Bugfixes and performance improvements

[mxmlnkn]

Hi there,

I'm trying to write a squashfs backend for ratarmount. While doing so, I implemented some fixes and performance improvements, which can be found here.

The workarounds include:

• [ ] Does not create the whole folder hierarchy in memory when only iterating over it to avoid high memory usage for SquashFS images with millions of files.
• [ ] Adds seekable, streamable file object accessor that can be opened given a single number.
• [ ] Uses libdeflate or ISA-L if installed, which a generally faster than the standard zlib.
• [x] Fixes lz4 support.
• [x] Adds lzma support.

Would you be interested in pull requests for any of these? I admit that the new file interface based on a simple int instead of a full inode object might be very specialized to my use case because I serialize it into a SQLite database, so I understand if you wouldn't want to merge that. The same could be said for the other points except the lz4 and lzma support, which I think should be merged upstream.

Additionally, it would be nice to:

• [ ] Add thread locks around the underlying file object so that multiple file objects can be opened and used from multiple threads concurrently.
• [ ] Add type-hinting
• [x] Add tests and a CI
• [ ] Move the build system wholly away from setup.py
• [ ] Remove support for Python <3.6 to add type-hinting and other newer amenities.

[matteomattei]

Thanks for your support... I will be more than happy to merge back your changes to the mainstream. Unfortunately I don't have too much time to dedicate to this repo but if you submit a pull request I will merge it. Can you try to isolate the changes which are not strictly related to your use case in the PR?