preparser + match-case = sadness

[yyyyx4]

sage: x = 2
sage: match x:
....:     case 1: print('hello')
....:     case 2: print('world')
....:
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [2], line 2
      1 match x:
----> 2     case Integer(1): print('hello')
      3     case Integer(2): print('world')

TypeError: sage.rings.integer.Integer() accepts 0 positional sub-patterns (1 given)

Component: python3

Issue created by migration from https://trac.sagemath.org/ticket/34678

[jhpalmieri]

comment:1

As a workaround, you can use case 1r: .... I agree that the preparser should be fixed to handle this situation, though.

[mkoeppe]

comment:2

I think this should be solved by defining __match_args__ for various of our classes, see https://peps.python.org/pep-0634/#class-patterns

[jhpalmieri]

comment:3

Something like this?

    __match_args__ = ("__match_self_prop__",)

    @property
    def __match_self_prop__(self):
        return self

(taken from that link)

That seems to work for this case, at least.

[jhpalmieri]

comment:4

How broadly could this be done? For SageObject? Element?

[jhpalmieri]

comment:5

One option is to modify Element, but I don't know if that's too broad:

diff --git a/src/sage/structure/element.pyx b/src/sage/structure/element.pyx
index b5d83ef71b..11981b1b7c 100644
--- a/src/sage/structure/element.pyx
+++ b/src/sage/structure/element.pyx
@@ -397,6 +397,25 @@ cdef class Element(SageObject):
         """
         self._parent = parent
 
+    # __match_args__ and __match_self_prop__ are to handle Python's
+    # match-case construction. See :trac:`34678` and
+    # https://peps.python.org/pep-0634/#class-patterns.
+    __match_args__ = ("__match_self_prop__",)
+
+    @property
+    def __match_self_prop__(self):
+        """
+        TESTS::
+
+            sage: x = 2
+            sage: match x:
+            ....:     case 1: print('hello')
+            ....:     case 2: print('world')
+            ....:
+            world
+        """
+        return self
+
     def _set_parent(self, parent):
         r"""
         Set the parent of ``self`` to ``parent``.

[jhpalmieri]

comment:6

The above patch doesn't work, and I think that I really don't understand the match-case construction. With the above patch:

┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 9.8.beta3, Release Date: 2022-10-30               │
│ Using Python 3.10.8. Type "help()" for help.                       │
└────────────────────────────────────────────────────────────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Warning: this is a prerelease version, and it may be unstable.     ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
sage: x = 3
sage: type(c)
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
Input In [2], in <cell line: 1>()
----> 1 type(c)

NameError: name 'c' is not defined
sage: match x:
....:     case 'b': print('hello')
....:     case c: print('world')
....: 
world
sage: c
3
sage: type(c)
<class 'sage.rings.integer.Integer'>

Why doesn't the match-case block raise an error, since c is not defined? How does c become defined?

[jhpalmieri]

comment:7

In fact I get this behavior in plain Python. I guess this is binding the undefined symbol c to x and proceeding from there, so this is not a problem with the above patch. However, it still doesn't work:

sage: x = 3
sage: match x:
....:     case 1: print('hello')
....:     case 3.2: print('world')
....: 
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Input In [21], in <cell line: 1>()
      1 match x:
      2     case Integer(1): print('hello')
----> 3     case RealNumber('3.2'): print('world')

TypeError: called match pattern must be a type

It accepts Sage Integers but not RealNumbers.

[mkoeppe]

comment:8

c becomes defined as a parameter in the match pattern.

[mkoeppe]

comment:9

The difference is that Integer is a class, but RealNumber is only a function.

[mkoeppe]

comment:10

I think we would be able to handle RealNumber by changing it to a class with a __classcall__ method.

[jhpalmieri]

comment:11

That may be more of a job than I have time for right now.