Error on pyreadstat.write_sav with ver 1.3.1

[albertxli]

Hi,

Reading an excel datafile, and trying to give it some labels for column headers and date values then save as .sav file. But ran into issue while writing into sav. please use this demo data to reproduce the error.

import pandas as pd
import pyreadstat


file_path = "demo.xlsx"
df = pd.read_excel(file_path)

output_path = "demo.sav"
pyreadstat.write_sav(df, output_path)

got the following error:

    pyreadstat.write_sav(df, output_path)
    ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "pyreadstat/pyreadstat.pyx", line 784, in pyreadstat.pyreadstat.write_sav
  File "pyreadstat/_readstat_writer.pyx", line 730, in pyreadstat._readstat_writer.run_write
  File "pyreadstat/_readstat_writer.pyx", line 333, in pyreadstat._readstat_writer.get_narwhals_column_types
  File "pyreadstat/_readstat_writer.pyx", line 227, in pyreadstat._readstat_writer.get_narwhals_str_series_max_length
AttributeError: 'datetime.datetime' object has no attribute 'encode'

I have ran the same code with pyreadstat 1.3.0 and save process was smooth without issues. It seems that the underlying narwhals library is complaining about the object data type from pandas. Let me know if you encounter the same problem with the demo data.

[ofajardo]

Thanks for the perfectly reproducible issue! The issue is now solved on the dev branch. Test it if you get a chance.

If you do, you will notice that your columns with mixed types are transformed to strings. Unfortunately you cannot save with mixed types (strings and dates, ints and strings etc) on a spss file, so the only way to harmonize that is to convert everything to a string, that gets saved to the sav file and that is what you will get back when reading the sav.

Otherwise it is a very interesting case, would you mind if I include the data in the pyreadstat tests?

[albertxli]

Glad to hear it's solved! When I was doing testing runs previously, I downloaded the package wheel from here. However, I only see the latest macOS version available there. Since I'm running on Windows, could you point me to where I should go to test the dev branch since I've never done the testing in such way?

And of course, It's totally fine to include the data in the pyreadstat tests.

[ofajardo]

Right, sorry, the pipeline producing the wheels failed because of change (bug?) introduced in a recent version of narwhals, that's why no new wheels yet. I have reported the issue here. I will wait a bit to see if they repair it and then produce the wheels.

[ofajardo]

@albertxli I finally managed to migrate the pipeline and generate new wheels for version 1.3.2 here. If you could test it would be great, then I would make a release.

[albertxli]

@ofajardo Thanks for the update, I just ran a quick test with the latest wheels, confirm the issue is gone and worked really well!

[ofajardo]

great, thanks for testing! version 1.3.2 is now out!