Cannot specify two `file` args

[halms]

Upfront note: I'm not sure if this issue is on the mise or the usage side, but I'm leaning usage.

I have the following mise task in a file:

#!/usr/bin/env python
#MISE description="Run a comparison of two files"
#USAGE arg "<file>"
#USAGE arg "<file>"

import sys
import pandas as pd

df1 = pd.read_parquet(sys.argv[1])
df2 = pd.read_parquet(sys.argv[2])

I want to be able to autocomplete two files when typing the file.
With the above script this works, but I can then never run the task because it always results in an error:

> mise r compare
[compare] ERROR Missing required arg: <file>
Missing required arg: <file>
[compare] ERROR task failed

> mise r compare file1
[compare] ERROR Missing required arg: <file>
[compare] ERROR task failed

> mise r compare file1 file2
[compare] ERROR Missing required arg: <file>
[compare] ERROR task failed

If I try to run with no args, I get the error message twice.
But no matter if I run with 1, 2, 3, 5 args, it always shows one error.

If I do

#!/usr/bin/env python
#MISE description="Run a comparison of two files"
#USAGE arg "<file1>"
#USAGE arg "<file2>"

or any other names for the args, I don't get file completion

[jdx]

they need to be uniquely named

[halms]

Yes, I was thinking so.

But it seems I cannot have them complete as files when I name them anything other than <file>.

The other option would be:

#USAGE arg "<file>" var=#true var_min=2 var_max=2

But that also doesn't complete.

So maybe my real issue is:
How to specify two arguments that both complete as files.

[jdx]

I've noticed that default completion behavior for shells seems to fall back to files but with usage the default shows nothing. I think usage should be patched to fall back to file completions since that seems to not be working.

To answer your actual question, idk off the top of my head but I thought there was some support for defining the completion type in usage.

[halms]

Yes, I agree, falling back to file completion (or basically the shell's default completion) if nothing else is specified sounds like a good option.
This would for many cases also lighten the need for explicit specs for simple scripts.
Currently it's just that if I do mise r task1 (where task1 doesn't have any args specified) and want to pass a file argument, I lose all completion otherwise provided by the shell and have to type/paste everything by hand.

I thought there was some support for defining the completion type in usage
My first thought was also that maybe something like

#USAGE arg "<file1>" complete_as=file
#USAGE arg "<file2>" complete_as=file

would be nice.

After digging into the code, I found that this works

#USAGE arg "<file1>"
#USAGE arg "<file2>"
#USAGE complete file1 type=file
#USAGE complete file2 type=file

or shorter

#USAGE arg "<file1>"; complete file1 type=file
#USAGE arg "<file2>"; complete file2 type=file

(but this isn't really clear from the docs)

Besides falling back to default shell completion, I see three possible ways for improvement (not mutually exclusive):

• Document the <file>, complete type= behaviors clearer
• Allow some shorthand to define the completion type in the arg spec
• Maybe modify the completion behavior of variable args (though I understand that is a harder problem)

[jdx]

#296