Kongplete predictors and enums

[shlomi-dr]

Hi Alec,

I am using an enum to specify what values should be allowed, and just started using Kongplete with custom predictions. Currently I repeat the enum values in the predictor's set, which is a bit of a chore, as they could get out of sync unintentionally.

I was thinking - could we tie these together? Instead of hard-coding the enum values, perhaps we could let them get picked up from the predictor? Something like:

var predictors = map[string]complete.Predictor{
  "vals":  complete.PredictSet("enum1", "enum2"),
}

type CLI struct {
   Val string `name:"val" enum:"vals" predictor:"vals"`
}

func main() {
   parser := kong.Must(&CLI, kong.WithEnumFromPredictors(predictors))
...
}

In any case, it would be convenient to provide the enum values in a similar way to the predictors, where we can programmatically set the values of the enum. Something like:

var enums = map[string][]strings{
  "vals": []string{"enum1", "enum2"},
}

type CLI struct {
   Val string `name:"val" enum:"vals"`
}

func main() {
   parser := kong.Must(&CLI, kong.WithEnums(enums))
...
}

Thank you!

[alecthomas]

I think it would be preferable if kongplete picked up the enum from Kong.

[shlomi-dr]

Here is a little method that connects kongplete with enums:

// pickEnumsFromPredictors converts kongplete's predictors map to Kong.Vars for
// use as enums. If no `specify` is provided, all predictors are used (including
// ones that probably shouldn't be enums, like file completions etc). Otherwise,
// only specified fields will be picked.
func pickEnumsFromPredictors(p map[string]complete.Predictor, specify ...string) kong.Vars {
   v := kong.Vars{}
   if len(specify) > 0 {
     for _, key := range specify {
       predictor, ok := p[key]
         if ok {
           v[key] = strings.Join(predictor.Predict(complete.Args{}), ",")
         }
     }
   } else {
     for s, predictor := range p {
       v[s] = strings.Join(predictor.Predict(complete.Args{}), ",")
     }
   }
   return v
}


var predictors = map[string]complete.Predictor{
  "vals":  complete.PredictSet("enum1", "enum2"),
}

var kongVars = PickEnumsFromPredictors(predictors)

type CLI struct {
   Val string `name:"val" enum:"${vals}" help:"allowed ${vals}" predictor:"vals"`
}

func main() {
   parser := kong.Must(&CLI, kongVars)
   kongplete.Complete(parser, kongplete.WithPredictors(predictors))
...
}

And with this, we can define the enums only once! Of course, this could also be reversed. Just a first attempt :)

[alecthomas]

But as a workaround you could use kong.Vars{} to interpolate vars into the Kong tags.

[shlomi-dr]

Ah! Again, I missed the obvious solution, which was right there on the Readme (which I read multiple times 🤦‍♂️) - to use kong.Vars{}.

Thank you for this, I think given that I can wire the kongplete's predictors to the enum myself using it. (see #314 (reply in thread) for example)

Here is the snippet from above, adjusted to the proper solution:

var kongVars = kong.Vars{
  "vals": "enum1,enum2",
}

type CLI struct {
   Val string `name:"val" enum:"${vals}" help:"allowed ${vals}"`
}

func main() {
   parser := kong.Must(&CLI, kongVars)
...
}

Thanks again for the quick response, this is awesome!