Hi Victor, and thanks for you reply.

On 9 Feb 2021, at 11:46, Victor Stinner wrote: If there are applications relying on the parameter name, it's better to trust the Python implementation, rather than the documentation.

I feared that answer :)

It would be better to make these parameters positional only in the first place, but now it's too late for such backward incompatible change :-(

Unfortunately, yes. Erlend

09.02.21 12:22, Erlend Aasland пише:

What's the recommended approach with issues like https://bugs.python.org/issue43094? Change the docs or the implementation? I did a quick search on bpo, but could not find similar past issues.

If the documentation and the C implemented function contradict about parameter name, we are free to treat the parameter as positional-only. User cannot pass the argument as keyword because the documented name does not work, and the real name is not exposed to the user. In this case, there are two similar functions, create_function() and create_aggregate(). Both are documented as having parameter "num_params", but real names are different: "narg" and "n_arg". It would be confusing to have different spelling of the same parameter in similar functions. I am sure that it would be difficult to remember how is it spelled in every case. Also, in Python terminology, the semantic of this name is the number of parameters, not the number of argument. So I think that in this particular case the chance of breaking some code is insignificant, but possible long-term harm of exposing bad parameter names may be significant.

On 2/9/21 9:15 PM, Serhiy Storchaka wrote:

09.02.21 12:22, Erlend Aasland пише:

...

What's the recommended approach with issues like https://bugs.python.org/issue43094? Change the docs or the implementation? I did a quick search on bpo, but could not find similar past issues.

If the documentation and the C implemented function contradict about parameter name, we are free to treat the parameter as positional-only. User cannot pass the argument as keyword because the documented name does not work, and the real name is not exposed to the user.

It is. Python will correct you if you try to use the documented name:

...

...

import sqlite3 connection = sqlite3.connect(':memory:') connection.create_function('testfunc', num_params=1, func=lambda arg: None) Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: function missing required argument 'narg' (pos 2)

On 10 Feb 2021, at 01:28, Petr Viktorin wrote:

On 2/9/21 9:15 PM, Serhiy Storchaka wrote:

...

...

What's the recommended approach with issues like https://bugs.python.org/issue43094? Change the docs or the implementation? I did a quick search on bpo, but could not find similar past issues. If the documentation and the C implemented function contradict about

09.02.21 12:22, Erlend Aasland пише: parameter name, we are free to treat the parameter as positional-only. User cannot pass the argument as keyword because the documented name does not work, and the real name is not exposed to the user.

It is. Python will correct you if you try to use the documented name:

Correct. However, I suspect that most users use positional arguments with these functions (since this is the first (?) report of the inconsistency) so I guess it's ok, in this case, to make the arguments positional only and normalise the argument naming. Erlend