Fixed wrong sample count being reported for certain ogg audio files. … #3579
Conversation
|
What about adding a test for a file that used to fail to open? |
|
If someone can provide a file with appropriate license that causes the issue then I can add it. |
If I want to provide you with files from my project for the need of tests, is there something in particular I should do ? |
|
We don't have some CLA set up to facilitate an easy adoption of resources. I guess one option is to publicly release the files with a reusable license. Elijah who created some assets used in the SFML examples has done so in the past: https://github.com/elijahfhopp/public-domain Maybe for tests it's less relevant? The main "issue" in a sense is distribution on platforms like Debian that only uses assets that are clearly licensed. Then again, they do run our test suite as well during the build step. Anyways, I don't want this to block the fix. Feel free to open a PR with a proposal on how to integrate the asset. |
Title.
It seems like the vorbis implementation has been somewhat buggy for some time.
After opening an ogg stream and reading its header information with
ov_info(), in certain cases (certain files) the internal stream read position isn't properly reset back to the beginning. This causes a subsequent call toov_pcm_total()to return a wrong total sample count. Reading in sample data usingov_read()based on this incorrect sample count will then return less samples than what was reported leading to the entire file failing to load. Explicitly resetting the stream read position back to the beginning of the file usingov_raw_seek()after callingov_info()seems to work around this problem.Assuming that this bug will realistically never get fixed, I don't have any reservations about building in this workaround. Glancing over the vorbis issue tracker it is unlikely most reported bugs will ever get fixed.
All the information necessary to reproduce the bug and test this fix can be found in #3578.
Fixes #3578