On Wed, Mar 1, 2017 at 11:25 AM, Serhiy Storchaka <storchaka@gmail.com> wrote:
On 28.02.17 23:17, Victor Stinner wrote:
My question is: would it make sense to implement this feature in
Python directly? If yes, what should be the syntax? Use "/" marker?
Use the @positional() decorator?

I'm strongly +1 for supporting positional-only parameters. The main benefit to me is that this allows to declare functions that takes arbitrary keyword arguments like Formatter.format() or MutableMapping.update(). Now we can't use even the "self" parameter and need to use a trick with parsing *args manually. This harms clearness and performance.

Agreed.
 
The problem with the "/" marker is that it looks ugly. There was an excuse for the "*" marker -- it came from omitting the name in "*args". The "*" prefix itself means an iterable unpacking, but "/" is not used neither as prefix nor suffix.

It's in a sense a pun -- * and / are "opposites" in mathematics, and so are the usages here.
 
Do you see concrete cases where it's a deliberate choice to deny
passing arguments as keywords?

dict.__init__(), dict.update(), partial.__new__() and partial.__call__() are obvious examples. There are others.

And there was performance reason. Just making the function supporting keyword arguments added an overhead even to calls with only positional arguments. This was changed recently, but I didn't checked whether some overhead is left.

Don't you like writing int(x="123") instead of int("123")? :-) (I know
that Serhiy Storshake hates the name of the "x" parameter of the int
constructor ;-))

I believe weird names like "x" was added when the support of "base" keyword was added due to the limitation of PyArg_ParseTupleAndKeywords(). All or nothing, either builtin function didn't support keyword arguments, or it supported passing by keyword for all arguments.

But now it is possible to support passing by keyword only the part of parameters. I want to propose to deprecate badly designed keyword names of builtins.

+1
 
By the way, I read that "/" marker is unknown by almost all Python
developers, and [...] syntax should be preferred, but
inspect.signature() doesn't support this syntax. Maybe we should fix
signature() and use [...] format instead?

[...] is not Python syntax too. And it is orthogonal to positional-only parameters. [...] doesn't mean that parameters are positional-only. They can be passed by keyword, but just don't have default value. On other side, mandatory parameters can be positional-only.

FWIW in typeshed we've started using double leading underscore as a convention for positional-only parameters, e.g. here:

https://github.com/python/typeshed/blob/master/stdlib/3/builtins.pyi#L936

FWIW I think 'self' should also be special-cased as positional-only. Nobody wants to write 'C.foo(self=C())'. :-)

--
--Guido van Rossum (python.org/~guido)