complete --force-files does not work without options

[rouge8]

fish version: 3.2.2

I'm trying to write completions for magic-wormhole and I'm struggling to complete paths outside of the current directory using __fish_complete_path with this completion file:

complete -e -c wormhole
complete --no-files -c wormhole

complete -c wormhole -l help
complete -c wormhole -l version -n "__fish_no_arguments"

complete -c wormhole -n __fish_use_subcommand -a receive -d "Receive a text message, file, or directory (from 'wormhole send')"

complete -c wormhole -n __fish_use_subcommand -a send -d "Send a text message, file, or directory"
complete -c wormhole -n "__fish_seen_subcommand_from send" -r -a "(__fish_complete_path)"

# This isn't a real option to wormhole but demonstrates the inconsistency
complete -c wormhole -l send -r -a "(__fish_complete_path)"

• If I type, wormhole send <tab> it starts completing files in the current directory, as expected.
• If I type wormhole send ~/<tab>, it doesn't complete anything, but I would expect it to complete the contents of my home directory
• If I type wormhole --send ~/<tab>, it starts completing the contents of my home directory

Is there a way to get wormhole send ~/<tab> to complete the contents of my home directory? I'm also very confused why -a "(__fish_complete_path)" is behaving differently with a long option vs a subcommand.

[faho]

__fish_complete_path completes directories. That is what it does - the brunt of it is command ls -dp, and the -d only lists directories.

What you want is to reenable files, like

complete -c wormhole -n "__fish_seen_subcommand_from send" -r -F

[rouge8]

__fish_complete_path completes directories

This doesn't seem right, I see it returning directories and files:

❯ __fish_complete_path ~/forks/fish-shell/
/Users/andy/forks/fish-shell/BSDmakefile	
/Users/andy/forks/fish-shell/CHANGELOG.rst	
/Users/andy/forks/fish-shell/CMakeLists.txt	
/Users/andy/forks/fish-shell/CODE_OF_CONDUCT.md	
/Users/andy/forks/fish-shell/CONTRIBUTING.rst	
/Users/andy/forks/fish-shell/COPYING	
/Users/andy/forks/fish-shell/Dockerfile	
/Users/andy/forks/fish-shell/GNUmakefile	
/Users/andy/forks/fish-shell/README.rst	
/Users/andy/forks/fish-shell/benchmarks/	
/Users/andy/forks/fish-shell/build/	
/Users/andy/forks/fish-shell/build,/	
/Users/andy/forks/fish-shell/build_tools/	
/Users/andy/forks/fish-shell/cmake/	
/Users/andy/forks/fish-shell/config_cmake.h.in	
/Users/andy/forks/fish-shell/debian/	
/Users/andy/forks/fish-shell/doc_src/	
/Users/andy/forks/fish-shell/docker/	
/Users/andy/forks/fish-shell/etc/	
/Users/andy/forks/fish-shell/fish.desktop	
/Users/andy/forks/fish-shell/fish.pc.in	
/Users/andy/forks/fish-shell/fish.png	
/Users/andy/forks/fish-shell/fish.spec.in	
/Users/andy/forks/fish-shell/osx/	
/Users/andy/forks/fish-shell/pcre2/	
/Users/andy/forks/fish-shell/po/	
/Users/andy/forks/fish-shell/share/	
/Users/andy/forks/fish-shell/src/	
/Users/andy/forks/fish-shell/tests/	

What you want is to reenable files, like

complete -c wormhole -n "__fish_seen_subcommand_from send" -r -F

With this, wormhole send <tab> doesn't complete anything at all. 😕

[faho]

Oh, sorry, no, I misunderstood ls man page. What it means is that, for directories given as arguments it won't descend.

So ls ~/ will list the contents of your HOME, ls -d ~/ will print $HOME.

What __fish_complete_path does is complete its first argument one level. So you'll have to do -a '(__fish_complete_path (commandline -ct))'

This is an awkward interface, but it's also unnecessary given that -F exists.

Unfortunately that has a bug and only works with options (so if -s or -l or -o) is given. Fix incoming.

[rouge8]

it's also unnecessary given that -F exists.

Unfortunately that has a bug and only works with options (so if -s or -l or -o) is given. Fix incoming.

Ah thanks that would explain that! I'm still confused by the behavior of __fish_complete_path with a subcommand vs an option though.

What __fish_complete_path does is complete its first argument one level. So you'll have to do -a '(__fish_complete_path (commandline -ct))'

This still doesn't complete anything for wormhole send ~/<tab>, but does for wormhole --send ~/<tab>:

complete -e -c wormhole
complete --no-files -c wormhole

complete -c wormhole -n __fish_use_subcommand -a send -d "Send a text message, file, or directory"
complete -c wormhole -n "__fish_seen_subcommand_from send" -r -a '(__fish_complete_path (commandline -ct))'

# This isn't a real option to wormhole but demonstrates the inconsistency
complete -c wormhole -l send -r -a '(__fish_complete_path (commandline -ct))'

e.g.:

❯ complete -C 'wormhole send ~/forks/fish-shell/'

vs.

❯ complete -C 'wormhole --send ~/forks/fish-shell/'
~/forks/fish-shell/benchmarks/
~/forks/fish-shell/BSDmakefile
~/forks/fish-shell/build,/
~/forks/fish-shell/build/
~/forks/fish-shell/build_tools/
~/forks/fish-shell/CHANGELOG.rst
~/forks/fish-shell/cmake/
~/forks/fish-shell/CMakeLists.txt
~/forks/fish-shell/CODE_OF_CONDUCT.md
~/forks/fish-shell/config_cmake.h.in
~/forks/fish-shell/CONTRIBUTING.rst
~/forks/fish-shell/COPYING
~/forks/fish-shell/debian/
~/forks/fish-shell/doc_src/
~/forks/fish-shell/docker/
~/forks/fish-shell/Dockerfile
~/forks/fish-shell/etc/
~/forks/fish-shell/fish.desktop
~/forks/fish-shell/fish.pc.in
~/forks/fish-shell/fish.png
~/forks/fish-shell/fish.spec.in
~/forks/fish-shell/GNUmakefile
~/forks/fish-shell/osx/
~/forks/fish-shell/pcre2/
~/forks/fish-shell/po/
~/forks/fish-shell/README.rst
~/forks/fish-shell/share/
~/forks/fish-shell/src/
~/forks/fish-shell/tests/

[faho]

This still doesn't complete anything for wormhole send ~/, but does for wormhole --send ~/:

Yes, this is because the ~ is taken literally instead of expanded. It's a general issue with scripted completions - things like ~ and $VARIABLE won't be expanded, and there is currently no safe way to do that - #5811 would help.

Anyway, I've just pushed a fix for that pesky --force-files issue, that's what you should be using in this case.

[rouge8]

Yes, this is because the ~ is taken literally instead of expanded. It's a general issue with scripted completions - things like ~ and $VARIABLE won't be expanded, and there is currently no safe way to do that - #5811 would help.

Ahhhhh, got it.

Anyway, I've just pushed a fix for that pesky --force-files issue, that's what you should be using in this case.

Thanks for the fast fix!