One thing to consider: if you do performance profiling, you'll find that date parsing is one of the slowest things this library does. The ciso8601 package is a C extension that speeds up this process greatly.

Right now it's easy to patch in ciso8601 for iso8601 (m3u8.parser.cast_date_time = ciso8601.parse_datetime).

I support dropping the iso8601 dependency, but it would be good to maintain the ability to patch in ciso8601.

According to the benchmarks on ciso8601, it's a significant improvement over iso8601 (almost 100x), but compared to the built-in function it's a lot closer, so the need to change that should go down a lot with this change and would still be completely possible

On a semi-related note, I added a PR for adding 3.12 to the list of tested versions and I noticed that ciso8601 doesn't have wheel packages for that yet, so it would need to be built from source until those are added

Looks like this isn't a viable solution until Python 3.11 so I'm guessing this needs to wait

FWIW, there is also the option of using backports.datetime_fromisoformat, which brings Python 3.11's parsing code to older versions of Python 3 (with pre-built wheels and high-performance).

I did another version of my PR that uses the built-in on 3.11 and later, but I really liked the idea of using the backports (for the consistent implementation and performance improvement), so I added that to the PR as well. Thanks for the suggestion!