Python: setting editbone length flips bone Z axis #88581

New Issue

Scurest · 2021-05-25T22:40:55+02:00

Scurest commented

2021-05-25 22:40:55 +02:00

System Information
Operating system: Linux
Graphics card: Intel HD

Blender Version
Broken: 3.0 (specifically I'm testing with f9567f6c63), also 2.83 (latest release fd30365201)
Worked: don't know

Short description of error

Setting editbone.length with Python to a specific value causes the bone's Z axis to flip.

Exact steps for others to reproduce the error
Open an empty .blend. Add an Armature. Turn on Armature > Viewport Display > Axes so you can see the bone axes. Switch to Edit mode.

Run this script

import bpy
ebone = bpy.data.armatures[0].edit_bones['Bone']

ebone.head = [0.0, 0.3999999761581421, 0.35000017285346985]
ebone.tail = [0.0, -0.6000005006790161, 0.34999972581863403]
ebone.align_roll( [1.509958593715055e-07, 1.7881393432617188e-07, 1.0000004768371582] )

The bone's Z axis is pointing up, as expected.

Now run

ebone.length =  0.11045359214368487

The bone's Z axis has flipped and is now pointing down! Now run the same line again. The Z axis flipped again, despite no change to the head/tail/roll! You can keep repeating this.

The pose position is also affected (so it isn't just a Viewport Display issue).

Other length values (eg. 0.11) do not trigger the issue.

Context

Discovered while investigating https://devtalk.blender.org/t/gltf-export-issue-upd-import-issue/18970. The magic numbers here are just the values the importer ends up with when importing that .glb file.

**System Information** Operating system: Linux Graphics card: Intel HD **Blender Version** Broken: 3.0 (specifically I'm testing with f9567f6c63), also 2.83 (latest release fd3036520101) Worked: don't know **Short description of error** Setting `editbone.length` with Python to a specific value causes the bone's Z axis to flip. **Exact steps for others to reproduce the error** Open an empty .blend. Add an Armature. Turn on Armature > Viewport Display > Axes so you can see the bone axes. Switch to Edit mode. Run this script ``` import bpy ebone = bpy.data.armatures[0].edit_bones['Bone'] ebone.head = [0.0, 0.3999999761581421, 0.35000017285346985] ebone.tail = [0.0, -0.6000005006790161, 0.34999972581863403] ebone.align_roll( [1.509958593715055e-07, 1.7881393432617188e-07, 1.0000004768371582] ) ``` The bone's Z axis is pointing up, as expected. Now run ``` ebone.length = 0.11045359214368487 ``` The bone's Z axis has flipped and is now pointing down! Now run the same line again. The Z axis flipped *again*, despite no change to the head/tail/roll! You can keep repeating this. The pose position is also affected (so it isn't just a Viewport Display issue). Other length values (eg. `0.11`) do not trigger the issue. ![flipping_z.png](https://archive.blender.org/developer/F10141842/flipping_z.png) **Context** Discovered while investigating https://devtalk.blender.org/t/gltf-export-issue-upd-import-issue/18970. The magic numbers here are just the values the importer ends up with when importing that .glb file.

Scurest commented

2021-05-25 22:40:55 +02:00

Added subscriber: @scurest

Julien Duroure commented

2021-05-26 07:32:38 +02:00

Added subscriber: @JulienDuroure

Philipp Oeser commented

2021-05-26 13:17:37 +02:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2021-05-26 13:17:37 +02:00

Changed status from 'Needs Triage' to: 'Confirmed'

Philipp Oeser commented

2021-05-26 13:17:37 +02:00

Can confirm this particular example (also the fact that other bone lengths dont give the same issues).

Heramba Limaye commented

2021-05-28 17:03:55 +02:00

Added subscriber: @Azrael69

Scurest commented

2021-05-29 00:21:43 +02:00

I tracked this down to the function vec_roll_to_mat3_normalized .

The formula for the matrix is poorly conditioned when nor is close to the -Y axis, so the code branches and the numbers in this issue are close to the branching threshold. Assigning to ebone.length slightly bumps the values so they cross the threshold.

So I'm now confident this is a dupe of #82455. The suggestion there (testing x*x+z*z>THETA_CRITICAL) fixes this issue.

(For some reason I can't get the "Close as Duplicate" widget to work. Sorry, can someone do it for me?)

I tracked this down to the function [vec_roll_to_mat3_normalized ](https://git.blender.org/gitweb/gitweb.cgi/blender.git/blob/9225fe933ae9904ea7bb646110308a921b37fc86:/source/blender/blenkernel/intern/armature.c#l2155). The formula for the matrix is poorly conditioned when nor is close to the -Y axis, so the code branches and the numbers in this issue are close to the branching threshold. Assigning to `ebone.length` slightly bumps the values so they cross the threshold. So I'm now confident this is a dupe of #82455. The suggestion there (testing `x*x+z*z>THETA_CRITICAL`) fixes this issue. (For some reason I can't get the "Close as Duplicate" widget to work. Sorry, can someone do it for me?)

Philipp Oeser commented

2021-05-29 08:22:23 +02:00

Thx for investigating, @scurest!

Philipp Oeser commented

2021-05-29 08:23:23 +02:00

Closed as duplicate of #82455

Philipp Oeser closed this issue

2021-05-29 08:23:23 +02:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Python: setting editbone length flips bone Z axis #88581