Add abstract methods to mjcf.Element to make it a bit friendlier to type-safe code.

nimrod-gileadi · copybara-github · commit ebb0f8944106 · 2022-03-25T05:44:05.000-07:00
PiperOrigin-RevId: 437225654
Change-Id: I259920123295eff2db8ce6c05d3d36ab8ee938a0
diff --git a/dm_control/mjcf/base.py b/dm_control/mjcf/base.py
@@ -29,3 +29,235 @@ class Element(metaclass=abc.ABCMeta):
   tree, all references held automatically become invalid.
   """
   __slots__ = []
+
+  @abc.abstractmethod
+  def get_init_stack(self):
+    """Gets the stack trace where this element was first initialized."""
+
+  @abc.abstractmethod
+  def get_last_modified_stacks_for_all_attributes(self):
+    """Gets a dict of stack traces where each attribute was last modified."""
+
+  @abc.abstractmethod
+  def is_same_as(self, other):
+    """Checks whether another element is semantically equivalent to this one.
+
+    Two elements are considered equivalent if they have the same
+    specification (i.e. same tag appearing in the same context), the same
+    attribute values, and all of their children are equivalent. The ordering
+    of non-repeated children is not important for this comparison, while
+    the ordering of repeated children are important only amongst the same
+    type* of children. In other words, for two bodies to be considered
+    equivalent, their child sites must appear in the same order, and their
+    child geoms must appear in the same order, but permutations between sites
+    and geoms are disregarded. (The only exception is in tendon definition,
+    where strict ordering of all children is necessary for equivalence.)
+
+    *Note that the notion of "same type" in this function is very loose:
+    for example different actuator element subtypes are treated as separate
+    types when children ordering is considered. Therefore, two <actuator>
+    elements might be considered equivalent even though they result in different
+    orderings of `mjData.ctrl` when compiled. As it stands, this function
+    is designed primarily as a testing aid and should not be used to guarantee
+    that models are actually identical.
+
+    Args:
+      other: An `mjcf.Element`
+
+    Returns:
+      `True` if `other` element is semantically equivalent to this one.
+    """
+
+  @property
+  @abc.abstractmethod
+  def tag(self):
+    pass
+
+  @property
+  @abc.abstractmethod
+  def spec(self):
+    pass
+
+  @property
+  @abc.abstractmethod
+  def parent(self):
+    pass
+
+  @property
+  @abc.abstractmethod
+  def namescope(self):
+    pass
+
+  @property
+  @abc.abstractmethod
+  def root(self):
+    pass
+
+  @abc.abstractmethod
+  def prefixed_identifier(self, prefix_root):
+    pass
+
+  @property
+  @abc.abstractmethod
+  def full_identifier(self):
+    """Fully-qualified identifier used for this element in the generated XML."""
+
+  @abc.abstractmethod
+  def find(self, namespace, identifier):
+    """Finds an element with a particular identifier.
+
+    This function allows the direct access to an arbitrarily deeply nested
+    child element by name, without the need to manually traverse through the
+    object tree. The `namespace` argument specifies the kind of element to
+    find. In most cases, this corresponds to the element's XML tag name.
+    However, if an element has multiple specialized tags, then the namespace
+    corresponds to the tag name of the most general element of that kind.
+    For example, `namespace='joint'` would search for `<joint>` and
+    `<freejoint>`, while `namespace='actuator'` would search for `<general>`,
+    `<motor>`, `<position>`, `<velocity>`, and `<cylinder>`.
+
+    Args:
+      namespace: A string specifying the namespace being searched. See the
+        docstring above for explanation.
+      identifier: The identifier string of the desired element.
+
+    Returns:
+      An `mjcf.Element` object, or `None` if an element with the specified
+      identifier is not found.
+
+    Raises:
+      ValueError: if either `namespace` or `identifier` is not a string, or if
+        `namespace` is not a valid namespace.
+    """
+
+  @abc.abstractmethod
+  def find_all(self, namespace,
+               immediate_children_only=False, exclude_attachments=False):
+    """Finds all elements of a particular kind.
+
+    The `namespace` argument specifies the kind of element to
+    find. In most cases, this corresponds to the element's XML tag name.
+    However, if an element has multiple specialized tags, then the namespace
+    corresponds to the tag name of the most general element of that kind.
+    For example, `namespace='joint'` would search for `<joint>` and
+    `<freejoint>`, while `namespace='actuator'` would search for `<general>`,
+    `<motor>`, `<position>`, `<velocity>`, and `<cylinder>`.
+
+    Args:
+      namespace: A string specifying the namespace being searched. See the
+        docstring above for explanation.
+      immediate_children_only: (optional) A boolean, if `True` then only
+        the immediate children of this element are returned.
+      exclude_attachments: (optional) A boolean, if `True` then elements
+        belonging to attached models are excluded.
+
+    Returns:
+      A list of `mjcf.Element`.
+
+    Raises:
+      ValueError: if `namespace` is not a valid namespace.
+    """
+
+  @abc.abstractmethod
+  def enter_scope(self, scope_identifier):
+    """Finds the root element of the given scope and returns it.
+
+    This function allows the access to a nested scope that is a child of this
+    element. The `scope_identifier` argument specifies the path to the child
+    scope element.
+
+    Args:
+      scope_identifier: The path of the desired scope element.
+
+    Returns:
+      An `mjcf.Element` object, or `None` if a scope element with the
+      specified path is not found.
+    """
+
+  @abc.abstractmethod
+  def get_attribute_xml_string(self, attribute_name, prefix_root=None):
+    pass
+
+  @abc.abstractmethod
+  def get_attributes(self):
+    pass
+
+  @abc.abstractmethod
+  def set_attributes(self, **kwargs):
+    pass
+
+  @abc.abstractmethod
+  def get_children(self, element_name):
+    pass
+
+  @abc.abstractmethod
+  def add(self, element_name, **kwargs):
+    """Add a new child element to this element.
+
+    Args:
+      element_name: The tag of the element to add.
+      **kwargs: Attributes of the new element being created.
+
+    Raises:
+      ValueError: If the 'element_name' is not a valid child, or if an invalid
+        attribute is specified in `kwargs`.
+
+    Returns:
+      An `mjcf.Element` corresponding to the newly created child element.
+    """
+
+  @abc.abstractmethod
+  def remove(self, affect_attachments=False):
+    """Removes this element from the model."""
+
+  @property
+  @abc.abstractmethod
+  def is_removed(self):
+    pass
+
+  @abc.abstractmethod
+  def all_children(self):
+    pass
+
+  @abc.abstractmethod
+  def to_xml(self, prefix_root=None, debug_context=None):
+    """Generates an etree._Element corresponding to this MJCF element.
+
+    Args:
+      prefix_root: (optional) A `NameScope` object to be treated as root
+        for the purpose of calculating the prefix.
+        If `None` then no prefix is included.
+      debug_context: (optional) A `debugging.DebugContext` object to which
+        the debugging information associated with the generated XML is written.
+        This is intended for internal use within PyMJCF; users should never need
+        manually pass this argument.
+
+    Returns:
+      An etree._Element object.
+    """
+
+  @abc.abstractmethod
+  def to_xml_string(self, prefix_root=None,
+                    self_only=False, pretty_print=True, debug_context=None):
+    """Generates an XML string corresponding to this MJCF element.
+
+    Args:
+      prefix_root: (optional) A `NameScope` object to be treated as root
+        for the purpose of calculating the prefix.
+        If `None` then no prefix is included.
+      self_only: (optional) A boolean, whether to generate an XML corresponding
+        only to this element without any children.
+      pretty_print: (optional) A boolean, whether to the XML string should be
+        properly indented.
+      debug_context: (optional) A `debugging.DebugContext` object to which
+        the debugging information associated with the generated XML is written.
+        This is intended for internal use within PyMJCF; users should never need
+        manually pass this argument.
+
+    Returns:
+      A string.
+    """
+
+  @abc.abstractmethod
+  def resolve_references(self):
+    pass
