Re: [PATCH v2 3/6] qapi/parser: add type hint annotations (QAPIDoc)

[Markus Armbruster]

John Snow <jsnow@redhat.com> writes:

> On Tue, Sep 7, 2021 at 6:44 AM Markus Armbruster <armbru@redhat.com> wrote:
>
>> John Snow <jsnow@redhat.com> writes:
>>
>> > Annotations do not change runtime behavior.
>> >
>> > This commit adds mostly annotations, but uses a TYPE_CHECKING runtime
>> > check to conditionally import dependencies, which only triggers during
>> > runs of mypy.
>>
>> Please add a reference to
>> https://mypy.readthedocs.io/en/latest/runtime_troubles.html#import-cycles
>>
>>
> OK.
>
>
>> > Signed-off-by: John Snow <jsnow@redhat.com>
>> >
>> > ---
>> >
>> > TopLevelExpr, an idea from previous drafts, makes a return here in order
>> > to give a semantic meaning to check_expr(). The type is intended to be
>> > used more in forthcoming commits (pt5c), but I opted to include it now
>> > instead of creating yet-another Dict[str, object] type hint that I would
>> > forget to change later.
>> >
>> > Signed-off-by: John Snow <jsnow@redhat.com>
>> > ---
>> >  scripts/qapi/parser.py | 77 ++++++++++++++++++++++++++----------------
>> >  1 file changed, 48 insertions(+), 29 deletions(-)
>> >
>> > diff --git a/scripts/qapi/parser.py b/scripts/qapi/parser.py
>> > index 3ddde318376..b1e2fa5c577 100644
>> > --- a/scripts/qapi/parser.py
>> > +++ b/scripts/qapi/parser.py
>> > @@ -18,6 +18,7 @@
>> >  import os
>> >  import re
>> >  from typing import (
>> > +    TYPE_CHECKING,
>> >      Dict,
>> >      List,
>> >      Optional,
>> > @@ -30,6 +31,15 @@
>> >  from .source import QAPISourceInfo
>> >
>> >
>> > +if TYPE_CHECKING:
>> > +    # pylint: disable=cyclic-import
>> > +    # TODO: Remove cycle. [schema -> expr -> parser -> schema]
>>
>
> WRT this todo: you mentioned that you'd prefer having some idea or plan for
> how to eliminate the cycle in order to let this band-aid fly. How about
> adding a 'doc' member to e.g. QAPISchemaFeature and doing the connection
> entirely inside of schema.py, and dropping connect_member() and
> connect_feature()?
>
> Would that be serviceable?

I guess it would.  One way to find out.

>> > +    from .schema import QAPISchemaFeature, QAPISchemaMember
>> > +
>> > +
>> > +#: Represents a single Top Level QAPI schema expression.
>> > +TopLevelExpr = Dict[str, object]
>>
>> Related: _ExprValue below, and _JSONObject in expr.py.  The latter's
>> comment gives the best rationale (except I don't get "the purpose of
>> this module is to interrogate that type").
>>
>>
> in expr.py, the purpose of that module (expr) is explicitly to interrogate
> (check, validate) the shape of arbitrary JSON objects. I am saying that a
> more strict definition specifically here in expr.py is not necessary
> because the entire purpose of expr.py is to, at runtime, verify the shape
> of any such variables that might be annotated that way. I am drawing some
> distinction to introspect.py, where we're generating that data ourselves --
> the stronger types are more viable there, because we know what they are
> already.

Let me try to write a clearer comment:

    # Deserialized JSON objects as returned by the parser.
    # This is a actually Dict[str, _JSONValue], where _JSONValue is
    # Union[bool, str, List[Self], Dict[str, Self]].  Since mypy lacks
    # recursive types, we can't define _JSONValue, and use object
    # instead.  Sad.
    _JSONObject = Dict[str, object]

> (again, sorry about mypy's lack of recursive typing, I hate it too, I
> promise)

We got to play the hand we've been dealt.

>> I think we'd like to have
>>
>> * A recursive type for JSON value (in our bastardized version of JSON)
>>
>>   This is Union of bool, str, List[Self], Dict[str, Self].  It's what
>>   .get_expr() returns.
>>
>>   Since mypy can't do recursive, we approximate with _ExprValue.
>>
>> * A recursive type for JSON object
>>
>>   This is the Dict branch of the above.  It's what check_keys() &
>>   friends take as argument.
>>
>>   We approximate with _JSONObject.
>>
>>   Same for the List branch would make sense if we had a use for the
>>   type.
>>
>> * A recursive type for TOP-LEVEL-EXPR
>>
>>   Actually the same type as the previous one, to be used only for the
>>   schema's top-level expressions.  It's the elements of
>>   QAPISchemaParser.exprs[], and what check_exprs() takes as argument.
>>
>>   We approximate with TopLevelExpr, but so far use it only for
>>   check_exprs().
>>
>>   Doesn't really improve type checking, but may serve as documentation.
>>
>>
> That's the intended effect here -- to help highlight which functions
> operate on those top-level expressions, and which might be invoked in more
> arbitrary cases.
> Consider also a hypothetical TOP-LEVEL-EXPR that is actually a bona-fide
> object with additional metadata, too. In these cases, the type really will
> be different.
>
>
>> Shouldn't these types be all defined in one place, namely right here?
>> Bonus: we need to explain the mypy sadness just once then.
>>
>> Shouldn't their names be more systematic?  _ExprValue, _JSONObject and
>> TopLevelExpr hardly suggest any relation...
>>
>>
> I drop _JSONObject in pt5c, it was a stop-gap. For the rest, I'll see about
> a more rigorous consolidation now that we're this far in.
>
> pt5c was intended as a "cleanup" step that did some of that consolidation
> of nearly-redundant types; but I wanted all of parser.py under the mypy gun
> *first*.
> Will you take a raincheck here and we'll focus on the consolidation in the
> next series? I agree it's worth doing.

Works for me.

> (I can add a 'FIXME' that will 100% need to be fixed before I move
> scripts/qapi under python/qemu/qapi -- the linter config there prohibits
> them, so you can be sure I can't ignore it.)

Makes sense.