Assigned action is lost if file is saved while in NLA tweaking mode on a linked object with linked animation data #106801

New Issue

System Information
Operating system: macOS-13.3-arm64-arm-64bit 64 Bits
Graphics card: Metal API Apple M1 Pro 1.2

Blender Version
Broken: version: 3.5.0, branch: blender-v3.5-release, commit date: 2023-03-29 02:56, hash: 1be25cfff18b
Worked: NA

Short description of error
NLA Editor loses original assigned action to a linked rig generated with rigify if file is saved in tweaking mode on a strip with a different action.

Entering tweaking mode temporarily changes the assigned action in Action Editor with the one the user wants to edit. Normally saving the file in this state and then reopening it will not be a problem, the user will find the interface exactly as it was saved, with tweaking mode active, and getting out of tweaking mode will reset the action in Action Editor to the previous assigned one.

However, if this is done to a linked rig generated with rigify (I could test and reproduce this bug on a human metarig and on a rig with just a pivot with the switchable parent property) reopening the file will result in tweaking mode being off and the termporary assigned action now being the permanent one in Action Editor (with the possible loss of the previous one).

Exact steps for others to reproduce the error
Complete list:
Create a file with a human metarig and generate rig with rigify.
Link the collection with the generated rig in a new file.
Now you have to make a library override, either using the drop down menus in the 3d view "Object -> Relations -> Make Library Override", or "Object -> Library Override -> Make", or by using the right click menu in the outliner going in "Library Override -> Make -> Selected and Content".
Create an action strip (Action_A).
Create a new action in action editor without creating a new strip (Action_B).
Enter tweaking mode on the previous action strip (Action_A).
Save file, then reopen.
Now the action in Action Editor will be Action_A, even though it should still be Action_B. Action_B should now still be in the file with 0 users.

Short list:
Open attached file:[animation_human_clean.blend]
Enter Tweaking mode on action strip, as shown in screenshot
Save file, then reopen.
You should find this result

By making a few tests I've been able to narrow it down to there being any bone with a "switchable parent" rigify property generating this bug.

*Edited the post to add clearer linking and library override instructios in the complete list.

**System Information** Operating system: macOS-13.3-arm64-arm-64bit 64 Bits Graphics card: Metal API Apple M1 Pro 1.2 **Blender Version** Broken: version: 3.5.0, branch: blender-v3.5-release, commit date: 2023-03-29 02:56, hash: `1be25cfff18b` Worked: NA **Short description of error** NLA Editor loses original assigned action to a linked rig generated with rigify if file is saved in tweaking mode on a strip with a different action. Entering tweaking mode temporarily changes the assigned action in Action Editor with the one the user wants to edit. Normally saving the file in this state and then reopening it will not be a problem, the user will find the interface exactly as it was saved, with tweaking mode active, and getting out of tweaking mode will reset the action in Action Editor to the previous assigned one. However, if this is done to a linked rig generated with rigify (I could test and reproduce this bug on a human metarig and on a rig with just a pivot with the switchable parent property) reopening the file will result in tweaking mode being off and the termporary assigned action now being the permanent one in Action Editor (with the possible loss of the previous one). **Exact steps for others to reproduce the error** Complete list: Create a file with a human metarig and generate rig with rigify. Link the collection with the generated rig in a new file. Now you have to make a library override, either using the drop down menus in the 3d view "Object -> Relations -> Make Library Override", or "Object -> Library Override -> Make", or by using the right click menu in the outliner going in "Library Override -> Make -> Selected and Content". Create an action strip (Action_A). Create a new action in action editor without creating a new strip (Action_B). Enter tweaking mode on the previous action strip (Action_A). Save file, then reopen. Now the action in Action Editor will be Action_A, even though it should still be Action_B. Action_B should now still be in the file with 0 users. Short list: Open attached file:[animation_human_clean.blend] ![01_base_file.png](/attachments/afd358f0-22fe-48c3-b71f-57c7e451acf4) Enter Tweaking mode on action strip, as shown in screenshot ![02_how_to_reproduce.png](/attachments/6aaef6e9-d903-43c3-a09f-2ec8c59f8fba) Save file, then reopen. You should find this result ![03_effect.png](/attachments/d4e43486-9e3b-41b1-96e5-48408b774338) By making a few tests I've been able to narrow it down to there being any bone with a "switchable parent" rigify property generating this bug. *Edited the post to add clearer linking and library override instructios in the complete list.

02_how_to_reproduce.png

1.4 MiB

linked_file_human_rig.blend

3.6 MiB

01_base_file.png

1.4 MiB

animation_human_clean.blend

3.0 MiB

03_effect.png

1.4 MiB

@bclark This is the issue @icappiello was talking about.

@angavrilov this seems something that you should have a look too. it seems to happen only when a rig contains a rigify space switching enabled sub-rig (basic pivot, limbs, spine)

Thanks, I didn't get the tag notification here :(
Looking.

Thanks, I didn't get the tag notification here :( Looking.

We need to update the ticket to include the entire process including how you made the library override.

I have edited the post by adding a few more informations on the library override.
I basically tested every method I knew to make them and always got the same result.

By the way, the added lines are these:
Link the collection with the generated rig in a new file.
Now you have to make a library override, either using the drop down menus in the 3d view "Object -> Relations -> Make Library Override", or "Object -> Library Override -> Make", or by using the right click menu in the outliner going in "Library Override -> Make -> Selected and Content".

If there's more to add, I'm here.

I have edited the post by adding a few more informations on the library override. I basically tested every method I knew to make them and always got the same result. By the way, the added lines are these: Link the collection with the generated rig in a new file. Now you have to make a library override, either using the drop down menus in the 3d view "Object -> Relations -> Make Library Override", or "Object -> Library Override -> Make", or by using the right click menu in the outliner going in "Library Override -> Make -> Selected and Content". If there's more to add, I'm here.

I've narrowed down the problematic case a bit. This doesn't happen with all rigify rigs. For example, if you create a metarig that only has the super_copy rig type in it, everything works fine.

It appears to be the case that this only happens with Rigify rigs that include rig types that trigger the creation of a weird sub-track on the NLA:

The rig types that trigger the creation of this sub-track at least include the limb.arm and limb.leg types. And this sub-track is created when Rigify generates the rig (i.e. the weird sub-track is already in the rig file, not just when linking).

It's not clear to me yet what that sub-track is (presumably an NLA feature I'm not familiar with?), nor is it at all clear to me why Rigify is messing with NLA tracks at all. But I think this at least narrows things down a little bit.

I've narrowed down the problematic case a bit. This doesn't happen with all rigify rigs. For example, if you create a metarig that only has the `super_copy` rig type in it, everything works fine. It appears to be the case that this only happens with Rigify rigs that include rig types that trigger the creation of a weird sub-track on the NLA: ![weird_sub_track.png](/attachments/9ad62e58-4947-45d5-9ce0-c47e0d28b23e) The rig types that trigger the creation of this sub-track at least include the `limb.arm` and `limb.leg` types. And this sub-track is created when Rigify generates the rig (i.e. the weird sub-track is already in the rig file, not just when linking). It's not clear to me yet what that sub-track is (presumably an NLA feature I'm not familiar with?), nor is it at all clear to me why Rigify is messing with NLA tracks at all. But I think this at least narrows things down a little bit.

weird_sub_track.png

9.3 KiB

With the help of @dr.sybren I figured out what that sub-track is, and was also able to create a minimal repro.

Minimal Repro

Files:

These were created by hand without Rigify. They have the same general structure as the repro files from @CorradoPiscitelli, and the same reported (short list) steps can be used to repro with these files.

Explanation

The weird sub-track turns out to be the animation data of the armature data block (not object).

The situation that triggers this bug is when the rig has animation data on both the armature object and the armature data, in the rig file itself. Having animation data on only the armature object or the armature data block is not enough. Both have to have animation data on them. I must have messed up my testing. The triggering case is specifically having animation data on the armature object. Whether it's on the armature data block or not is irrelevant. (Which makes a lot more sense.)

(Side note: although this does mean the bug isn't specific to Rigify, it's still mysterious to me why Rigify is adding animation data to the armature data block. So that might be worth investigating separately.)

(Edit: turns out the reason for Rigify adding animation data to the armature data block is for drivers. For example, to let the animator tweak the curvature of rubber-hose limbs, since that's all b-bone based and the b-bone properties are on the armature data.)

With the help of @dr.sybren I figured out what that sub-track is, and was also able to create a minimal repro. ### Minimal Repro Files: - [`rig.blend`](/attachments/ddd62354-56ba-41d7-a6b8-0f57693db462) - [`animation.blend`](/attachments/3ffa3b34-c5bf-4c2c-a178-1d787c8ef70e) These were created by hand without Rigify. They have the same general structure as the repro files from @CorradoPiscitelli, and the same reported (short list) steps can be used to repro with these files. ### Explanation The weird sub-track turns out to be the animation data of the *armature data block* (not object). ~~The situation that triggers this bug is when the rig has animation data on both the armature object *and* the armature data, in the rig file itself. Having animation data on only the armature object *or* the armature data block is not enough. Both have to have animation data on them.~~ I must have messed up my testing. The triggering case is specifically having animation data on the armature *object*. Whether it's on the armature data block or not is irrelevant. (Which makes a lot more sense.) (Side note: although this does mean the bug isn't specific to Rigify, it's still mysterious to me why Rigify is adding animation data to the armature data block. So that might be worth investigating separately.) (Edit: turns out the reason for Rigify adding animation data to the armature data block is for drivers. For example, to let the animator tweak the curvature of rubber-hose limbs, since that's all b-bone based and the b-bone properties are on the armature data.)

rig.blend

800 KiB

animation.blend

831 KiB

Thanks for tracking that down!
What is our bug/priority level for this now that we have it narrowed down to reproduce?

Thanks for tracking that down! What is our bug/priority level for this now that we have it narrowed down to reproduce?

What is our bug/priority level for this now that we have it narrowed down to reproduce?

There's some potential for data loss here, with the following situation:

Someone saves the file while in tweak mode.
Later they open it up again, continue working without noticing anything is off yet. But Action_B is no longer used by anything (zero users).
They save again. And Blender doesn't save Action_B, so it's lost on next open.

It's a little unlikely, since the situation will probably be noticed by the user at step 2 above. And if they're using the NLA, they're probably aware of how actions work and can re-assign Action_B without trouble. But I think it's probably worth making this high priority nevertheless. Data loss is one of the worst things you can do to a user.

I'll keep looking at it today, and see if I can fix it, or at least narrow it down enough to figure out who best to ping about it.

> What is our bug/priority level for this now that we have it narrowed down to reproduce? There's some potential for data loss here, with the following situation: 1. Someone saves the file while in tweak mode. 2. Later they open it up again, continue working without noticing anything is off yet. But Action_B is no longer used by anything (zero users). 3. They save again. And Blender doesn't save Action_B, so it's lost on next open. It's a *little* unlikely, since the situation will probably be noticed by the user at step 2 above. And if they're using the NLA, they're probably aware of how actions work and can re-assign Action_B without trouble. But I think it's probably worth making this high priority nevertheless. Data loss is one of the worst things you can do to a user. I'll keep looking at it today, and see if I can fix it, or at least narrow it down enough to figure out who best to ping about it.

Thanks to blender-asset-tracer, I've confirmed that the bug is happening on file save, not load. Action_B is already replaced by Action_A in the on-disk saved file.

Thanks to [`blender-asset-tracer`](https://pypi.org/project/blender-asset-tracer/), I've confirmed that the bug is happening on file *save*, not load. Action_B is already replaced by Action_A in the on-disk saved file.

Nevermind, I was reading the data wrong (*facepalm*). It appears that the saved data is actually correct. So I think whatever is happening is happening on load.

Specifically, the on-disk blend file after saving has the following animdata fields set correctly:

action: set to Action_A
tmpact: set to Action_B
flag: set to 4 (which indicates being in tweak mode, and that the main action has been temporarily moved to tmpact while the tweaking action has been moved to action).

Nevermind, I was reading the data wrong (\*facepalm\*). It appears that the saved data is actually correct. So I think whatever is happening is happening on load. Specifically, the on-disk blend file after saving has the following animdata fields set correctly: - `action`: set to `Action_A` - `tmpact`: set to `Action_B` - `flag`: set to 4 (which indicates being in tweak mode, and that the main action has been temporarily moved to `tmpact` while the tweaking action has been moved to `action`).

I re-tested the triggering cases, and I must have messed up my previous testing, because it appears that only the armature object's animation data is relevant here. This time around I tried all the following permutations in the rig file:

Object w/o animdata, Armature w/o animdata: no bug.
Object w/o animdata, Armature w animdata: no bug.
Object w animdata, Armature w/o animdata: bug triggered.
Object w animdata, Armature w animdata: bug triggered.

Investigating further, it looks like what's going on here is that because the tmpact field of animdata isn't exposed in RNA, the override system doesn't know anything about it and thus it doesn't get preserved in the linking file if there's animdata in the rig file to overwrite it.

The solution (assuming I've identified the source of the bug accurately) is to expose tmpact to RNA as well, so that its state gets properly preserved and overridden locally.

I re-tested the triggering cases, and I must have messed up my previous testing, because it appears that only the armature *object*'s animation data is relevant here. This time around I tried all the following permutations in the rig file: - Object w/o animdata, Armature w/o animdata: no bug. - Object w/o animdata, Armature w animdata: no bug. - Object w animdata, Armature w/o animdata: bug triggered. - Object w animdata, Armature w animdata: bug triggered. Investigating further, it looks like what's going on here is that because the `tmpact` field of animdata isn't exposed in RNA, the override system doesn't know anything about it and thus it doesn't get preserved in the linking file if there's animdata in the rig file to overwrite it. The solution (assuming I've identified the source of the bug accurately) is to expose `tmpact` to RNA as well, so that its state gets properly preserved and overridden locally.

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Assigned action is lost if file is saved while in NLA tweaking mode on a linked object with linked animation data #106801

Minimal Repro

Explanation