polars-ls 0.2.4

List directory contents as Polars DataFrames

pols

List directory contents as Polars DataFrames.

Installation

The polars-ls package can be installed with either polars or polars-lts-cpu using the extras by those names:

pip install polars-ls[polars]
pip install polars-ls[polars-lts-cpu]

If Polars is already installed, you can simply pip install polars-ls.

User guidance

Names are relative

Counter to the typical pathlib.Path notion of a name, the names in ls and hence pols are more relative names: hence . is a valid name (if you try accessing the .name attribute of a pathlib Path, it'll come back as "").

>>> cwd = Path.cwd()
>>> cwd / "."
PosixPath('/home/louis/dev/pols')
>>> cwd / ".."
PosixPath('/home/louis/dev/pols/..')
>>> (cwd / ".").name
'pols'
>>> (cwd / "..").name
'..'
>>> Path(".").name
''

Individual files and directories don't mix

The way ls works is that individual files get collected in one 'set' of results and directories in another, and never the two shall meet. If you ls a few files and one or more directories, you'll get one set of reults with all the files and one set for each of the folders. This is because of the previous point: the names shown are relative to the directory 'root' (if you're specifying files individually, the current working directory is the assumed directory 'root', and of course absolute paths always show as absolute so their 'root' is shown too).

(Even if the individual files are in different folders: it's because merging files with different roots whose relative names are being shown would be invalid)

$ ls README.md src src/pols/__init__.py 
README.md  src/pols/__init__.py

src:
pols

To the same effect, the results are grouped in a list of dicts, where the key is the source (either the empty string for the individual files, or the directory root). This allows an identical printout style to ls:

$ ls -A ../.py*
../.python-version

../.pytest_cache:
CACHEDIR.TAG  .gitignore  README.md  v
$ pols -A ../.py*
shape: (1, 1)
┌────────────────────┐
│ name               │
│ ---                │
│ str                │
╞════════════════════╡
│ ../.python-version │
└────────────────────┘
../.pytest_cache:
shape: (4, 1)
┌──────────────┐
│ name         │
│ ---          │
│ str          │
╞══════════════╡
│ README.md    │
│ v            │
│ .gitignore   │
│ CACHEDIR.TAG │
└──────────────┘

Globs (Kleene stars) go 1 level deep

You can use ** in ls and pols but in both cases you only actually get one level, unlike other tools (and Python's glob).

$ ls src/pols/**.py
src/pols/cli.py  src/pols/__init__.py  src/pols/pols.py
$ ls src/pols/*/*.py
src/pols/features/a.py  src/pols/features/A.py  src/pols/features/hide.py
src/pols/features/__init__.py  src/pols/features/p.py

Patterns that don't match will error non-fatally

It's allowed to not match a file, just like in ls:

$ ls *.yaml *.toml *.md
ls: cannot access '*.yaml': No such file or directory
 pyproject.toml   README.md

$ pols *.yaml *.toml *.md
pols: cannot access '*.yaml': No such file or directory
shape: (2, 1)
┌────────────────┐
│ name           │
│ ---            │
│ str            │
╞════════════════╡
│ pyproject.toml │
│ README.md      │
└────────────────┘

OSErrors like FileNotFoundError are non-fatal but can be thrown with raise_on_access

If you want such errors to be fatal, pass raise_on_acecss (--raise-on-access on the command line):

$ pols *.yaml *.toml *.md --raise-on-access
pols: cannot access '*.yaml': No such file or directory
Traceback (most recent call last):
...
FileNotFoundError: No such file or directory

Note that the file expansion and preparation is done before any printing or DataFrame operations, so these errors won't occur mid-way through any Polars computations.

Differences from ls

The design is intended to keep as closely as possible to GNU coreutils ls.

So far one particular divergence is that command line order does not affect pols (e.g. -aA vs -Aa), but there is no real way to implement this with boolean parameters. It could be detected for a CLI but for now I'm prioritising feature completeness over CLI complexity (it is autogenerated from the function signature with argh).

Another is that hide is not disabled by a/A because there is no need to, and this enables filtering hidden files minus some pattern. In ls, --hide silently fails if passed with -a.