All of lore.kernel.org
 help / color / mirror / Atom feed
From: Kevin Wolf <kwolf@redhat.com>
To: Markus Armbruster <armbru@redhat.com>
Cc: pbonzini@redhat.com, berrange@redhat.com, qemu-devel@nongnu.org,
	eblake@redhat.com, ehabkost@redhat.com
Subject: Re: [RFC PATCH 07/12] qapi: Allow defining QOM classes
Date: Fri, 10 Dec 2021 18:53:47 +0100	[thread overview]
Message-ID: <YbOUKw1GhRn5mVuT@redhat.com> (raw)
In-Reply-To: <871r379oku.fsf@dusky.pond.sub.org>

Am 23.11.2021 um 11:02 hat Markus Armbruster geschrieben:
> Kevin Wolf <kwolf@redhat.com> writes:
> 
> > Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> > ---
> >  scripts/qapi/expr.py   | 28 +++++++++++++++++-
> >  scripts/qapi/schema.py | 66 ++++++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 93 insertions(+), 1 deletion(-)
> 
> Missing: docs/devel/qapi-code-gen.rst update.  I understand why, but it
> does make review harder.
> 
> >
> > diff --git a/scripts/qapi/expr.py b/scripts/qapi/expr.py
> > index 3cb389e875..77550629f3 100644
> > --- a/scripts/qapi/expr.py
> > +++ b/scripts/qapi/expr.py
> > @@ -181,6 +181,8 @@ def check_defn_name_str(name: str, info: QAPISourceInfo, meta: str) -> None:
> >      """
> >      if meta == 'event':
> >          check_name_upper(name, info, meta)
> > +    elif meta == 'class':
> > +        check_name_str(name, info, meta)
> 
> This permits arbitrary QAPI names.  I figure we'll want to define and
> enforce a suitable naming convention.

It's the QOM type name. So if you want to enforce something stricter, it
needs to be defined for QOM first. I seem to remember that arbitrary
QAPI names is already stricter than arbitrary QOM names, but I was
hoping that it is close enough to what QOM classes actually use in
practice. I fully expect to have a list of exceptions at some point when
converting devices.

> >      elif meta == 'command':
> >          check_name_lower(
> >              name, info, meta,
> > @@ -557,6 +559,24 @@ def check_alternate(expr: _JSONObject, info: QAPISourceInfo) -> None:
> >          check_type(value['type'], info, source)
> >  
> >  
> > +def check_class(expr: _JSONObject, info: QAPISourceInfo) -> None:
> > +    """
> > +    Normalize and validate this expression as a ``class`` definition.
> > +
> > +    :param expr: The expression to validate.
> > +    :param info: QAPI schema source file information.
> > +
> > +    :raise QAPISemError: When ``expr`` is not a valid ``class``.
> > +    :return: None, ``expr`` is normalized in-place as needed.
> > +    """
> > +    config = expr.get('config')
> > +    config_boxed = expr.get('config-boxed', False)
> > +
> > +    if config_boxed and config is None:
> > +        raise QAPISemError(info, "'boxed': true requires 'config'")
> 
> You check 'config-boxed', but the error message talks about 'boxed'.
> 
> > +    check_type(config, info, "'config'", allow_dict=not config_boxed)
> > +
> > +
> >  def check_command(expr: _JSONObject, info: QAPISourceInfo) -> None:
> >      """
> >      Normalize and validate this expression as a ``command`` definition.
> > @@ -627,7 +647,7 @@ def check_exprs(exprs: List[_JSONObject]) -> List[_JSONObject]:
> >              continue
> >  
> >          metas = expr.keys() & {'enum', 'struct', 'union', 'alternate',
> > -                               'command', 'event'}
> > +                               'class', 'command', 'event'}
> >          if len(metas) != 1:
> >              raise QAPISemError(
> >                  info,
> > @@ -671,6 +691,12 @@ def check_exprs(exprs: List[_JSONObject]) -> List[_JSONObject]:
> >                         ['struct', 'data'], ['base', 'if', 'features'])
> >              normalize_members(expr['data'])
> >              check_struct(expr, info)
> > +        elif meta == 'class':
> > +            check_keys(expr, info, meta,
> > +                       ['class'], ['if', 'features', 'parent', 'config',
> > +                        'config-boxed'])
> > +            normalize_members(expr.get('config'))
> > +            check_class(expr, info)
> >          elif meta == 'command':
> >              check_keys(expr, info, meta,
> >                         ['command'],
> > diff --git a/scripts/qapi/schema.py b/scripts/qapi/schema.py
> > index b7b3fc0ce4..ebf69341d7 100644
> > --- a/scripts/qapi/schema.py
> > +++ b/scripts/qapi/schema.py
> > @@ -155,6 +155,9 @@ def visit_object_type_flat(self, name, info, ifcond, features,
> >      def visit_alternate_type(self, name, info, ifcond, features, variants):
> >          pass
> >  
> > +    def visit_class(self, entity):
> > +        pass
> > +
> >      def visit_command(self, name, info, ifcond, features,
> >                        arg_type, ret_type, gen, success_response, boxed,
> >                        allow_oob, allow_preconfig, coroutine):
> > @@ -766,6 +769,50 @@ def __init__(self, name, info, typ, ifcond=None):
> >          super().__init__(name, info, typ, False, ifcond)
> >  
> >  
> > +class QAPISchemaClass(QAPISchemaEntity):
> > +    meta = 'class'
> > +
> > +    def __init__(self, name, info, doc, ifcond, features, parent,
> > +                 config_type, config_boxed):
> > +        super().__init__(name, info, doc, ifcond, features)
> > +
> > +        assert not parent or isinstance(parent, str)
> 
> I can't see what ensures this.

Indeed, check_class() fails to check this. You agree that it is the
place where it should be checked?

> > +        assert not config_type or isinstance(config_type, str)
> > +        self._parent_name = parent
> > +        self.parent = None
> > +        self._config_type_name = config_type
> > +        self.config_type = None
> > +        self.config_boxed = config_boxed
> > +
> > +    def check(self, schema):
> > +        super().check(schema)
> > +
> > +        if self._parent_name:
> > +            self.parent = schema.lookup_entity(self._parent_name,
> > +                                               QAPISchemaClass)
> > +            if not self.parent:
> > +                raise QAPISemError(
> > +                    self.info,
> > +                    "Unknown parent class '%s'" % self._parent_name)
> > +
> > +        if self._config_type_name:
> > +            self.config_type = schema.resolve_type(
> > +                self._config_type_name, self.info, "class 'config'")
> 
> "class's 'config'" for consistency with other uses of .resolve_type().
> 
> > +            if not isinstance(self.config_type, QAPISchemaObjectType):
> > +                raise QAPISemError(
> > +                    self.info,
> > +                    "class 'config' cannot take %s"
> > +                    % self.config_type.describe())
> > +            if self.config_type.variants and not self.boxed:
> 
> self.config_boxed?
> 
> > +                raise QAPISemError(
> > +                    self.info,
> > +                    "class 'config' can take %s only with 'boxed': true"
> 
> 'config-boxed'?

Yes, I renamed it and of course forgot some instances. Can't wait for
the patches that would make mypy catch at least 'self.boxed'.

> > +                    % self.config_type.describe())
> > +
> > +    def visit(self, visitor):
> > +        super().visit(visitor)
> > +        visitor.visit_class(self)
> > +
> >  class QAPISchemaCommand(QAPISchemaEntity):
> >      meta = 'command'
> >  
> > @@ -1110,6 +1157,23 @@ def _def_alternate_type(self, expr, info, doc):
> >                                      QAPISchemaVariants(
> >                                          None, info, tag_member, variants)))
> >  
> > +    def _def_class(self, expr, info, doc):
> > +        name = expr['class']
> > +        ifcond = QAPISchemaIfCond(expr.get('if'))
> > +        features = self._make_features(expr.get('features'), info)
> > +        parent = expr.get('parent')
> > +        config_type = expr.get('config')
> > +        config_boxed = expr.get('config-boxed')
> > +
> > +        if isinstance(config_type, OrderedDict):
> > +            config_type = self._make_implicit_object_type(
> > +                name, info, ifcond,
> > +                'config', self._make_members(config_type, info))
> 
> Does QAPISchemaMember.describe() need an update for this?

Looks like it.

Kevin



  reply	other threads:[~2021-12-10 17:55 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-03 17:29 [RFC PATCH 00/12] QOM/QAPI integration part 1 Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 01/12] qapi: Add visit_next_struct_member() Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 02/12] qom: Create object_configure() Kevin Wolf
2021-11-23 15:23   ` Markus Armbruster
2021-12-14  9:52     ` Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 03/12] qom: Make object_configure() public Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 04/12] qom: Add instance_config() to TypeInfo Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 05/12] rng-random: Implement .instance_config Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 06/12] rng-backend: " Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 07/12] qapi: Allow defining QOM classes Kevin Wolf
2021-11-23 10:02   ` Markus Armbruster
2021-12-10 17:53     ` Kevin Wolf [this message]
2021-11-03 17:29 ` [RFC PATCH 08/12] qapi: Create qom-config:... type for classes Kevin Wolf
2021-11-23 13:00   ` Markus Armbruster
2021-12-10 17:41     ` Kevin Wolf
2021-11-03 17:29 ` [RFC PATCH 09/12] qapi/qom: Convert rng-backend/random to class Kevin Wolf
2021-11-23 13:15   ` Markus Armbruster
2021-12-10 17:57     ` Kevin Wolf
2021-11-03 17:30 ` [RFC PATCH 10/12] qapi: Generate QOM config marshalling code Kevin Wolf
2021-11-23 14:16   ` Markus Armbruster
2021-12-10 16:50     ` Kevin Wolf
2021-11-03 17:30 ` [RFC PATCH 11/12] qapi/qom: Add class definition for rng-builtin Kevin Wolf
2021-11-03 17:30 ` [RFC PATCH 12/12] qapi/qom: Add class definition for rng-egd Kevin Wolf
2021-11-03 21:26 ` [RFC PATCH 00/12] QOM/QAPI integration part 1 Paolo Bonzini
2021-11-04  9:07   ` Kevin Wolf
2021-11-04 12:39     ` Paolo Bonzini
2021-11-04 14:26       ` Kevin Wolf
2021-11-04 14:49         ` Paolo Bonzini
2021-11-04 15:51           ` Kevin Wolf
2021-11-04 15:52     ` Damien Hedde
2021-11-05 13:55       ` Kevin Wolf
2021-11-23 16:05 ` Markus Armbruster
2021-12-14 10:23   ` Kevin Wolf
2021-12-14 10:40     ` Peter Maydell
2021-12-14 11:52       ` Kevin Wolf
2021-12-14 14:45     ` Markus Armbruster
2021-12-14 16:00       ` Kevin Wolf

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=YbOUKw1GhRn5mVuT@redhat.com \
    --to=kwolf@redhat.com \
    --cc=armbru@redhat.com \
    --cc=berrange@redhat.com \
    --cc=eblake@redhat.com \
    --cc=ehabkost@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.