all

if (sys->resource.p_sys->front_surface != sys->resource.p_sys->surface) DirectXDestroySurface(sys->resource.p_sys->surface);

to

Code: Select all

if (sys->resource.p_sys->front_surface != sys->resource.p_sys->surface) { DirectXDestroySurface(sys->resource.p_sys->surface); msg_Dbg(vd, "Patch is working"); } else { msg_Dbg(vd, "Patch is not working"); }

and the line Patch is working appears in the logs. Let's see tomorrow!

[Jean-Baptiste Kempf]

be careful, messages may leak

[AEtH]

Well, the patch works ok but it did not solve the issue, after about 3500 iterations the player crashed as always. I'll take a deeper look at the code, let's see if I can see anything.

[Jean-Baptiste Kempf]

did you run it through gdb ?

[AEtH]

Not yet, I'm working with directx.c. So far I detected a bug, is really strange. As you can see in this piece of code inside Open:

Code: Select all

/* FIXME */ sys->use_overlay = false; //var_CreateGetBool(vd, "overlay"); /* FIXME */

overlay mode is broken. I uncommented that line and yes, it is broken. But, and this is strange, I saw in the logs this line:

Code: Select all

[GENERIC] [DIRECTX] ------------> YUV overlay surface created successfully

and again, that brought me to this code inside DirectXCreatePictureResource:

Code: Select all

bool allow_overlay = var_InheritBool(vd, "overlay");

which should be

Code: Select all

bool allow_overlay = false; //var_InheritBool(vd, "overlay");

just while overlay mode is broken.

Right now I'm running a test with overlay disabled, let's see the results.

[Jean-Baptiste Kempf]

Btw, the patch is inside VLC 1.1.1 and libVLC 1.1.1 (just released).

[AEtH]

I printed directx.c source code, both 1.0.6 and 1.1.1 versions. This is insane!!!

Well, let's dig out!!! (I somebody tells me that I'm not trying to solve the issue by myself, I could explode! )

[garlantinapple]

the canvas handle is not released, and when the player starts playing again, another canvas handle is used.

[Jean-Baptiste Kempf]

canvas? on windows? You mean the HWND ?