Includes some loop condition fixes after fuzzing.
The previously identified regression (#43641) is now fixed upstream.
(cherry picked from commit 1bb6491992)
Co-authored-by: Gordon MacPherson <gordon@gordonite.tech>
Co-authored-by: Andrea Catania <info@andreacatania.com>
Co-authored-by: K. S. Ernest (iFire) Lee <ernest.lee@chibifire.com>
This is a complete rewrite of the importer. It will give more deterministic behaviour and has been sponsored by IMVU inc, over 1 year has gone into the development of this importer to remove the burden of the FBX SDK.
This was my project for 1 entire year and I really enjoyed the opportunity to add to Godot.
Along the road of implementing fixes we implemented fbx pivots, animations and inheritance type handling, which in most cases works properly.
We have implemented animation and mesh skinning too this should work out of the box, if there are issues let us know.
It's designed so that you can expand this with ease, and fix bugs easily too.
It can import from Autodesk Maya and import into Godot, with pivots.
There are bits we could polish but for now this is good enough.
Additional fixes made before upstreaming:
- fixed memory leaks
- ensure consistent ordering on mac linux and windows for fbx tree. (very important for material import to be deterministic)
- disabled incorrect warnings for fbx_material
- added compatibility code for /RootNode/ so compat is not broken
- Optimise FBX - directly import triangles
- remove debug messages
- add messages for mesh id, mesh re-import is sometimes slow and we need to know what mesh is being worked on
- Document no longer uses unordered maps
- Removed some usages of &GetRequiredToken replaced with safe *GetRequiredToken() function
- Added parser debugging
- Added ERR_FAIL_CONDS for unsupported mesh formats (we can add these later super easy to do now)
- Add memory debugging for the Tokens and the TokenParser to make it safe
- Add memory initialisation to mesh.cpp surface_tool.h and mesh.h
- Initialise boolean flags properly
- Refactored to correct naming for the fbx_mesh_data.h so you know what data you are working on
- Disabled corruption caused by the FIXME:
- Fixed document reading indexes and index_to_direct vs indexes mode
- Fixed UV1 and UV2 coordinates
- Fixed importer failing to import version 7700 files
- Replaced memory handling in the FBX Document with pointers, before it was dereferencing invalid memory.
- Fixed typed properties
- Improved Document API
- Fixed bug with ProcessDOMConnection() not working with the bool flag set to true.
- Fixed FBX skinning not deforming for more than one single mesh
- Fixed FBX skeleton mapping and skin mapping not being applied properly (now retrieved from document skin list)
- Fixed set_bone_pose being used in final version()
- Fixed material properties exceeding 1.0.
- FBX Document parser revamped to use safe memory practices, and with graceful error messages.
- ScopePtr, TokenPtr and various internal types have been fleshed out to use proper typedefs across the codebase.
- Fixed memory leaks caused by token cleanup failing (now explicit cleanup step, no shared_ptr, etc)
- Fixed bug with PropertyTable not reading all properties and not cleaning up properly.
- Fixed smoothing groups not working
- Fixed normal duplications
- Fixed duplication check for pre-existing coordinates.
- Fixed performance of vertex lookup in large meshes being slow, using lookup table separate to the data for indexing, this reduces import time from 10 minutes of bistro down to 30 seconds.
- Fixed includes requiring absolute path in headers and cpp files using CPPPath.
Bugs/Features wish list:
- locator bones
- quat anim key interpolation (most fbx maya files have euler rotations from blender and maya, nobody uses this)
- some rigs skins scale up when SSC enabled inconsistently per bone
- some skins can disappear entirely
- material mapping needs expanded, but this will be done for 4.0 as it requires rewrite.
Workarounds for issues found until we patch them:
- mesh -> clear skin can resolve most of the bugs above.
- locators can be worked around by removing them before exporting your rig.
- some material properties wont always import, this is okay to override in the material properties.
**If you are having issues or need support fear not!**
Please provide minimal rigs which can reproduce issues as we can't spend a lot of time investigating each rig. We need a small example which breaks and we can then sort the problem. In some cases this is not possible so its okay to privately send models to us via IRC or a ticket and we can provide an email address, we won't reveal or disclose privately sent rig files to any companies, or to companies I work for, they will not be shared, only tested and bugs will be drawn up from the conclusions. Also include identifying information about what you did and how it didn't work. Please file each file separately in a bug report, unless the problem is the same.
This was sponsored by IMVU, and a special thanks to everyone who supported this project.
Signed-off-by: Gordon MacPherson <gordon@gordonite.tech>
Upstream development restarted after 13 years. Changes:
2020-02-02: Version 0.5.0
Minor speed improvement on the decompressor.
Prevent memory violation when decompressing corrupted input.
2020-01-10: Version 0.4.0
Only code & infrastructure clean-up, no new functionality.
(cherry picked from commit 5167c9186a)
The previous code would always use SSE2 intrinsics, which is not valid
on UWP ARM platforms (and likely not on some x86 platforms either).
The patch has been submitted upstream too:
https://github.com/richgel999/jpeg-compressor/pull/13
(cherry picked from commit 3806efbaa7)
The comment mentioned a conflict with libwebsockets, but we actually
still get this conflict even now that we don't use libwebsockets.
Not sure what component is clashing but we should basically just keep
this patch.
Follow-up to #36823.
(cherry picked from commit 8189abd64a)
For some weird reason 'git apply' does not error out when it does nothing,
so I missed that I did not apply the patch properly in #36823...
This broke the UWP 32-bit x86 build.
(cherry picked from commit 9a727714ee)
- Improve the SCsub to allow unbundling and remove unnecessary code.
- Move files around to match upstream source.
- Re-sync with upstream commit 308db73d0b3c2d1870cd3e465eaa283692a4cf23
to ensure we don't have local modifications.
- Doesn't actually build against current version 5.0.1 due to the lack
of the new ArmaturePopulate API that Gordon authored. We'll have to
wait for a public release with that API (5.1?) to enable unbundling.
(cherry picked from commit 9d8a9ea826)
The problem could be related to different byte ordering when copying
the interface address over the binding address.
(cherry picked from commit e85330231c)
Fixes CVE-2019-18222.
`include/mbedtls/version.h` was modified to include the forgotten
version bump to 2.16.4, fixed upstream by ARMmbed/mbedtls#2992.
Happy new year to the wonderful Godot community!
We're starting a new decade with a well-established, non-profit, free
and open source game engine, and tons of further improvements in the
pipeline from hundreds of contributors.
Godot will keep getting better, and we're looking forward to all the
games that the community will keep developing and releasing with it.
This is a very outdated copy of Bullet's btScalar.h,
we're probably only discovering the tip of the bad
cross-platform compatibility of the unmaintained vhacd.
On some systems, including Alpine Linux, musl is used instead of
glibc. This commit patches the third-party V-HACD module to provide
a macro not provided by musl.
Fixes#31555.
- Added correct config file for android dllmaps.
- Fix __Internal DllImports with a dlopen fallback.
- Add missing P/Invoke functions and internal calls expected by the monodroid BCL and our custom version of the 'Android.Runtime.AndroidEnvironment' class (this last one can be found in the godot-mono-builds repo).
- Make sure to set 'btls' instead of 'legacy' as the default TLS provider on Android.
This reverts commit e00426c512.
The way we handle platform-specific intrinsics is not good, so the
current state will not compile on armv8. This commit also requires
SSE4.1 support, which is likely not a good idea for portable binaries.
We'll have to redo this with more caution after 3.2 is released, or
we might simply drop opus as we're only using it as dependency for
theora right now.
Fixes#33606.
Removes the custom code for VS2012 support in config.h,
and reduces diff with upstream config.h.
We still have many custom defines for ARM optimizations
which we probably don't use properly, since we don't
compile the included asm code, so a thorough review and
cleanup would be welcome.
Basic skin support
Various fixes
- Fixes bind mount id and mesh index
- Fixed duplicate nodes being created
- Prevented leak when instances being freed during re-import.
- Improved camera and light transform import
- skeleton handling and technical debt removal
- ASSIMP: bone nodes were unlinked from bones by this code
- bone_add working can distinguish between armatutes
- Updated transform to be the correct offset
- Added safety for state.root node errors
- Fixed memory leak with leaf bones
- Implemented children re-parenting for mesh template
- import_animation fixes to basic skeleton data
- Adds some more debug messages
- Fixed Godot import segfault
- Fix build failing on mono
- Clear resources we use which are no longer required after import
- Fixed bone duplication issue
- Working skeleton_bone_map which can lookup armatures properly now.
- Fixed stack being used up when mesh swapped & Fixed bone ID
Additional notes:
We use a mesh template which is a fake node to instance the initial
mesh nodes . This is to ensure the entire tree can be built.
We replace mesh node templates with the real mesh after the
skeleton is available, since this makes it ensure that the fully
built skeleton exists with all bones, all nodes, etc.
The bone stack is a stack which pops when it finds bones,
this overcomes duplicate bones with the same names.
FBX has lots of these because animation armature has bone names like bone001
and another armature will also have bone001
Fixed errors in node path assignment
Simple explanation:
- Every mesh uses a node from the stack
- Node stack was empties before completed
- Every time node not found, stack must be rebuilt to maintain correct armature order :)
Additional fixes:
- Fixes destructor in assimp
- Implements aiNode* mArmature in bone data
- Implements aiNode* mParent in bone data
- Fixes parent ID on bones.
Implemented skeleton assignment in generate_mesh_indicies
This is the only place we can safely do a lookup for the skeleton for the mesh.h
I used a pointer reference so we can pass this back out, since the skeleton assignment happens inside the function.
Added mesh re-parenting to the armature node this is a permanent feature and must be enforced, just like GLTF2 specification.
Fixed import_animation spawning tracks per skin
The DynamicFont implementation currently in use is based on
FreeType, which provides much better visual quality.
This old implementation wasn't exposed anywhere, so this shouldn't
break compatibility.
This decreases binary sizes by a few kilobytes.
This reads materials properly from Maya and expands on existing functionality to make this work properly.
aiTextureType_SHININESS no longer used as not appropriate for PBR texture as it is legacy.
This fix will be also present in assimp soon.
Issues fixed:
- Updated assimp to latest and backported fixes into godot.
- Fixed file scale being ignored from FBX file.
- Fixed bone removal
- Implemented proper armature binding
- Fixed recursion not always going through the entire path
- Implemented assimp global scaling system
- Fixed assimp global scale process to support unit conversion
- Implemented proper fbx scaling
- Fixed asserts caused by missing faces in some models which could crash
- Fixed valid bone removal
- Fixed root node being overwriten by assimp which caused data loss
- Fixed armature construction so that it works with multiple roots
- Implemented basic support for FBX standard materials
- Refactoring to improve code quality and improve function reuse.
- Simplified node creation from assimp scene into subsections: create_light, create_mesh, create_bone.
- Creating meshes is now done after hierarchy is created so that the skeleton is always available.
- Added support to assimp to support file scale in all formats which call SetFileScale.
- Many other fixes provided into assimp.
Known issues:
- FBX pivots from Maya do not currently work. (workaround: for now use blender import and export to remove pivot tracks)
- Hierarchy creates an extra node for each mesh - this was done intentionally but we intended to do a pass to remove these as they're a required node.
- When an animated mesh has not executed any animation the rest pose is wrong.
Co-authored-by: K. S. Ernest (iFire) Lee <ernest.lee@chibifire.com>
Rémi Verschelde
bruvzg
Gordon MacPherson
volzhs
Fabio Alessandrelli
Marcel Admiraal
Pedro J. Estébanez
CodeforEvolution
James Westman
Ignacio Etcheverry
Jonathan Mannancheril
Jonathan Mannancheril
Hugo Locurcio
K. S. Ernest Lee
follower