GraphQL support / accessor plugin system

[maartenbreddels]

I recently added graphql support in vaex (another dataframe library), and just added support for pandas as well: vaexio/vaex#446

However, this is only available after importing vaex.graphql, e.g.:

In vaex, I use a plugin system using the entry point system, e.g:

entry_points={
        'vaex.dataframe.accessor': ['graphql = vaex.graphql:DataFrameAccessorGraphQL'],
    },

Is this maybe something pandas can do as well? That would mean that simply installing vaex.graphql would enable this.

Note: in vaex I lazily import the modules, so if df.graphql never gets accessed, vaex.graphql will never be imported.

[jbrockmendel]

@datapythonista would this fit into the plugin idea you mentioned elsewhere?

[datapythonista]

That looks interesting. I don't know much about graphql, and the plugin system I proposed so far is for IO, but I think would be good to have this in pandas.

But the way it's implemented it should be as easy as decorating that function or class to make it a DataFrame accessor I think. That would solve the immediate problem.

And using entry points to extend pandas is surely something we can consider. I'm in my phone, and travelling, so won't be able to have a look in the next few days probably, but if you want to create an issue for that with more details on what you implemented that would be great.

And if you have ideas on how to make extensions compatible both with pandas and vaex, even better.