Bug #13277, reported by Juho (bebraw) Vepsalainen (thanks!) - a typo made pynodes w/o input socket definitions fail with an out of range error when accessing the output sockets in the script.
- "Memoryblock free: attempt to free NULL pointer" messages related to the Material node (Add->Input->Material). Deleting it and quitting Blender would bring 2 or 3 of these warnings. Trivial fix (check if NULL) in blenkernel/intern/node.c, though Nodes devs may be interested. Found while testing to fix the following bug:
== PyNodes ==
- Bug #11715 reported by Alexanter Feterman:
http://projects.blender.org/tracker/?func=detail&atid=125&aid=11715&group_id=9
Adding a Dynamic node and setting it to a pynode script would crash Blender if no Material Node (MatNode) was present.
Thanks Alexander for reporting and Brecht for assigning it to me.
Bug #10104 reported by bebraw: missing check for how many node sockets were being created by a pynode script. Too many (more than MAX_SOCKET == 64) would crash Blender.
http://projects.blender.org/tracker/?func=detail&atid=125&aid=10104&group_id=9
Notes: moved the MAX_SOCKET define from node.c to BKE_node.h so I could use it in Node.c. Also improved error reporting in pynodes when errors occur in the init stage.
Thanks Juho (bebraw), Tom (assigned the bug to me) and Brecht (mentioned the MAX_SOCKET define).
1) Added support for a var called "__node__" in pynode scripts, that can be used to point to the desired pynode object, to make sure it is chosen.
2) Fixed a semi-obscure crash that could happen when reparsing a pynode script that was used by multiple nodes in different scenes. Memory corruption would happen if the reparsing failed during the recreation of the pynode object, not when executing it.
Finishing a few needed things before the release: changed ShadeInput vars access from a single getAttribute function to individual methods, one for each var, for faster access (needed, since these are accessed over and over during rendering).
Very repetitive code, so I used a few macros to write it. Compiles w/o warnings here (linux, gcc 4.3.0), please report any problems.
This is just an internal change, the Python API is still the same.
Ongoing updates to the Blender.Node Python API:
- Changed Blender.Node.node to Blender.Node.Scripted to be more specific and conform to bpython API.
- Added a Blender.Node.Socket type to be used to define node sockets in a PyNode script. Also, socket type is inferred from the value(s) passed in, instead of also being defined by the script author.
- Added attr access to input and output sockets in the __call__ method. Ex: an input socket called 'color' can be accessed as self.input.color now.
These changes break existing pynode scripts, which shouldn't be a problem yet, of course, since we're still finishing this feature for 2.46. The wiki page and sample .blends have already been updated:
http://wiki.blender.org/index.php/BlenderDev/PyNodeshttp://wiki.blender.org/index.php/BlenderDev/PyNodes/API
1) Trying to bring back compatibility with Python2.3.
2) Adding some stubs to compile blender player again on linux.
Please tell me if Blender still doesn't compile with py 2.3 or if the player isn't compiling. There was a binreloc related stub I needed to add, so probably the player wasn't compiling before the pynodes commit.
Thanks PanzerMKZ for reporting and testing part of the fix to py 2.3.
I was careful in selectively rolling back revisions, but if you've committed changes unrelated to BPY mixed with BPY changes, I might have reverted those too, so please double check.
