Support additional args to namedtuple() #5215
Conversation
| @@ -1,7 +1,7 @@ | |||
| [case testNamedTupleUsedAsTuple] | |||
| from collections import namedtuple | |||
|
|
|||
| X = namedtuple('X', ['x', 'y']) | |||
| X = namedtuple('X', 'x y') | |||
There was a problem hiding this comment.
My change ends up requiring list to exist in all of these tests. It was easier to just change the namedtuple call than to add [builtins fixtures/list.pyi] everywhere. I left a few with the list to make sure we test that code path.
test-data/unit/check-namedtuple.test
Outdated
|
|
||
| [builtins fixtures/bool.pyi] | ||
|
|
||
| [case testNamedTupleDefaults-skip] |
There was a problem hiding this comment.
will work on enabling this next (#4788)
| @@ -13,14 +13,6 @@ s = b'foo' | |||
| from typing import TypeVar | |||
| T = TypeVar(u'T') | |||
|
|
|||
| [case testNamedTupleUnicode] | |||
There was a problem hiding this comment.
This no longer worked because my stub for namedtuple just has str. Apparently we don't support Text in the test stubs, so I just moved this test case to the pythoneval tests.
| @@ -27,3 +27,4 @@ class int: | |||
| class str: pass | |||
| class bool: pass | |||
| class function: pass | |||
| class ellipsis: pass | |||
There was a problem hiding this comment.
This was needed because a few tests that import collections didn't have ... available in their fixtures, and my stub for namedtuple uses it.
| @@ -138,10 +138,6 @@ MypyFile:1( | |||
| from collections import namedtuple | |||
| N = namedtuple('N') # E: Too few arguments for namedtuple() | |||
|
|
|||
| [case testNamedTupleWithTooManyArguments] | |||
There was a problem hiding this comment.
This test stopped working because this error is now caught during type checking rather than semanal.
mypy/semanal_namedtuple.py
Outdated
| return items, types, num_defaults, ok | ||
|
|
||
| def extract_defaults_from_arg(self, call: CallExpr, arg: Expression) -> int: | ||
| if not isinstance(arg, (ListExpr, TupleExpr)): |
There was a problem hiding this comment.
Technically the defaults argument can be any iterable, but it doesn't seem particularly useful to support that.
|
Did anyone else review this yet? It seems oddly stalled (I wonder what happened around June 13, a number of PRs saw their last activity around that date). |
|
No. As far as I can see, it's ready for review. |
|
I can review it later today. |
|
Thanks Ivan!
|
mypy/checkexpr.py
Outdated
| @@ -200,8 +200,14 @@ def analyze_var_ref(self, var: Var, context: Context) -> Type: | |||
| def visit_call_expr(self, e: CallExpr, allow_none_return: bool = False) -> Type: | |||
| """Type check a call expression.""" | |||
| if e.analyzed: | |||
| if isinstance(e.analyzed, NamedTupleExpr) and not e.analyzed.is_typed: | |||
| # Type check the arguments, but ignore the results. | |||
There was a problem hiding this comment.
Could you please add a comment here that generating correct errors in call here depends on the typeshed stub?
mypy/semanal_namedtuple.py
Outdated
| info = self.build_namedtuple_typeinfo(name, items, types, {}) | ||
| if num_defaults > 0: | ||
| default_items = { | ||
| arg_name: EllipsisExpr() |
There was a problem hiding this comment.
This is a bit hacky so at least deserves a comment why we use EllipsisExpr() instead of using actual expressions.
If possible, I would just use the actual expressions here.
There was a problem hiding this comment.
I think I can get the actual expressions, but I'm not sure I understand why that would be better. It would make the code more complicated and I don't see much of an advantage.
There was a problem hiding this comment.
If this is really complex then it maybe doesn't make sense. In general now that we work on mypyc it is simpler if we keep all real info somewhere in the AST.
There was a problem hiding this comment.
Oh yes, this makes sense in the context of mypyc. It isn't too bad actually, I'll just make the change.
mypy/semanal_namedtuple.py
Outdated
| if call.arg_kinds != [ARG_POS, ARG_POS]: | ||
| # Typed namedtuple doesn't support additional arguments. | ||
| if fullname == 'typing.NamedTuple': | ||
| return self.fail_namedtuple_arg("Too many arguments for namedtuple()", call) |
There was a problem hiding this comment.
I would capitalize namedtuple -> NamedTuple due to the proposed change.
test-data/unit/check-namedtuple.test
Outdated
| A = namedtuple('A', 'a b') | ||
| B = namedtuple('B', 'a b', rename=1) | ||
| C = namedtuple('C', 'a b', rename='not a bool') # E: Argument "rename" to "namedtuple" has incompatible type "str"; expected "int" | ||
| # This works correctly but also produces "mypy/test-data/unit/lib-stub/collections.pyi:3: note: "namedtuple" defined here" |
There was a problem hiding this comment.
You can add an [out] section at the end that lists all error verbatim instead of using # E:
| typename: str, | ||
| field_names: Union[str, Iterable[str]], | ||
| *, | ||
| # really int but many tests don't have bool available |
There was a problem hiding this comment.
I think this should read "really bool".
|
Thanks for the review! I will push an update soon. |
|
Will fix the CI failure later today. |
mypy/semanal_namedtuple.py
Outdated
| if len(defaults) > 0: | ||
| default_items = { | ||
| # We don't have the actual argument expressions any more, so we | ||
| # just use Ellipsis for all defaults. |
There was a problem hiding this comment.
This comment is now outdated.
|
Thanks for catching that! I'll close the other ones too. |
By letting typeshed do the work.
Fixes #2124, #2127, #4788.