Migrate to click

[dlqqq]

Description

• Closes Use argument parsers to run AI commands #117

Uses click to parse arguments. Notable features:

• Declarative handling of subcommands. Adding new subcommands is done easily by using the @line_magic_parser.command() decorator.
• Support for subcommand arguments. Use the @click.option() or @click.argument() decorators to add new arguments to a subcommand.
• Automatically generated help strings.
• Mitigates the issues of Python version compatibility involved with using argparse, which may include breaking changes version-to-version.

Demo

Screen.Recording.2023-05-23.at.4.00.26.PM.mov

Breaking changes

Language models can no longer be invoked by a line magic. The reason for this is that parsers do not handle variadic arguments correctly when using subcommands. Both argparse and click fail for such a use-case. So for example, %ai list openai will make click think that you want to execute the list model with the prompt openai, rather than dispatching to openai. I really, really tried getting this to work, with one hacky workaround after another. Unfortunately, in the current landscape of CLI parsers in Python, it looks like there's no way of doing this while also parsing subcommands declaratively.

However, I believe this is a good breaking change to make. Users rarely have enough horizontal space to fit their entire prompt into a single line, along with the magic command itself and its arguments. This helps keep the first line of each magic short, which also avoids clipping the cell toolbar in JupyterLab. This also makes building the parsers easier, since it allows us to have two separate parsers: one for the cell magic and one for the line magic.

[JasonWeill]

Please modify the user docs to reflect the breaking change. You might need to modify (and check in changes to) the example workbooks, too.

[dlqqq]

I've updated the documentation to reflect this change. 👍