Expand ~ in file names, like pandas and xarray

[ecomodeller]

Problem

MIKE IO passes file names straight through to mikecore without expanding a leading ~, so ~/data/wl.dfs0 doesn't work. pandas and xarray both expand ~ on read and write, and MIKE IO aims to be consistent with them.

Reads fail with a FileNotFoundError naming the unexpanded path — annoying, but diagnosable:

>>> mikeio.read("~/data/random.dfs0")
FileNotFoundError: ~/data/random.dfs0

Writes are worse — they fail silently. Dataset.to_dfs creates a directory literally named ~ in the current working directory and writes the file there. Nothing lands in $HOME and no error is raised:

>>> ds.to_dfs("~/data/out.dfs0")   # no exception
>>> os.path.exists("~")            # a literal '~' directory, in the cwd
True

A ~ directory inside a repo is also a nuisance to clean up — rm -rf ~ is exactly the command you don't want someone reaching for.

The failure mode was inconsistent across formats, too: PfsDocument.write and Mesh.write use Python's open()/write_text() and so did raise FileNotFoundError, while the dfs writers went through mikecore and silently mis-wrote.

Fix

Add mikeio._path.normalize_path() — expanduser(fspath(...)), mirroring pandas' stringify_path — and apply it at the public API boundaries:

• mikeio.open / mikeio.read, via the Dfs0/Dfs1/Dfs2/Dfs3 constructors, _get_dfsu_info (Dfsu2DH, DfsuLayered, DfsuSpectral), the dfsu factory, and Mesh
• Dataset.to_dfs and DataArray.to_dfs, plus write_dfs0/1/2/3 and write_dfsu
• Mesh.write, GeometryFM2D.to_mesh, GeometryFM3D.to_mesh, Grid2D.to_mesh
• read_pfs and PfsDocument.write
• every generic.py function: scale, fill_corrupt, sum, add, diff, concat, extract, avg_time, quantile, change_datatype, transform
• extract_track when given a file name

normalize_path is a no-op for paths without a leading ~ and is idempotent, so this is backwards-compatible. Because it returns a str, a number of now-redundant str(filename) wrappers are dropped.

Only expanduser is applied — not abspath. pandas does the same, and it keeps _filename (which shows up in repr) as the user wrote it.

Tests

New tests/test_path_expansion.py (35 tests) points $HOME at a tmp_path via monkeypatch and covers read, open, write, mesh, pfs, track, and every generic function. Each write test also asserts no literal ~ directory is created.

Verified these are not vacuous: 32 of the 35 fail on main.

Full suite: 878 passed. mypy clean, ruff check shows only the 3 pre-existing errors on main.

🤖 Generated with Claude Code

[ecomodeller]

@copilot is this behaviour and implementation similar to pandas

[Copilot]

@copilot is this behaviour and implementation similar to pandas

Yes — this is aligned with pandas. pandas’ stringify_path path normalization expands ~ via os.path.expanduser (without converting to absolute paths), and our normalize_path does the same (os.fspath(...) + expanduser) for both read and write entry points.