Geskriptete Objekte werden jedes Mal neu aufgebaut, wenn ein FCStd Dokument geöffnet wird. Zu diesem Zweck behält das Dokument eine Referenz auf das Modul und die Python Klasse, die zur Erstellung des Objekts verwendet wurden, zusammen mit seinen Eigenschaften.
Attribute der Klasse, die zur Erstellung des Objekts verwendet wurde, können ebenfalls gespeichert, d.h. "serialisiert" werden. Dies kann weiter durch die Methoden dumps
und loads
der Klasse gesteuert werden.
Standardmäßig werden in einer Objektklasse die Attribute aus dem __dict__
Wörterbuch der Klasse gespeichert.
# various_states.py
class VariousStates:
def __init__(self, obj):
obj.addProperty("App::PropertyLength", "Length")
obj.addProperty("App::PropertyArea", "Area")
obj.Length = 15
obj.Area = 300
obj.Proxy = self
Type = dict()
Type["Version"] = "Custom"
Type["Release"] = "production"
self.Type = Type
self.Type = "Custom"
self.ver = "0.18"
self.color = (0, 0, 1)
self.width = 2.5
def execute(self, obj):
pass
Mit dieser Klasse kann ein Objekt erstellt und unter mein_Dokument.FCstd gespeichert werden. Wenn dem neuen Objekt kein bestimmter Ansichtsanbieter zugewiesen ist, wird seine Proxy Klasse einfach auf einen anderen Wert als None
gesetzt, in diesem Fall auf 1
.
import FreeCAD as App
import various_states
doc = App.newDocument()
doc.FileName = "my_document.FCStd"
obj = doc.addObject("Part::FeaturePython", "Custom")
various_states.VariousStates(obj)
if App.GuiUp:
obj.ViewObject.Proxy = 1
doc.recompute()
doc.save()
Wenn wir die Datei erneut öffnen, können wir das Wörterbuch der Klasse des Objekts einsehen.
>>> obj = App.ActiveDocument.Custom
>>> print(obj.Proxy)
<various_states.VariousStates object at 0x7f0a899bde10>
>>> print(obj.Proxy.__dict__)
{'Type': {'Version': 'Custom', 'Release': 'production'}, 'ver': '0.18', 'color': [0, 0, 1], 'width': 2.5}
Wir sehen, dass alle Attribute, die in der Klasse mit self
beginnen, gespeichert wurden. Diese können von unterschiedlichem Typ sein, einschließlich Zeichenfolge, Liste, Fließkommazahl und Wörterbuch. Das ursprüngliche Tupel für self.color
wurde in eine Liste konvertiert, aber ansonsten wurden alle Attribute gleich geladen.
Wir können eine Klasse ähnlich der ersten definieren, die bestimmte Attribute zum Speichern implementiert.
# various_states.py
class CustomStates:
def __init__(self, obj):
obj.addProperty("App::PropertyLength", "Length")
obj.addProperty("App::PropertyArea", "Area")
obj.Length = 15
obj.Area = 300
obj.Proxy = self
Type = dict()
Type["Version"] = "Custom"
Type["Release"] = "production"
self.Type = Type
self.ver = "0.18"
self.color = (0, 0, 1)
self.width = 2.5
def execute(self, obj):
pass
def dumps(self):
return self.color, self.width
def loads(self, state):
self.color = state[0]
self.width = state[1]
The return value of dumps
is the object that will be serialized. This can be a single value, or a tuple of values. When the object is restored, the class calls the loads
method, passing the state
variable with the serialized content. In this case state
is a tuple that is unpacked into the respective variables to reconstruct the "state" that original existed.
state = (self.color, self.width)
state = ((0, 0, 1), 2.5)
We can create an object with this class, and save the document, just like in the previous example. When we re-open the file we can inspect the dictionary of the object's class.
>>> obj2 = App.ActiveDocument.Custom2
>>> print(obj2.Proxy)
<various_states.CustomStates object at 0x7fb399496630>
>>> print(obj2.Proxy.__dict__)
{'color': [0, 0, 1], 'width': 2.5}
The original tuple for self.color
was converted to a list, but otherwise the information was recovered fine. Instead of restoring all attributes like in the previous case, only the attributes that we specified in dumps
and loads
were restored.
Normally, scripted objects should use properties to store information, as these are automatically restored when the document is opened.
However, sometimes the class restore internal information which is not intended to be modified, but which is helpful to inspect.
For example, most objects of the Draft Workbench set up a Type
attribute that can be used to determine the type of object that is under use.
class DraftObject:
def __init__(self, obj, _type):
self.Type = _type
def dumps(self):
return self.Type
def loads(self, state):
if state:
self.Type = state
All objects have a TypeId
property, but for scripted objects this property is not useful, as it always refers to the parent C++ class, for example, Part::Part2DObjectPython
or Part::FeaturePython
. Therefore, having this additional Proxy.Type
attribute in the class is useful to treat each object in a particular way.
Version information can be stored inside the class in order to verify the origin of an object.
class CustomObject:
def __init__(self, obj, _type):
self.Type = _type
self.version = "0.18"
def dumps(self):
return self.Type, self.version
def loads(self, state):
if state:
self.Type = state[0]
self.version = state[1]
If the structure of the class changes, that is, if its properties or methods change, are renamed, or are removed, we could test the version attribute in order to migrate the older object to a new set of properties or to a new class. This can be done by implementing the onDocumentRestored
method, as explained in Scripted objects migration.
class CustomObject:
def onDocumentRestored(self, obj):
if hasattr(obj.Proxy, "version") and obj.Proxy.version:
if obj.Proxy.version == "0.18":
self.migrate_from_018(obj)
def migrate_from_018(self, obj):
...
dumps
and loads
in the forum.