Fix check_default_type_multi and check_default_type_multi2

[kshramt]

No description provided.

[carlobaldassi]

I'm looking into that, but I suspect that in check_default_type_multi it is the error message which is wrong. I'm almost sure check_default_type_multi2 is correct. Can you provide a use case of what you have tried?

[kshramt]

1. arg_type <: eltype(default) in check_default_type_multi is inconsistent with typeof(default) <: arg_type in check_default_type.
2. An use case is:

using ArgParse

settings = ArgParseSettings()
@add_arg_table(settings, "arg", action=>:store_arg, nargs=>2, default=>["p", "q"]) # error

[carlobaldassi]

Ok, I pushed a change which makes the use case work. Thanks for reporting. The thing which still makes me a little uncomfortable is that now you can have a default such as ["p", "q"] which is a Vector{ASCIIString}, but if you pass p and q as arguments to your program you'll get {"p", "q"}, i.e. a Vector{Any}. I was thinking of issuing a warning in these situations, but maybe that'd be nagging. Comments welcome.

About the difference between check_default_type and check_default_type_multi: it's not inconsistent, and the reason is that there's a difference between the eltype of a Vector and the type of each of its elements. The previous version was checking that arg_type <: eltype(default) as a way to ensure that pushing something of type arg_type to the vector was allowed; this was indended for the multi-action case (e.g. action = :append_arg) and I therefore relaxed it for the multi-nargs case; notice that in both cases it is still checked that isa(x, arg_type) holds for each element x of the vector, which corresponds to the test in check_default_type.
(Same argument for check_default_type_multi2.)

[kshramt]

The thing which still makes me a little uncomfortable is that now you can have a default such as ["p", "q"] which is a Vector{ASCIIString}, but if you pass p and q as arguments to your program you'll get {"p", "q"}, i.e. a Vector{Any}.

I am not sure if this behavior causes any problems.

The previous version was checking that arg_type <: eltype(default) as a way to ensure that pushing something of type arg_type to the vector was allowed; this was indended for the multi-action case (e.g. action = :append_arg) and I therefore relaxed it for the multi-nargs case

I understand. Thank you for your explanations and modifications!