[Python-ideas] Re: Add __match_args__ to complex and slice

March 28, 2022

      Some background: the reason we decided not to do this is because cases are allowed to omit positional sub-patterns when matching. For example, it would be possible to write this:

```
...
...
...
match n:
...     case complex(42j):
...         ...
...

Read that carefully! It would match any complex number with a *real* part equal to 42j. Which is to say, it would actually match *nothing*.

I personally prefer the (admittedly wordier) status quo:

...
...
...
match n:
...     case complex(real=r, imag=i):
...         ...
...

The problem with slice is much bigger, though: the expected attributes actually change depending on how many "args" there are, so a static mapping just isn't possible. Again, the status quo is much more explicit, for better or worse:

...
...
...
match i:
...     case slice(start=lo, stop=hi, step=by):
...         pass
...

At one point, the original pattern matching proposal (and implementation) included support for a `__match_args_required__` attribute that could optionally be provided alongside `__match_args__`. With that, you could then set `complex.__match_args__ = ("real", "imag")` and `complex.__match_args_required__ = 2`, and get an error like:

...
...
...
match n:
...     case complex(42j):
...        ...
... 
TypeError: complex() requires 2 positional sub-patterns (1 given)

Similarly, you could set `slice.__match_args__ = ("start", "stop", "step")` and `slice.__match_args_required__ = 2`, and get an error like:

...
...
...
match i:
...     case slice(stop):
...        ...
... 
TypeError: slice() requires at least 2 positional sub-patterns (1 given)

We eventually dropped `__match_args_required__` to simplify the proposal, though. For anyone curious, all of our discussions during this period are archived in the issue tracker at: https://github.com/gvanrossum/patma/issues.

Brandt

[Python-ideas] Re: Add __match_args__ to complex and slice

Brandt Bucher