Fix ogg with extra buffer at end #95
Conversation
|
CC: @Wohlstand for the sample-accurate seek references. |
|
The function that I added is |
| @@ -260,6 +260,7 @@ typedef struct __SOUND_SAMPLEINTERNAL__ | |||
| Uint32 buffer_size; | |||
| void *decoder_private; | |||
| Sint32 total_time; | |||
| Uint32 total_length; | |||
There was a problem hiding this comment.
You add a new member to the struct here, but only init it in VORBIS_open. I think you're supposed to init it at least to some default value in each function that initializes this struct.
|
If anyone wants to compare, here is the same fix in stb_vorbis side. main...ericoporto:SDL_sound-1:fix-stb-vorbis-interleaved
Ok, I tried this and it gives a different third value and will still be wrong, just slightly less wrong (it will append 2 ms instead of 5 ms) |
If this fixes the issue, stb_vorbis patching should be preferred, IMO. |
|
On first look seems ok. I'll try to check some after sleep. |
|
Closing this in favor of #97 |
There is a bug in stb_vorbis that I am not sure how to fix
SDL_sound/src/stb_vorbis.h
Lines 5651 to 5658 in fdcecaf
The n that is added here at the end can be bigger than the size of the buffer. The
current_playback_locwas added in nothings/stb#1295, it will be wrong at the end, but it's correct before the last call where there are samples, so we can use that to workaround and get the true final length.The other approach would be to push the truncation in stb_vorbis, inside
stb_vorbis_get_samples_float_interleaved, the issue there is it would get stb_vorbis more different from it's upstream. I think the actual issue isstb_vorbis_get_frame_floatis called and the result is dismissed, but that result would be the only other way to know the actual remaining length.fix #92