[Ivorforce]

Note

This code was implemented by AI.

Adds support for memory:// to rdrecord, rdsamp and rdheader, addressing #491 (comment). Supersedes #491.

Adding this was already almost done via fsspec. What was needed was Windows support for path-joining POSIX paths and adding it to the supported protocols list.

Besides that the PR adds documentation and tests.

[tompollard]

Thanks for picking this up! A couple of minor things:

memory://100 doesn't work yet. rdheader and _rdrecord split the path first and then look for the protocol in the directory part (record.py:1864 and record.py:2205). With no directory, that part is just memory:, so it's treated as a local path and you get an error:

FileNotFoundError: ... '<cwd>/memory:/100.hea'

I think checking record_name for the protocol instead, in rdheader, _rdrecord and _infer_sig_len, would fix it. Then we could drop the paragraph in docs/io.rst that asks people to use a directory.

The test fixture's cleanup can fail. If nothing was uploaded, fs.rm("/wfdb-test", recursive=True) raises FileNotFoundError. When a test fails, you then also get an error from the cleanup that hides the real problem. Adding an fs.exists("/wfdb-test") check first should sort it.

Please could you take a look at these?

A couple of other minor things that we could address here or in new PRs:

• rdheader and _infer_sig_len still choose between posixpath.join and os.path.join themselves, so they could use the new _join_path too.
• The threading warning in the docs is correct, because fsspec gives every reader the same BytesIO. Later we might read memory files with fs.cat_file() into a new BytesIO, so each reader gets its own copy and the warning can go.