feat: dask expr cast

[aidoskanapyanov]

What type of PR is this? (check all applicable)

• 💾 Refactor
• ✨ Feature
• 🐛 Bug Fix
• 🔧 Optimization
• 📝 Documentation
• ✅ Test
• 🐳 Other

Related issues

• Related issue feat: add methods to Dask backend #637
• Closes #

Checklist

• Code follows style guide (ruff)
• Tests added
• Documented the changes

If you have comments or can explain your changes, please do so below.

Implemented without preserving a dtype backend for now. Need to dig more into that :P

[FBruzzesi]

I love this! Thanks for the effort! 🙌🏼
I left a few nitpicks here and there 😇

Regarding the failing test due to missing coverage, branches can't be all reach in the same test run 🥲

[aidoskanapyanov]

Thanks for the review! :)

Some CI failures:

• Coverage is dropping for some reason for pytest-38 (3.8, windows-latest)
• Also, modin constructor is not failing, but marked as failing in pytest

[FBruzzesi]

• Coverage is dropping for some reason for pytest-38 (3.8, windows-latest)

I will take a look and see if I can spot anything

• Also, modin constructor is not failing, but marked as failing in pytest

That's typically good news even though... why is that happening now 😂?

Edit: Feel free to remove such block and check against CI. Locally I was able to get all green dots, but I can't replicate on windows.

[aidoskanapyanov]

I wonder why windows and linux pytest coverage CI show different results. For some reason windows run is flagging lots of lines as uncovered, whereas linux is flagging them as green.

For example, on my linux machine coverage report is saying all good for these lines:

while the windows CI run is flagging lines 84-121:

2024-08-22T18:19:06.5566828Z narwhals\_dask\utils.py                             72     60     46      0    10%   18-45, 51-75, 79-80, 84-121

[aidoskanapyanov]

I wonder why windows and linux pytest coverage CI show different results. For some reason windows run is flagging lots of lines as uncovered, whereas linux is flagging them as green.

Yeah, nevermind that, I figure CI coverage reports are a bit more involved than I understand rn. Gotta dig into it first.

[FBruzzesi]

Hey there!

I think the problem with coverage for python 3.8 is that dask is entirely skipped (min python version is 3.9), therefore as we increase total number of functions, the coverage will gradually decrease.

I think you just unluckily hit the threshold number to push the total below 90%.

As all other workflows are passing, I would not worry too much about it. I think we will need to be smarter about it or come with another threshold value.

[FBruzzesi]

We adjusted the threshold to 85%, now some other unrelated CI error appeared 😂

uv pip install --pre --extra-index-url https://pypi.anaconda.org/scientific-python-nightly-wheels/simple pandas --system
shell: /usr/bin/bash -e {0}
env:
pythonLocation: /opt/hostedtoolcache/Python/3.12.5/x64
PKG_CONFIG_PATH: /opt/hostedtoolcache/Python/3.12.5/x64/lib/pkgconfig
Python_ROOT_DIR: /opt/hostedtoolcache/Python/3.12.5/x64
Python2_ROOT_DIR: /opt/hostedtoolcache/Python/3.12.5/x64
Python3_ROOT_DIR: /opt/hostedtoolcache/Python/3.12.5/x64
LD_LIBRARY_PATH: /opt/hostedtoolcache/Python/3.12.5/x64/lib
error: HTTP status server error (503 Service Unavailable) for url (https://pypi.anaconda.org/scientific-python-nightly-wheels/simple/pandas/)
Error: Process completed with exit code 2.

[aidoskanapyanov]

Thanks @FBruzzesi !
Yeah, server error on the anaconda side by the looks of it. I've rerun the CI manually, it's passing now 😄