Complete subcommands after local flag

[Luap99]

Currently the custom completion does not complete subcommands after a local flag was given.
I think this is wrong since a program can use subcommands even when there were local flags on the root/parent command given.

cobra/custom_completions.go

Lines 283 to 287 in b95db64

	// Complete subcommand names, including the help command
	if len(finalArgs) == 0 && !foundLocalNonPersistentFlag {
	// We only complete sub-commands if:
	// - there are no arguments on the command-line and
	// - there are no local, non-peristent flag on the command-line

[Luap99]

@marckhouzam Is there a special reason for this? Otherwise i would like to change this behavior.

[marckhouzam]

Is there a special reason for this? Otherwise i would like to change this behavior.

Yes this is on purpose for local flags. If a command has a local flags it does not get inherited by a subcommand, so it would be wrong to suggest subcommands as they will fail with the error "unknown flag".

However for persistent flags, which are inherited by subcommands, we do complete subcommands.

Makes sense?

[Luap99]

Yes that makes sence.

However in the podman it works like this:
podman --remote ps -a remote is a local flag on the root cmd here.

I found out that this works because TraverseChildren is set to true on the root cmd so i would say the completion logic should respect TraverseChildren too.

I can open a PR if you agree?

[marckhouzam]

Aha. I'm not familiar with the TraverseChildren logic. You may have found a scenario that was missed. Yes please go ahead and file a PR. Thanks!

[marckhouzam]

@Luap99 could you check if the case you mention works with bash completion?

[Luap99]

@Luap99 could you check if the case you mention works with bash completion?

Yes it does work but regardless of whether TraverseChildren is set or not.

[marckhouzam]

That is strange. I mimicked the local flag behavior on what was done in the current bash completion... See #288

[Luap99]

I also noticed there is bug when you have a local boolean flag there is no subcommand completion but after you provide one argument subcommands are suggested. This doesn't happen with a Persistent Flag.

[Luap99]

I also noticed there is bug when you have a local boolean flag there is no subcommand completion but after you provide one argument subcommands are suggested.

Nevermind my test setup was wrong ;)
It works for local boolean flags, but for local flags with arguments subcommands are still getting completed. So it is inconsistent.

[marckhouzam]

@Luap99 could you check if the case you mention works with bash completion?

Yes it does work but regardless of whether TraverseChildren is set or not.

I've looked further into this and it seems to be a bug with the current bash completion.
When using a boolean flag or the = form for a flag (e.g. integer), sub-commands are not shown after a local flag as expected, but when using the no-= form, it does complete:

bash-5.0$ ./testprog --local 1 [tab][tab]
child1      completion  help
# Sub-commands are completed when they shouldn't

bash-5.0$ ./testprog --local=1 [tab][tab]
[file completion is done]
# Sub-commands are NOT completed as expected

bash-5.0$ ./testprog --bool [tab][tab]
[file completion is done]
# Sub-commands are NOT completed as expected

My guess is that this bug was introduced along the way as the bash completion script evolved.
I will open an issue to keep track of this.

[marckhouzam]

I've opened #1172 for the discrepancy with bash.