* External caches didn't load for smoke straight away. Smoke caches store all necessary info in every file, so no need to try to look for an info file.
would stop the entire menu from drawing if they raised an exception.
now print the exception and continue.
Also added a verbose argument for bpy.utils.(un)register_module() to help test whats being registered.
- 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
after discussion over the mailing-list it seems that it really makes no sense to restrict that. Specially since we have the "Property" playback mode that usually will be a float.
Since there is no need for do_version or whatoever I don't see a problem. Any Thoughts? Maube to allow it only through the Python API?
This removes auto-registration, committed by Martin r30961.
Realize this is a contentious topic but Brecht and myself both would rather opt-in registration.
TODO:
- addons need updating.
- class list will be modified to use weakrefs (should have been done for existing system too).
- will move bpy.types.(un)register functions into bpy.utils.(un)register_class, currently including these functions in a type list is internally ugly, scripts which loop over types also need to check for these.
Good discovery this one.
It appeared that "pose mode" changes were not registered in our undo system.
That way the first operator you try to redo after exit posemode fails.
Adding a byte rect to float ImBuf was always freeing the mipmap levels.
Removed this convention since it crashes renders + image texture draw.
Proper ownership handling of ImBuf is high on the wish list :)
The backdrop zoom factor for new node-editor instances was not set
(i.e. was default initialised to 0). Now, this gets set to 1.0.
Also, set the property default in RNA to match this.
conections
It seems that some of the Outliner hacks used while building the tree
was causing problems, as Make Single User (and potentially other code
working with ID-data, specifically with the "newid" value there) was
making use of the variable used there for other purposes, leading to
memory corruption.
This bug also occurred in 2.4x, though when I tested there, it crashed
immediately.
Ton, you may want to double-check this bug!
tested that correcting invalid meshes works by generating random meshes and checking that only the first call to mesh.validate() makes changes.
found 2 bugs in mesh validation.
- face sorting array wasn't assigned correct indices.
- removing invalid edges used wrong comparison.
Recoded animation channel sorting code. In particular, the old code
didn't handle "islands" of selected items well (i.e. a chain of
several connected items in a row), with some of these cases having
unpredictable results.
There were also some bugs in the way some of the rearranging methods
worked, allowing some invalid operations to be performed. Some of
these probably triggered errors such as some channels getting stuck,
and so on.
from Shinsuke Irie (irie) with some minor edits.
Shinsuke's description from the tracker:
---
I have implemented GVFS framework support of blender-thumbnailer.py which allows some file managers like Nautilus and Thunar to show thumbnails in trash or network directories. If Python's gio module is available, the thumbnailer uses it to access to filesystems mounted via GVFS. This change shouldn't affect desktop environments other than GNOME and XFCE.
A function gvfs_open() in this patch is defined to solve a stupid incompatibility between Python file object and GIO Seekable object.
On Ubuntu 10.10, I confirmed thumbnails can be generated for file://, trash://, sftp://, and smb://.
detects...
- invalid vertex range for edges/faces
- duplicate indices in edge/face
- duplicate edges/faces in mesh
- missing edges data in faces
At the moment it doesn't correct errors, but eventually it will do this.
- Tidying up some inconsistent formatting
- Names of old IPO blocks are now included in the names used for new
actions. These are included after a "CDA:" prefix, (i.e. "_C_onverted
_D_ata _A_ction:"), which makes it easier to browse through these
actions later.
One-liner fix - a missing "OPERATOR_FINISHED" on the select operator
was causing problems renaming markers and potentially with other
operations too!
To find this bug, I added debug method to dump the list of markers to
console. This has revealed some troublesome things about the way
markers are organised, which IMO need to be addressed.
-> 2.50
Actionified ShapeKey IPO-blocks (i.e. "Shape Key Actions") would have
an action channel with the hardcoded name, "Shape", and this action
would be assigned to Object level (although ShapeKey blocks had their
own IPO-block slot, only Objects could have actions, so actionifying
ShapeKey IPO-blocks would wrap a ShapeKey block's IPO's to an Object-
level action).
Hence, the path conversions code would wrongly interpret this action
channel as referring to a Pose Channel instead, thus creating some
invalid paths with a 'pose.bones["Shape"]' prefix wrongly getting
tacked on. To ensure that the converted animation can work out of the
box, a 'data.shape_keys' prefix is now used instead so that these
actions can still be Object-rooted while still being able to correctly
control the Shape Keys. This is because there's no easy way to
identify and then shift such action from Object-level to ShapeKey-
level within the conversion code. The consequence though is that such
converted ShapeKey actions CAN ONLY BE USED THROUGH OBJECT LEVEL (i.e.
via Action NOT ShapeKey editor).
Secondly, the Action/ShapeKey editor version patching code has been
modified so that if a ShapeKey editor view was active when loading an
old 2.4x file, the action gets cleared from the view. This is because
of this didn't make semantic sense: the ShapeKey editor is for
ShapeKey-rooted actions, while the Action Editor is for Object-rooted
actions. The converted files though let Object-level actions be shown
in either one.
