|
From: | John Snow |
Subject: | Re: [PATCH v5 24/36] qapi/source.py: add type hint annotations |
Date: | Fri, 9 Oct 2020 10:30:30 -0400 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.11.0 |
On 10/8/20 4:42 AM, Markus Armbruster wrote:
John Snow <jsnow@redhat.com> writes:On 10/7/20 7:55 AM, Markus Armbruster wrote:John Snow <jsnow@redhat.com> writes:Annotations do not change runtime behavior. This commit *only* adds annotations. A note on typing of __init__: mypy requires init functions with no parameters to document a return type of None to be considered fully typed. In the case when there are input parameters, None may be omitted. Since __init__ may never return any value, it is preferred to omit the return annotation whenever possible. Signed-off-by: John Snow <jsnow@redhat.com> Reviewed-by: Eduardo Habkost <ehabkost@redhat.com> Reviewed-by: Cleber Rosa <crosa@redhat.com> Tested-by: Cleber Rosa <crosa@redhat.com> --- scripts/qapi/mypy.ini | 5 ----- scripts/qapi/source.py | 31 ++++++++++++++++++------------- 2 files changed, 18 insertions(+), 18 deletions(-) diff --git a/scripts/qapi/mypy.ini b/scripts/qapi/mypy.ini index 8ab9ac52cc4..1b8555dfa39 100644 --- a/scripts/qapi/mypy.ini +++ b/scripts/qapi/mypy.ini @@ -34,11 +34,6 @@ disallow_untyped_defs = False disallow_incomplete_defs = False check_untyped_defs = False -[mypy-qapi.source] -disallow_untyped_defs = False -disallow_incomplete_defs = False -check_untyped_defs = False - [mypy-qapi.types] disallow_untyped_defs = False disallow_incomplete_defs = False diff --git a/scripts/qapi/source.py b/scripts/qapi/source.py index e97b9a8e15e..1cc6a5b82dc 100644 --- a/scripts/qapi/source.py +++ b/scripts/qapi/source.py @@ -11,37 +11,42 @@ import copy import sys +from typing import List, Optional, TypeVarclass QAPISchemaPragma:- def __init__(self): + def __init__(self) -> None: # Are documentation comments required? self.doc_required = False # Whitelist of commands allowed to return a non-dictionary - self.returns_whitelist = [] + self.returns_whitelist: List[str] = [] # Whitelist of entities allowed to violate case conventions - self.name_case_whitelist = [] + self.name_case_whitelist: List[str] = []class QAPISourceInfo:- def __init__(self, fname, line, parent): + T = TypeVar('T', bound='QAPISourceInfo') + + def __init__(self: T, fname: str, line: int, parent: Optional[T]):More ignorant questions (I'm abusing the review process to learn Python type hints)... Why do you need to annotate self here, but not elsewhere?This is admittedly me being a little extra, but I thought it was a good way to show a pattern for people who maybe hadn't been exposed to it yet. This is a pattern that allows for subclassing. I am stating that this __init__ method takes a parent of the same type as itself, whatever that happens to actually be. T is a TypeVar that we can use for Generics. By declaring the type of self as that TypeVar, we bind T to self's type. When we annotate the parent as a T, we are enforcing that the parent we receive is of the same type as ourselves. (Yep, we don't actually subclass QAPISourceInfo and I don't have plans to. It felt slightly icky to hard-code the class type name, though. I try to avoid that whenever I can. I'm not sure I would go so far as to call it a code smell or an antipattern, but it's something I tend to avoid anyway.)Say I define class QSISub as a direct subclass of QAPISourceInfo, and let it inherit __init__(). What's the type of QSISub.__init__()'s parameter @parent? According to my reading of your explanation, it's QSISub. Correct?
That's right.(I'm realizing that this is maybe not a constraint that we should even anticipate here, because maybe we don't wish to say that the parent should always be of the same type. but hey, it led to a good mypy lesson.
I'm going to edit it to do the simpler thing for now and leave well enough alone. There's another chance to see an interesting pattern of TypeVars in the error series in part 4 that I think is actually more explicitly appropriate.)
If we used QAPISourceInfo instead of T for @parent, then it would be QAPISourceInfo. Correct?
Yup! Here's a little sample program that shows what this kind of typing does: ``` from typing import TypeVar, Optional class Example: T = TypeVar('T', bound='Example') def __init__(self: T, parent: Optional[T] = None): self.parent = parent class SubExample(Example): pass x = Example() y = Example(x) z = SubExample() a = Example(x) # OK b = Example(z) # OK c = SubExample(x) # BZZZT d = SubExample(z) # OK ``` If you check this with mypy, you'll get this error: ```test.py:17: error: Argument 1 to "SubExample" has incompatible type "Example"; expected "Optional[SubExample]"
Found 1 error in 1 file (checked 1 source file) ```
Now, perhaps any QAPISourceInfo will do as @parent, perhaps it really needs to be a QSISub. We can't know when we write QAPISourceInfo. But we don't *have* to get this exactly right for all future subclasses, because I can always override __init__() when inheritance doesn't give me the __init__() I want. Correct?
You could, but I suggested on IRC the other day that I am not fully comfortable with the LSP rules that mypy (sometimes?) enforces. I tend not to want to override init with narrower types if I can avoid it, but it is true that we do this quite a lot in the codebase already.
(I believe I have seen mypy throw errors about this on occasion, but I can't pinpoint the exact circumstances in which it does. It's a point of confusion for me.)
Say I override __init__(), and have it call super().__init__(). I have to pass it a QAPISourceInfo @parent. A QSISub will do (it's a subtype). Correct? One more: is bound='QAPISourceInfo' strictly needed?
I'm not sure if bound is strictly required or not. mypy docs just use it outright and don't mention why:
https://mypy.readthedocs.io/en/stable/generics.html#generic-methods-and-generic-self Generally, this constrains a TypeVar to some upper bound, see --https://mypy.readthedocs.io/en/stable/generics.html#type-variables-with-upper-bounds
-- but in this case, we're only using that TypeVar for an init method that only exists for this type and below, so it might be redundant.
I modified my example program to remove the bound and it appears to fail in the same exact way, so it might be pointless in this case. It might have a stronger use if you want to re-use the 'T' typevar elsewhere where it isn't implicitly bound by the 'self' argument. Maybe it has implications for multi-inheritance too, I've not tested it to that level.
Thanks for the good question! --js
[Prev in Thread] | Current Thread | [Next in Thread] |