[Python-3000] PEP3102 Keyword-Only Arguments

[Guido van Rossum]

Not remembering the PEP in detail, I agree with Jim's resolution of all these.

I guess the right rule is that all positional arguments come first
(first the regular ones, then * or *args). Then come the keyword
arguments, again, first the regular ones (name=value), then **kwds.

I believe the PEP doesn't address the opposite use case: positional
arguments that should *not* be specified as keyword arguments. For
example, I might want to write

  def foo(a, b): ...

but I don't want callers to be able to call it as foo(b=1, a=2) or
even foo(a=2, b=1).

A realistic example is the write() method of file objects. We really
don't want people starting to say f.write(s="abc") because even if
that works for the current file type you're using, it won't work if an
instance of some other class implementing write() is substituted --
write() is always documented as an API taking a positional argument,
so different "compatible" classes are likely to have different
argument names. Currently this is enforced because the default file
type is implemented in C and it doesn't have keyword arguments; but in
Py3k it may well be implemented in Python and then we currently have
no decent way to say "this should really be a positional argument".
(There's an analogy to forcing keyword arguments using **, using *args
for all arguments and parsing that explicitly -- but that's tedious
for a fairly common use case.)

Perhaps we can use ** without following identifier to signal this?
It's not entirely analogous to * without following identifier, but at
least somewhat similar.

--Guido

On 8/12/06, Jim Jewett <jimjjewett at gmail.com> wrote:
> On 8/11/06, Jiwon Seo <seojiwon at gmail.com> wrote:
> > When we have keyword-only arguments, do we allow 'keyword dictionary'
> > argument? If that's the case, where would we want to place
> > keyword-only arguments?
>
> > Are we going to allow any of followings?
>
> > 1. def foo(a, b,  *, key1=None, key2=None, **map)
>
> Seems perfectly reasonable.
>
> I think the controversy was over whether or not to allow keyword-only
> without a default.
>
> > 2. def foo(a, b, *,  **map, key1=None, key2=None)
>
> Seems backward, though I suppose we could adjust if we needed to.
>
> > 3. def foo(a, b, *, **map)
>
> What would the * even mean, since there aren't any named keywords to separate?
>
> -jJ
> _______________________________________________
> Python-3000 mailing list
> Python-3000 at python.org
> http://mail.python.org/mailman/listinfo/python-3000
> Unsubscribe: http://mail.python.org/mailman/options/python-3000/guido%40python.org
>


-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)