The edit_assigned_product() function now removes the annotation from all
existing IfcRelAssignsToProduct relationships instead of just one, preventing
annotations from being incorrectly assigned to multiple products simultaneously.
This fixes issues where leader lines wouldn't update correctly because the
annotation retained old product assignments.
Previously, when multiple CSS selectors were comma-separated in a rule,
only the last selector received the unique drawing ID suffix. This caused
style conflicts when multiple drawings were placed on the same sheet.
Now all selectors in comma-separated lists receive the unique suffix,
ensuring proper style isolation between drawings.
Example:
Before: .cut.Status-DEMOLISH, .cut.Option-D.d2156 { ... }
After: .cut.Status-DEMOLISH.d2156, .cut.Option-D.d2156 { ... }
In Blender 5.0 some theme attributes were removed and moved to other places and code broke.
In the commit I've also switched the code to less dynamic approach, so static type checker could caught potential errors.
`get_active_stair_calculated_params` wasn't updated when zero tread runs support was added and it's still was relying on a convention that `0.0` custom tread run = no custom tread run.
`s0 = V_(custom_first_last_tread_run[0] or tread_run, tread_rise) + td_vector` was assuming `0` means "no custom first tread run", not it means actually "zero first tread run", so could should be adjusted
before - https://files.catbox.moe/5ovqnh.mp4
after - https://files.catbox.moe/srorl2.mp4
Before f52aafd default `(0,0)` meant that there are no custom tread runs, but after the change it actually means that runs for first and last treads are 0. So default value should be adjusted.
Added explicit `None` value for clarity.
Since it's not illegal in Blender and can be helpful - e.g. when you were assigning class to ifc space, you would lose selection completely and would need to reselect object to continue working on it.
After 3aa5ddb07 `select_products` adds to current selection, instead of resetting it.
So couple tests started to fail since they create new objects and `view_layer` is not yet updated (previously `bpy.ops.object.select_all(action="DESELECT")` was also implicitly updating view layer).
```
FAILED test/tool/test_spatial.py::TestSelectProducts::test_select_products - AssertionError: assert bpy.data.objects['Object'] in []
FAILED test/tool/test_system.py::TestSelectSystemProducts::test_run - AssertionError: assert bpy.data.objects['Object'] in []
```
Long story short - since 5.0 `np.array(Vector())` is now producing `np.float32` instead of `np.float64`. So we have to provide `dtype` explicitly to support both <5.0 and >= 5.0.
See https://projects.blender.org/blender/blender/issues/149283
It's probably will be very popular bug, when someone would try to migrate to Blender 5.0 using old Bonsai.
Example traceback:
```
File "Blender\5.0\extensions\.local\lib\python3.11\site-packages\bonsai\bim\prop.py", line 34, in <module>
import bonsai.bim.schema
File "Blender\5.0\extensions\.local\lib\python3.11\site-packages\bonsai\bim\schema.py", line 23, in <module>
import bpy_restrict_state
ModuleNotFoundError: No module named 'bpy_restrict_state'
```
This also deprecates one more instance of the add_representation API
call. As this is slowly cleaned up we can move towards "regenerating"
annotations, not just generating from scratch.
Thomas Krijnen