Jonas Haag
Jonas Haag
I also sometimes get this (with MP3 files) ```pytb Traceback (most recent call last): ... File "...librosa/core/audio.py", line 157, in load y = sf_desc.read(frames=frame_duration, dtype=dtype, always_2d=False).T File ".../soundfile.py", line 865,...
Yep, seems to be an issue with error reporting in threads. When using the following stress testing script, I also get the occasional `Not a valid SNDFILE* pointer`. ```py import...
Yeah maybe it is a problem with PySoundfile, but definitely not in librosa, as is evident from the latest code snippet I posted.
Ok, looks like it is the same issue as reported here: https://github.com/erikd/libsndfile/issues/279 Error variables not being thread safe. Is this something that you are willing to change in libsndfile (eg....
Is this something you would be willing to review and merge if I contribute it?
I wonder if there is anything that I did wrong to get such snippy responses from you. (They feel snippy to me at least.) It seems not to be your...
@evpobr any thoughts that you can share?
Here are some of my ideas on how to fix it. Curious what you think about them. - A) We change `sf_error` to be a thread local. Pros: simple; Cons:...
It’s not a problem of non-atomic writes. The problem is that sf_errno is a global variable and context switches can happen between sf_errno is set (by the library) and read...
By "return", you mean "write to user-specified pointer"? Otherwise, how would you "return" an error code from, say, `const char *sf_count_t sf_seek(...)` or `const char *sf_get_string(...)`? What do you think...