PointerProperty in UI slows down extremely #91387

New Issue

Matthias Thurau · 2021-09-13T20:58:39+02:00

Matthias Thurau commented

2021-09-13 20:58:39 +02:00

System Information
Windows 10

Blender Version
Blender 2.83.17

Short description of error
Creating a python script that utilizes bpy.props.PointerProperty suddenly slows down when rendered in the UI. This happens proportional to the amount of elements in a custom CollectionProperty in bpy.types.scene.

Exact steps for others to reproduce the error

Run the script attached
You should see a two categories in your 3d viewport.
The category UI called "INT PROPERTY" has normal behavior and speed.
the category UI called "POINTER PROPERTY" is extremely slow.
(change the amount of items in line 161 of the script to slow down even more.)

This is a simplified version of the problem that occours for a huge addon developed in my company. We have hundreds of items in the bpy.context.scene.CollectionProperty as well as tens of UIs with tens of PointerPropertys. We are running into huge performance problems.

bug_pointer_property.py

**System Information** Windows 10 **Blender Version** Blender 2.83.17 **Short description of error** Creating a python script that utilizes bpy.props.PointerProperty suddenly slows down when rendered in the UI. This happens proportional to the amount of elements in a custom CollectionProperty in bpy.types.scene. **Exact steps for others to reproduce the error** - Run the script attached - You should see a two categories in your 3d viewport. - The category UI called "INT PROPERTY" has normal behavior and speed. - the category UI called "POINTER PROPERTY" is extremely slow. (change the amount of items in line 161 of the script to slow down even more.) This is a simplified version of the problem that occours for a huge addon developed in my company. We have hundreds of items in the bpy.context.scene.CollectionProperty as well as tens of UIs with tens of PointerPropertys. We are running into huge performance problems. [bug_pointer_property.py](https://archive.blender.org/developer/F10409601/bug_pointer_property.py)

Matthias Thurau commented

2021-09-13 20:58:39 +02:00

Added subscriber: @Zweistein

Pratik Borhade commented

2021-09-14 09:20:43 +02:00

Added subscriber: @PratikPB2123

Philipp Oeser commented

2021-09-14 15:27:16 +02:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2021-09-14 15:27:16 +02:00

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

Philipp Oeser commented

2021-09-14 15:27:16 +02:00

Can confirm it is extremely slow, seems like it spends most of its time trying to figure out if the button is animated:

Usually performance issues are not considered bugs, but I will confirm this for now and see if #user_interface or #animation_rigging can classify this.

Can confirm it is extremely slow, seems like it spends most of its time trying to figure out if the button is animated: ![image.png](https://archive.blender.org/developer/F10412563/image.png) Usually performance issues are not considered bugs, but I will confirm this for now and see if #user_interface or #animation_rigging can classify this.

Philipp Oeser commented

2021-09-14 16:09:03 +02:00

Added subscribers: @mont29, @dr.sybren

Philipp Oeser commented

2021-09-14 16:09:03 +02:00

@mont29, @dr.sybren : Can the whole BKE_animdata_driver_path_hack from 8cb1b35bee go? (since this seems to be from times where there could be FCurves outside AnimData)

I also dont quite get how (all of a sudden) AnimData can be non-NULL:

we get NULL from BKE_animdata_from_id https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$428
we call BKE_animdata_driver_path_hack
why should it ever be non-NULL afterwards? (BKE_animdata_from_id is called again after BKE_animdata_driver_path_hack)
if it stays NULL, all the rest of BKE_fcurve_find_by_rna_context_ui will be skipped anyways...

If the BKE_animdata_driver_path_hack stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly):
P2392: T91387_snippet



diff --git a/source/blender/blenkernel/intern/fcurve.c b/source/blender/blenkernel/intern/fcurve.c
index 8e9c504dcbf..376dfdd86bf 100644
--- a/source/blender/blenkernel/intern/fcurve.c
+++ b/source/blender/blenkernel/intern/fcurve.c
@@ -431,11 +431,13 @@ FCurve *BKE_fcurve_find_by_rna_context_ui(bContext *C,
         C ? 2 : 1);
     char *path = NULL;
 
+    /*
     if (!adt && C) {
       path = BKE_animdata_driver_path_hack(C, &tptr, prop, NULL);
       adt = BKE_animdata_from_id(tptr.owner_id);
       step--;
     }
+    */
 
     /* Standard F-Curve - Animation (Action) or Drivers. */
     while (adt && step--) {

@mont29, @dr.sybren : Can the whole `BKE_animdata_driver_path_hack` from 8cb1b35bee go? (since this seems to be from times where there could be FCurves outside AnimData) I also dont quite get how (all of a sudden) AnimData can be non-NULL: - we get NULL from `BKE_animdata_from_id` https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$428 - we call `BKE_animdata_driver_path_hack` - why should it ever be non-NULL afterwards? (`BKE_animdata_from_id` is called again after `BKE_animdata_driver_path_hack`) - if it stays NULL, all the rest of `BKE_fcurve_find_by_rna_context_ui` will be skipped anyways... If the `BKE_animdata_driver_path_hack` stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly): [P2392: T91387_snippet](https://archive.blender.org/developer/P2392.txt) ``` diff --git a/source/blender/blenkernel/intern/fcurve.c b/source/blender/blenkernel/intern/fcurve.c index 8e9c504dcbf..376dfdd86bf 100644 --- a/source/blender/blenkernel/intern/fcurve.c +++ b/source/blender/blenkernel/intern/fcurve.c @@ -431,11 +431,13 @@ FCurve *BKE_fcurve_find_by_rna_context_ui(bContext *C, C ? 2 : 1); char *path = NULL; + /* if (!adt && C) { path = BKE_animdata_driver_path_hack(C, &tptr, prop, NULL); adt = BKE_animdata_from_id(tptr.owner_id); step--; } + */ /* Standard F-Curve - Animation (Action) or Drivers. */ while (adt && step--) { ```

Philipp Oeser commented

2021-09-14 16:16:44 +02:00

In #91387#1219956, @lichtwerk wrote:
If the BKE_animdata_driver_path_hack stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly):

Well, at least when it is not animated/driven ;)
As soon as that is the case, we get the similar bottleneck a bit further down [that is already commented to be slow -- also needs to call RNA_path_from_ID_to_property], but at least we are not calling it twice!
https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$447

> In #91387#1219956, @lichtwerk wrote: > If the `BKE_animdata_driver_path_hack` stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly): Well, at least when it is not animated/driven ;) As soon as that is the case, we get the similar bottleneck a bit further down [that is already commented to be slow -- also needs to call `RNA_path_from_ID_to_property`], but at least we are not calling it twice! https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$447

Hans Goudey commented

2021-09-21 06:22:20 +02:00

Added subscriber: @HooglyBoogly

Matthias Thurau commented

2021-12-20 20:36:15 +01:00

I really hoped and tested this bug with Blender 3.0.0 and I have to confirm that it is sadly still there.

Is there a known workaround?

I really hoped and tested this bug with Blender 3.0.0 and I have to confirm that it is sadly still there. Is there a known workaround?

Matthias Thurau commented

2021-12-20 20:50:37 +01:00

Yes, cool, there is a workaround:
If i change the pointed-to IntProperty "options", then it seems to be fast:

#create property group
class MyPointerProperty(bpy.types.PropertyGroup):
    my_int_property: bpy.props.IntProperty(options={'HIDDEN'})

The default is options={'ANIMATABLE'} as per documentation.

Yes, cool, there is a workaround: If i change the pointed-to IntProperty "options", then it seems to be fast: ``` #create property group class MyPointerProperty(bpy.types.PropertyGroup): my_int_property: bpy.props.IntProperty(options={'HIDDEN'}) ``` The default is options={'ANIMATABLE'} as per documentation.

Sybren A. Stüvel commented

2021-12-21 17:06:46 +01:00

In #91387#1276064, @Zweistein wrote:
I really hoped and tested this bug with Blender 3.0.0 and I have to confirm that it is sadly still there.

The task is still open, and not closed as "Resolved". There is no reason to think it's been resolved already.

> In #91387#1276064, @Zweistein wrote: > I really hoped and tested this bug with Blender 3.0.0 and I have to confirm that it is sadly still there. The task is still open, and not closed as "Resolved". There is no reason to think it's been resolved already.

blender-admin commented

2021-12-23 13:49:58 +01:00

This issue was referenced by 025c921416

This issue was referenced by 025c9214165ca9e43d3c281c326bf4887c32d642

Sybren A. Stüvel commented

2021-12-23 14:16:39 +01:00

In #91387#1219956, @lichtwerk wrote:
Can the whole BKE_animdata_driver_path_hack from 8cb1b35bee go?

Yes! It's gone since 025c921416 :)
It's replaced with a call to RNA_path_from_ID_to_property(), though, so the code isn't going to be any faster. It's much clearer, though.

I also dont quite get how (all of a sudden) AnimData can be non-NULL:

we get NULL from BKE_animdata_from_id https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$428

we call BKE_animdata_driver_path_hack

why should it ever be non-NULL afterwards? (BKE_animdata_from_id is called again after BKE_animdata_driver_path_hack)

if it stays NULL, all the rest of BKE_fcurve_find_by_rna_context_ui will be skipped anyways...

Some years ago BKE_animdata_driver_path_hack() still did a lot more, and could probably have caused the anim data to appear.
Now, though, RNA_path_from_ID_to_property() it's not changing tptr so the NULL pointer stays.

If the BKE_animdata_driver_path_hack stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly):
P2392: T91387_snippet



diff --git a/source/blender/blenkernel/intern/fcurve.c b/source/blender/blenkernel/intern/fcurve.c
index 8e9c504dcbf..376dfdd86bf 100644
--- a/source/blender/blenkernel/intern/fcurve.c
+++ b/source/blender/blenkernel/intern/fcurve.c
@@ -431,11 +431,13 @@ FCurve *BKE_fcurve_find_by_rna_context_ui(bContext *C,
         C ? 2 : 1);
     char *path = NULL;
 
+    /*
     if (!adt && C) {
       path = BKE_animdata_driver_path_hack(C, &tptr, prop, NULL);
       adt = BKE_animdata_from_id(tptr.owner_id);
       step--;
     }
+    */
 
     /* Standard F-Curve - Animation (Action) or Drivers. */
     while (adt && step--) {

Yup, nuke it!

> In #91387#1219956, @lichtwerk wrote: > Can the whole `BKE_animdata_driver_path_hack` from 8cb1b35bee go? Yes! It's gone since 025c921416 :) It's replaced with a call to `RNA_path_from_ID_to_property()`, though, so the code isn't going to be any faster. It's much clearer, though. > I also dont quite get how (all of a sudden) AnimData can be non-NULL: > - we get NULL from `BKE_animdata_from_id` https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/fcurve.c$428 > - we call `BKE_animdata_driver_path_hack` > - why should it ever be non-NULL afterwards? (`BKE_animdata_from_id` is called again after `BKE_animdata_driver_path_hack`) > - if it stays NULL, all the rest of `BKE_fcurve_find_by_rna_context_ui` will be skipped anyways... Some years ago `BKE_animdata_driver_path_hack()` still did a lot more, and could probably have caused the anim data to appear. Now, though, `RNA_path_from_ID_to_property()` it's not changing `tptr` so the `NULL` pointer stays. > If the `BKE_animdata_driver_path_hack` stuff can go, the example script from this report gets super snappy again (and drivers still seems to work properly): > [P2392: T91387_snippet](https://archive.blender.org/developer/P2392.txt) ``` diff --git a/source/blender/blenkernel/intern/fcurve.c b/source/blender/blenkernel/intern/fcurve.c index 8e9c504dcbf..376dfdd86bf 100644 --- a/source/blender/blenkernel/intern/fcurve.c +++ b/source/blender/blenkernel/intern/fcurve.c @@ -431,11 +431,13 @@ FCurve *BKE_fcurve_find_by_rna_context_ui(bContext *C, C ? 2 : 1); char *path = NULL; + /* if (!adt && C) { path = BKE_animdata_driver_path_hack(C, &tptr, prop, NULL); adt = BKE_animdata_from_id(tptr.owner_id); step--; } + */ /* Standard F-Curve - Animation (Action) or Drivers. */ while (adt && step--) { ``` Yup, nuke it!

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

PointerProperty in UI slows down extremely #91387