Teach argcomplete that command subcommand calls command-subcommand about argcomplete HOT 10 CLOSED

There isn't currently a built in way to do this.

Could you provide an example of how you implement this in the argparse configuration? Do you use parse_known_args() and examine the remainder?

You could write a completer that calls command-subcommand while passing through the appropriate environment variables and args, then returns the result, but I'm not sure if that's the nicest way.

Another possibility would be to have the Python portion of argcomplete exit with an exit code that specifies "call the supplied executable name for completions instead", and make the shellcode portion understand that.

from argcomplete.

No, there is no argparsing. All it does is if the subcommand is now known, it looks for command-subcommand on the PATH and calls out to that instead in a subprocess (with appropriately changed sys.argv).

from argcomplete.

Oh maybe that isn't entirely true. It uses a custom subclass of ArgumentParser with an overridden error method: https://github.com/conda/conda/blob/master/conda/cli/conda_argparse.py#L35.

from argcomplete.

Use of the argparse error handler in this way precludes compatibility with argcomplete (at least for the purpose of returning these completions), because the Argparse class is being extended in a way incompatible with how argcomplete extends it.

I recommend enumerating the command-subcommand combinations as actual argparse actions, instead of overriding the error mechanism.

from argcomplete.

I'll take a look. Is it possible to not search the PATH unless a command is given that isn't found with that way?

from argcomplete.

#!/usr/bin/env python3
# PYTHON_ARGCOMPLETE_OK

import argparse, argcomplete, subprocess

def complete_subparser_action(prefix, **kwargs):
    internal_actions = ['foo', 'bar']
    if prefix and any(action.startswith(prefix) for action in internal_actions):
        return internal_actions
    else:
        external_actions = subprocess.check_output("compgen -c "+prefix, shell=True, executable='/bin/bash')
        external_actions = [a.decode() for a in external_actions.splitlines()]
        return internal_actions + external_actions

parser = argparse.ArgumentParser()
subparsers = parser.add_subparsers()
subparsers.completer = complete_subparser_action
subparsers.add_parser('foo')
subparsers.add_parser('bar')
argcomplete.autocomplete(parser)
args = parser.parse_args()

print(args)

from argcomplete.

Thanks, I'll look at doing something like that.

By the way, you should probably avoid using shell=True, as it can introduce security vulnerabilities via shell injection (although I couldn't figure out how to actually do that in this case).

from argcomplete.

This is sample code. I have to use shell=True, because otherwise compgen is not available. And yes, you should quote and escape it.

from argcomplete.

I have a similar issue, I'm using calling out subshells to do autocompletion: https://github.com/openaps/openaps/blob/master/bin/openaps

However, after the subprocess has done it's completion, the original wrapper is adding the top-level completion opts again, (after the last nargs=argparse.REMAINDER in the top level shell). Does anyone know how to prevent that from happening?

from argcomplete.

I ended up with the following implementation, which subclasses argcomplete.CompletionFinder and overrides __call__ conda/conda#1443. I also implement calling it automatically inside of a custom subclass of ArgumentParser, so that extensions only have to import and use that custom subclass to get completion.

from argcomplete.