fish --no-execute reports "Unknown command" for functions

[paynechu]

I reported here before.
dag/vim-fish#16 (comment)

[KamilaBorowska]

eval is a function, and --no-execute doesn't try to check whatever function exists or not.

~ $ fish --no-execute --help
fish: Unknown command '__fish_print_help'
Standard input: __fish_print_help fish
                ^

(ok, that is a bug)

This ought to be fixed, but in certain cases executing code is needed (some functions, like . aren't defined in files, but in initialization script).

[zanchey]

--no-execute doesn't work in fish 2.0 for me either:

> fish --no-execute echo
open: No such file or directory

[KamilaBorowska]

@zanchey:

Of course it doesn't. You want to check syntax of echo file, but no such file exists (then again, the error message could be less confusing).

~ $ fish --no-execute -c echo
~ $

[zanchey]

Ah, of course.

[lilyball]

This bit me when updating the tests, because test.fish runs fish -n on itself. This meant that I couldn't use type inside the test file.

I'm inclined to say that when -n is set, fish should treat unknown commands as valid.

The real issue here is functions that are loaded from $fish_function_path, as that never ends up getting set, which means autoload can't trigger. Presumably functions provided by a file that's sourced from the config file won't work either, but I haven't tested that.

[krader1961]

Relabeling this from "bug" to "enhancement" because the current behavior seems reasonable and is not in contradiction of the documentation. Whether we can provide a better experience is TBD.

[actionless]

what about having sorta shim loaded when running fish -n

so in that shim abbr and alias would be defined

[ulidtko]

Relabeling this from "bug" to "enhancement" because the current behavior seems reasonable and is not in contradiction of the documentation

Please let me disagree. The current behavior, by far, doesn't seem reasonable.

There can be no valid reason to flag errors when "syntax checking" code which works correctly, runs perfectly fine; this isn't linting.

--no-execute is broken, and should be clearly labelled as such. See #6484.

[actionless]

thanks a lot! works for me now