Returning zero completions leads to completer being called three times

[r4victor]

We're trying to use argcomplete to complete resource names. If we return an empty list, this causes completer to being called three times (tested on zsh). This is a problem since calling the completer involves a network request. Minimal example:

#!/usr/bin/env python
import argparse
import argcomplete
from argcomplete.completers import BaseCompleter


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("name").completer = NameCompleter()

    argcomplete.autocomplete(parser, always_complete_options=False)


class NameCompleter(BaseCompleter):

    def __call__(self, **kwargs):
        argcomplete.warn("Called completer")
        return []


main()

Then:

✗ eval "$(register-python-argcomplete clicomp.py)"
✗ ./clicomp.py <tab>
Called completer

Called completer

Called completer

I haven't found a logic in argcomplete that causes the repeated calls. Do you know if it's shell-related? What should a completer return in case of no completions to avoid repeated invocations? Thanks.

[kislyuk]

Thank you for reporting this @r4victor. I agree the observed behavior is a concern and should not be expected.

@evanunderscore I think this is a side effect of #467 Never mind, something was wrong with my environment. I'm not sure what exactly causes this yet.

[r4victor]

@kislyuk, thanks for looking into this!

We were testing it in fresh zsh environments and found that it only occurs with oh-my-zsh (dstackai/dstack#2285).

It'd be great if it's possible to fix this on the argcomplete side.

[solovyevt]

Great project, @kislyuk!

A more specific context for this issue: dstackai/dstack#2285 (comment)