🐛 `ipydatagrid>=1.3` doesn't support integer pandas column names

[jgunstone]

... as unwanted column created with string name matching the integer on set_cell_value

I'm just trying to update ipyautoui to ipydatagrid>1.3.1 and came across this tricky little issue:

Describe the bug

a new column with the string name of the edited column is created. This only happens when retrieving the data in a new cell.

To Reproduce

from ipydatagrid import DataGrid
import pandas as pd

df = pd.DataFrame({0:{"a":"a", "b": 1}, 1:{"a":"abba", "b": 2}})
display(df)
gr = DataGrid(df)
assert list(gr.data.columns) == [0, 1]
column_name, primary_key_value, new_value = 1, 'a', "my new string"
gr.set_cell_value(column_name, primary_key_value, new_value)
display(gr.data)
assert list(gr._data["data"]) == ["key", 0, 1, "ipydguuid"]

display(gr._data["data"].columns)
gr._data["data"]

Expected behavior Behaves as expected for non-integer column names

Environment (please complete the following information):

• Operating System and Version: [e.g. Ubuntu Linux 22.04]
• Browser Chrome
• ipydatagrid=1.3.1
• jupyterlab=4.1.8

@martinRenou - maybe you could help?

[jgunstone]

feels like a weird jupyter related bug so I just tried with jupyterlab=4.2 https://github.com/jupyterlab/jupyterlab/issues/15801

and it worked!

looks like that is slated for this week so I'll just hang tight until its out and try again

[jgunstone]

just tried with the jupyterlab==4.2 and unfortunately I'm getting the same original error

if I make the col names strings not numbers the bad behaviour stops, i.e.

it seems as though on the JS side of ipydatagrid (or in lumino?) the integer col names are being coerced to strings and an additional column is added.

looking through the js side of things:

https://github.com/bloomberg/ipydatagrid/blob/23b6287f63c99e375d3aa09fb8eadeed9389129f/js/core/viewbasedjsonmodel.ts#L564-L573

https://github.com/bloomberg/ipydatagrid/blob/23b6287f63c99e375d3aa09fb8eadeed9389129f/js/core/viewbasedjsonmodel.ts#L205-L219

https://github.com/jupyterlab/lumino/blob/43ae67c5cfe90b04acff3bbdd12cce9bb7a879bf/packages/datagrid/src/jsonmodel.ts#L150-L158

it appears from the above that there is a requirement to use a string for the name (though this wasn't previously enforced). FYI the referenced frictionless schema spec doesn't explicitly require the column name to be a string.

Does ipydatagrid require the pandas DataFrame to have string col names? should it?