okay, implemented Matrix on top >>> no error anylonger

but the angles are still wrong ill test abit longer tomorrow, don't have time for that now

what i can say:

-> unscaled model (armature fits the model):

if the model stays in blender this way:

[front]--[mid]--[back]

it is ingame this way:

[front]
  |         
[mid]
  |
[back]

-> scaled model by 0.5 (armature twisted by 180°):

if the model stays in blender this way:

[front]--[mid]--[back]
[head-bone]--[torsobone]--[backbone]

it is ingame this way:

[front]   [backbone]
  |         |
[mid]   [torsobone]
  |         |
[back] [headbone]

i tested different formulas and checked result in Tom's XmodelUtils.

Seems to be enough to use an identity matrix for the root bone:

X 1.000000, 0.000000, 0.000000
Y 0.000000, 1.000000, 0.000000
Z 0.000000, 0.000000, 1.000000

Code - No need to import Matrix anymore:

            if i == 0:
                b_matrix = ((1,0,0),(0,1,0),(0,0,1))
            else:
                b_matrix = a_matrix * bone.matrix_local

I'm still unsure about bone rotations... Roll angles seem to play a role when animating, different values may distort models differently. I really need some Maya test files to compare!