fix for crash when iterating over a collection which allocates the collection and frees on when finished.
The ability for BPy_StructRNA to hold a reference to other PyObject's was added to support this.
Previously the api just converted the collection to a list and got the iterator from the list to return.
This has the advantage that it uses minimal memory on large collections where before it would make an array.
Though the main reason for this change is to support a bugfix for collections which free memory when they are done, this currently crashes the python api since once the list is built, the data is freed which is used by the list items in some cases (dynamic enums for eg).
Second method for not having python crash blender on invalid access (ifdef'd out ATM, so no functional change).
This uses a weakref list per ID, and invalidates all members of that list when the ID is freed.
the list is not stores in the ID pointer but using a hash table since storing python in DNA data is not acceptable.
This is more correct then the previous method but shows down execution of scripts significantly since its always adding and removing from lists when data is created and freed.
This uses pythons GC so its no overhead during runtime but makes removing ID's slower.
Commented definition 'USE_PYRNA_INVALIDATE_GC' for now, so no functional change.
- KeyingSetInfo classes are now collected like Panels, Operators etc so bpy.utils.register_module() can be used.
- move bpy.types.register() to bpy.utils.register_class
- operators which reload G.main would crash blender if called from python and then accessed bpy.data.*
- WM_read_homefile_exec was setting the contexts Scene to NULL as a signal for the event system, this didnt work in background mode, crashing when property update functions expected scene to be set.
- now writing to RNA is disabled when inside render() call.
- disallow calling operators when writes are disabled.
Rendering runs in a thread so running operators from the thread is not safe unless rendering becomes a blocking operator again.
applied to python api and exotic.c, removed some args being passed down which were not needed.
keyword args for new mathutils types were being ignored when they should raise an error.
The reason this didnt work is all sibclasses of pythons type() or PyType_Type in C, have to have their size set to sizeof(PyHeapTypeObject) rather then sizeof(PyTypeObject) as you might expect.
This is strange since its not a heap-class (defined in pythons runtime), but a static C type, so Im not sure about this, and cant find any documentation but it seems to work ok.
An example of how this is useful - an importer mixin could define the filepath properties and a generic invoke function which can run the subclasses exec for each selected file.
- Panels and Menus now skip the property check when registering.
- renamed _idproperties_ to _idprops_ in function names, function names were getting very long.
- removed the immediate option from C/api and now store in python only, when python loads modules it sets it to False.
- unloading a module would clear the entire TypeMap for all modules, only remove the module types that is being unloaded.
- added some checks for bad class registering, report errors rather then crashing.
F8 key enabled again, useful for script UI development.
- keying set freeing wasnt freeing from all scenes and the builtin list.
- PointerProperty() cant refer to a removed python srna type (fixed in rigify and netrender).
- Added a check for freeing a type used by a PointerProperty but its very slow, makes reloading take ~10sec. Only enabled this in debug mode for now.
Netrender register() function isnt re-registering the property, probably because the module is cached by python and not re-run.
split PropertyRNA off into 3 types, base type, collection and array, since array and collections needed internal checks inside almost every function its better to have the, as subclassed to the property type. This makes introspection more useful.
Also made printing of structs and properties prettier giveing type and length.
Fixed and completed support for returning multiple values. This includes support for returning arrays, both fixed and dynamically sized. The way this is achieved is by storing an additional int value next to the dynamic parameter in the ParameterList stack which gets passed to the C function as an additional parameter. In the case of return parameters it is duty of the C function to set this int to the correct length value for the dynamic parameter (which makes sense). Note that for the dynamic output/return parameter it is assumed the function has allocated that memory (which gets freed automatically).
Also, I cleaned the makesrna's bridge function generation code a bit and renamed PROP_RETURN to PROP_OUTPUT, which represents better the reality now that there are multiple returns. The function now to mark multiple returns (outputs) is RNA_def_function_output.
For an example, look at Action.get_frame_range in rna_action_api.c, by the way Aligorith I removed the #ifdef for this function now that there's support for returning arrays, feel free to modify (the function seems to work).
written from scratch by Daniel Salazar (zanqdo). added own modifications.
New property type
bpy.props.FloatVectorProperty(), only difference with float is it takes a 'size' argument and optional 'default' sequence of floats.
moved bpy.props.* functions out of bpy_rna.c into their own C file.
bpy.types.register(MacroClass)
instead of
bpy.ops.add_macro(MacroClass)
The rest is unchanged.
Also remove some now unused code for the old registration methods (there's still some remaining).
- use python malloc's in bpy_array.c
- automatically blending bone locations is disabled if the target bone has locked location
- neck had incorrect roll
Added a group example
C = bpy.context
ob = C.active_object
bpy.data.groups[0].objects.add(ob)
- add_to_group and rem_from_group now take optional scene and base flags and deal with updating the object & base flags
- operators that add objects to groups were setting ob->recalc= OB_RECALC_OB; looks like its not needed.
- previously add() ignored python args, now add and remove are called like any other FunctionRNA from python.
- made the pyrna api use tp_getset's for collestions active/add()/remove()
- python defined classes will be used when available (otherwise automaically generated metaclasses are made as before)
- use properties rather then functions for python defined rna class's
- call the classes getattr AFTER doing an RNA lookup, avoids setting and clearing exceptions for most attribute lookups, tested UI scripts are ~25% faster.
- extending rna py classes this way is a nicer alternative to modifying the generated metaclasses in place.
Example class
--- snip
class Object(bpy.types.ID):
def _get_children(self):
return [child for child in bpy.data.objects if child.parent == self]
children = property(_get_children)
--- snip
The C initialization function looks in bpy_types.py for classes matching RNA structure names, using them when available.
This means all objects in python will be instances of these classes.
Python properties/funcs defined in ID py class will also be available for subclasses for eg. (Group Mesh etc)
# Before
[
bpy.props.StringProperty(attr="path", name="File Path", description="File path used for exporting the PLY file", maxlen= 1024, default= ""),
bpy.props.BoolProperty(attr="use_modifiers", name="Apply Modifiers", description="Apply Modifiers to the exported mesh", default= True),
bpy.props.BoolProperty(attr="use_normals", name="Export Normals", description="Export Normals for smooth and hard shaded faces", default= True),
bpy.props.BoolProperty(attr="use_uvs", name="Export UVs", description="Exort the active UV layer", default= True),
bpy.props.BoolProperty(attr="use_colors", name="Export Vertex Colors", description="Exort the active vertex color layer", default= True)
]
# After
path = StringProperty(attr="", name="File Path", description="File path used for exporting the PLY file", maxlen= 1024, default= "")
use_modifiers = BoolProperty(attr="", name="Apply Modifiers", description="Apply Modifiers to the exported mesh", default= True)
use_normals = BoolProperty(attr="", name="Export Normals", description="Export Normals for smooth and hard shaded faces", default= True)
use_uvs = BoolProperty(attr="", name="Export UVs", description="Exort the active UV layer", default= True)
use_colors = BoolProperty(attr="", name="Export Vertex Colors", description="Exort the active vertex color layer", default= True)
Multidim. arrays can now be modified at any level, for example:
struc.arrayprop = x
struc.arrayprop[i] = x
struc.arrayprop[i][j] = x
struc.arrayprop[i][j][k] = x
etc...
Approriate rvalue type/length checking is done.
To ensure all works correctly, I wrote automated tests in release/test/rna_array.py.
These tests cover: array/item access, assignment on different levels, tests that proper exceptions are thrown on invalid item access/assignment.
The tests use properties of the RNA Test struct defined in rna_test.c. This struct is only compiled when building with BF_UNIT_TEST=1 scons arg.
Currently unit tests are run manually by loading the script in the Text Editor.
Here's the output I have: http://www.pasteall.org/7644
Things to improve here:
- better exception messages when multidim. array assignment fails. Those we have currently are not very useful for multidim.
- add tests for slice assignment
Example code: http://www.pasteall.org/7332/c.
New API functions: http://www.pasteall.org/7330/c.
Maximum number of dimensions is currently limited to 3, but can be increased arbitrarily if needed.
What this means for ID property access:
* MeshFace.verts - dynamic array, size 3 or 4 depending on MFace.v4
* MeshTextureFace.uv - dynamic, 2-dimensional array, size depends on MFace.v4
* Object.matrix - 2-dimensional array
What this means for functions:
* more intuitive API possibility, for example:
Mesh.add_vertices([(x, y, z), (x, y, z), ...])
Mesh.add_faces([(1, 2, 3), (4, 5, 6), ...])
Python part is not complete yet, e.g. it is possible to:
MeshFace.verts = (1, 2, 3) # even if Mesh.verts is (1, 2, 3, 4) and vice-versa
MeshTextureFace.uv = [(0.0, 0.0)] * 4 # only if a corresponding MFace is a quad
but the following won't work:
MeshTextureFace.uv[3] = (0.0, 0.0) # setting uv[3] modifies MTFace.uv[1][0] instead of MTFace.uv[3]
possible from python, but it's still work in progress.
Pointers and collections are restricted to types derived from
IDPropertyGroup (same as for operators), because RNA knows how to
allocate/deallocate those.
Collections have .add() and .remove(number) functions that can be
used. The remove function should be fixed to take an other argument
than a number.
With the IDPropertyGroup restriction, pointers are more like nested
structs. They don't have add(), remove() yet, not sure where to put
them. Currently the pointer / nested struct is automatically allocated
in the get() function, this needs to be fixed, rule is that RNA get()
will not change any data for thread safety.
Also, it is only possible to add properties to structs after they have
been registered, which needs to be improved as well.
Example code:
http://www.pasteall.org/7201/python
you can copy operator strings from buttons or the reporting interface and run them in the console.
- Ctrl+C over an operator button copies its python string to the clipboard.
- Paste in the console (1 line only for now).
- operators run from python no longer require all arguments.
calling rna functions with the wrong argument type would raise an error like..
expected a string type
Added an error prefix so now the message is..
TypeError: UILayout.item_enumO(): error with argument 3, "value" - expected a string type
Merging changes made by Arystanbek in the soc-2009-kazanbas branch,
plus some things modified and added by me.
* Operator exec is called execute in python now, due to conflicts
with python exec keyword.
* Operator invoke/execute now get context argument.
* Fix crash executing operators due to bpy_import_main_set not being
set with Main pointer.
* The bpy.props module now has the FloatProperty/IntProperty/
StringProperty/BoolProperty functions to define RNA properties for
operators.
* Operators now have an __operator__ property to get the actual RNA
operator pointers, this is only temporary though.
* bpy.ops.add now allows the operator to be already registered, it
will simply overwrite the existing one.
* Both the ui and io directories are now scanned and run on startup.
- Support for python to convert a PyObject into a collection (uses a list of dicts - quite verbose :/)
- Operators can now take collection args when called from python.
- Support for printing operators that use collections (macro recording).
- Added RNA_pointer_as_string which prints all pointer prop values as a python dict.
Example that can run in the in test.py (F7 key)
bpy.ops.VIEW3D_OT_select_lasso(path=[{"loc":(0, 0), "time":0}, {"loc":(1000, 0), "time":0}, {"loc":(1000, 1000), "time":0}], type='SELECT')
for some reason lasso locations always print as 0,0. Need to look into why this is.
