Julia syntax: calling a function with kwargs

[JohnAAbbott]

What is the syntax recommendation for calling a function with kwargs?
I was somewhat surprised to find the following example on the julialang website

julia> foo(;kwargs...) = dump(kwargs)
foo (generic function with 1 method)

julia> foo(a=1, b=2, c=3)
Base.Pairs{Symbol, Int64, Tuple{Symbol, Symbol, Symbol}, NamedTuple{(:a, :b, :c), Tuple{Int64, Int64, Int64}}}
  data: NamedTuple{(:a, :b, :c), Tuple{Int64, Int64, Int64}}
    a: Int64 1
    b: Int64 2
    c: Int64 3
  itr: Tuple{Symbol, Symbol, Symbol}
    1: Symbol a
    2: Symbol b
    3: Symbol c

Why is the call not foo(;a=1, b=2, c=3)? It seems that putting in the semicolon produces the same result as without.
Is there a recommendation for which syntax to use in Oscar code?
[This question was triggered by line 585 in mpoly-affine-algebras.jl where a kwarg is specified without a preceding semicolon]

[thofma]

It is usually a good idea to separate the keyword arguments with a semicolon, since otherwise one might not always get what one wants:

julia> sum(i for i in 1:2, init = 5)
3

julia> sum(i for i in 1:2; init = 5)
8

[JohnAAbbott]

Ouch! That's a subtle difference. I wonder what the first call does...? I tried @which but that was not so helpful.
Anyway I'll correct line 585 in mpoly-affine-algebras.jl (and maybe think of a regexp to find other instances where a comma was used instead of a semicolon).

[JohnAAbbott]

In light of the comment by @thofma it seems to be a really good idea to ensure that we use a semicolon before specifying kwargs, otherwise we will have code which "looks correct" but might not do what one reasonably expects (i.e. somewhat evil bugs).
A fairly simple regexp for finding source code lines which are suspicious is ([^);]*= (at least that seems to work in Emacs), but this probably does not work in all cases... Any suggestions for a better regexp?

[JohnAAbbott]

Here is another (Emacs) regexp: it has more false positives, but fewer false negatives. The regexp ([^;^J]*[_a-zA-Z]+ *=
It looks for an open bracket not followed by a semicolon, but with something looking like identifier = ...

[JohnAAbbott]

In PR #2614 @lgoettgens points out that when using a semicolon, one can use short-cut assignments for kwargs (if kwarg-name and assigned value are written identically). This makes the source code shorter... not sure if it makes the code clearer though.. Advice???
For example:

return ideal(homogenization(gens(groebner_basis(I; ordering=ordering)), var; pos=pos))

could be more succinctly written as

return ideal(homogenization(gens(groebner_basis(I; ordering)), var; pos))

[but then the kwargs do not stand out to the "casual reader"]

[thofma]

I think this is subjective. I personally don't find it clearer at all.

[JohnAAbbott]

Since no-one has expressed a strong opinion either way, I'll go with what I prefer (in my code), namely pos=pos etc.
If we later decide that the shorter form is better, it should be easy to automatically convert pos=pos into just pos.