[dataclass_transform] support implicit default for "init" parameter in field specifiers #14870
Conversation
…n field specifiers
| if not isinstance(call, CallExpr): | ||
| return True | ||
|
|
||
| specifier_type = _get_callee_type(call) |
There was a problem hiding this comment.
I'm reusing this existing helper from mypy.plugins.common, but unfortunately it doesn't truly support overloads:
if isinstance(callee_node, (Var, SYMBOL_FUNCBASE_TYPES)) and callee_node.type:
callee_node_type = get_proper_type(callee_node.type)
if isinstance(callee_node_type, Overloaded):
# We take the last overload.
return callee_node_type.items[-1]
elif isinstance(callee_node_type, CallableType):
return callee_node_type
I'm not sure if there's an existing method that can resolve the appropriate overload or if some new plumbing will need to be added to MyPy.
There was a problem hiding this comment.
We can't generally resolve overloads in a dataclass plugin, since semantic analysis is not complete yet, and certain type operations aren't strictly speaking available yet. We could perhaps do a best-effort resolution, but I'm not sure if it's worth the effort. Do you know of any use case where this is needed? Perhaps we can come up with a simplified implementation that is good enough for common use cases.
There was a problem hiding this comment.
@JukkaL I'm not sure if anything is using it in the wild, thought he PEP calls it out with an explicit example: https://peps.python.org/pep-0681/#field-specifier-parameters We've discussed this some in #14293 (comment), so I can try asking there if there any concrete use cases to be aware of.
This comment has been minimized.
This comment has been minimized.
|
According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉 |
wesleywright
changed the title
|
I'll try finishing this up so that we can include it in the 1.2 release. |
|
Closing in favor of #15010, which builds on top of this and adds overload resolution. |
…n field specifiers (#15010) (Basic functionality was implemented by @wesleywright in #14870. I added overload resolution.) This note from PEP 681 was missed in the initial implementation of field specifiers: > If unspecified, init defaults to True. Field specifier functions can use overloads that implicitly specify the value of init using a literal bool value type (Literal[False] or Literal[True]). This commit adds support for reading a default from the declared type of the `init` parameter if possible. Otherwise, it continues to use the typical default of `True`. The implementation was non-trivial, since regular overload resolution can't be used in the dataclass plugin, which is applied before type checking. As a workaround, I added a simple overload resolution helper that should be enough to support typical use cases. It doesn't do full overload resolution using types, but it knows about `None`, `Literal[True]` and `Literal[False]` and a few other things. --------- Co-authored-by: Wesley Collin Wright <wesleyw@dropbox.com>
…n field specifiers (#15010) (Basic functionality was implemented by @wesleywright in #14870. I added overload resolution.) This note from PEP 681 was missed in the initial implementation of field specifiers: > If unspecified, init defaults to True. Field specifier functions can use overloads that implicitly specify the value of init using a literal bool value type (Literal[False] or Literal[True]). This commit adds support for reading a default from the declared type of the `init` parameter if possible. Otherwise, it continues to use the typical default of `True`. The implementation was non-trivial, since regular overload resolution can't be used in the dataclass plugin, which is applied before type checking. As a workaround, I added a simple overload resolution helper that should be enough to support typical use cases. It doesn't do full overload resolution using types, but it knows about `None`, `Literal[True]` and `Literal[False]` and a few other things. --------- Co-authored-by: Wesley Collin Wright <wesleyw@dropbox.com>
This note from PEP 681 was missed in the initial implementation of field specifiers:
This commit adds support for reading a default from the declared type of the
initparameter if possible. Otherwise, it continues to use the typical default ofTrue.