The approach you've chosen overall does the job.
It seems to closely mimic the behavior of urllib.parse.urlsplit (in a simplified manner as you've pointed out). This might be a good thing in some cases, where a generalized solution preserving the original behavior is actually needed. However, I am not sure if we have such case here.
As we essentially need to support only urls of types: b2id:// and b2://bucket/file, is there a simpler way to achieve that with less abstractions and code-induced liability?

@olzhasar-reef I replaced the SplitB2Result with a simple namedtuple which should be sufficient. I considered hardcoding the schemes into the regex but that would produce less useful error messages (invalid B2 URI instead of Unsupported URI scheme) so I would prefer to leave that as is.

@irene-sheen-reef
My point is that this whole url split and unsplit flow feels like an unnecessary overkill for our use case.
As scheme, netloc and path are just intermediary tools, we might as well just parse exactly what's needed straight up.

E.g.:

b2id_pattern = re.compile(r'^b2id://(?P[a-zA-Z0-9:_-]+)$')
b2_pattern = re.compile(r'^b2://(?P[a-z0-9-]+)/(?P.+)$')

And pick the appropriate one based on the prefix.

Are there any downsides compared to the approach you've suggested?