@NicolasIRAGNE i've fix encoding issue on windows, we now use utf-16 le instead of utf-16.

I let you decide what we need to do from now.

I'll take a look whenever I have a bit of time, I'm not sure what this does but it seems better than initial check

But I am curious as to what the actual error is. Do we read too many characters? What happens if the file is just a huge numbers of ascii chars?

Also,#375 did this fix the problem as well?

But I am curious as to what the actual error is. Do we read too many characters? What happens if the file is just a huge numbers of ascii chars?

The current problem:

1. we read a chunk of a file
2. we wrongly determine binaries using a too much simple method
3. this leads to include bytes within the final output context
4. the tiktoken package, which uses rust, panic when unwrapping parsing result

Also,#375 did this fix the problem as well?

Yes. This is the main purpose of this with another one : The context should now ignore a lot more (if it's not all) binaries, so the context should be a lot more usable for LLMs on various repositories.