Clarification on thread safety

[vikigenius]

The docs say PyMuPDF does not support running on multiple threads.

I am running in async context

async def handle_request(path: str):
    await asyncio.to_thread(extract_pdf(path))

def extract_pdf(path: str):
    with pymupdf.open(path) as pdf:
        do_extract()

Is this pattern inherently unsafe because of the asyncio.to_thread? Even though no pymupdf objects are shared across thread boundaries?

[JorjMcKie]

This is no issue, but either a Discussions item or even better should be discussed on our Forum.

[JorjMcKie]

We can not guarantee correct working in a threaded environment. The critical detail is the cache memory internally maintained by our base library MuPDF.
Parallel task will have problems even when not updating anything.
So at a minimum, all calls to extract_pdf() must be queued by some mechanism to ensure only one instance of it is running at any point in time.

[vikigenius]

Does that apply even to different document objects, if I understand correctly there is a global cache, so you can't access even multiple documents, one document per thread, simultaneously.

[JorjMcKie]

If you absolutely need parallel processing you should fallback to the multiprocessing module. Every process would have its own copy of MuPDF cache.