This needs to be fixed. I did a test. I loaded a character into an empty scene and did a simple cycle of operations: I selected and deselected the object 10 times. It took 23 seconds to undo to the initial state in 2.8. Same operation in 2.79 takes 3.5 seconds.
Just jumping in here to confirm it s a major issue for production work. I’ve been using blender for small home project and the slow undo is rarely noticeable but it seems to be directly linked to scene file to me (and where the scene is stored). As soon as I work on production on a large set with lots of high poly assets I can start counting seconds. On one scene I have here of 2gb my undo time is 17 secs. That’s working form local hard drive , as it’s slower even from network. The issue starts to be obvious from a file size of a few hundred megs I’d say. Undo in edit mode is fine (but node grease pencil draw mode :(, or object mode ect… ). It’s really annoying as undo is something that you use without thinking specially when switching softwares and for me it’s the main thing that would make me think twice about usuing blender for a full production process. Right now I can already see myself feeling very uneasy if I undo by mistake during an asset review while everybody waits and makes small talk . How can we put this issue in the spotlight? Any ideas?
I still believe that if instead of the current undo, they put on a database that records only the values of the objects moved (not necessarily also vertices in edit mode, because I noticed that the undo is not too problematic in edit mode ) they would solve this problem in a flash.
So I just tested the patch implemented in master last week that should help with undo, but it’s still super slow, check the video, I just moved an object, nothing more, the undo for this must be instant to avoid breaking the speed of work.
I finally got around learning blender and actually liking a lot about it! But now after having worked on a bigger project with it, i am actually considering if i should get a modo liscence or something. Undo is actually unusable - This should be the #1 development issue
This is exactly the problem I am facing right now. As long you don’t go for something serious/highpoly you don’t have a problem, but as soon the project gets more complex every undo just nerves. . Beside of this blender is great. There are still issues in subd performance and multires but undo is the biggest stone on the way for my taste.
Blender 2.83 alpha is out with supposed undo improvements but this is no different from other builds
This should be priority number 1 in first 2.8 release.
undo experiments is in a parallel branch under development, and the results obtained are already exciting. undo-experiments-swap-reread-datablocks
it is likely that we will see the new undo system in blender 2.83 …
it’s just a matter of figuring out if it will be solid enough for this release
Current ‘final’ branch is called uuid-undo-experiments-swap-reread-datablocks, and is almost master-ready (besides some known random crashes happening in depsgraph that need to be fixed).
It is still planned for 2.83, but as an experimental feature (to be enabled in preferences), since it is very unlikely we can harden it enough to enable it by default in this release.
Please also note that using this feature in its current state will expose you to rather severe risks (most common issues will be crashes, but there could also be some .blend file corruptions in some corner cases).
Great! Is it possible that since 2.83 will be LTS relase it will be non-experimental or on by default at some point? I mean, that way future new users will see a better performance without having to look for some hidden setting.
I just tested it, I did the same test as the last time, it’s the same scene of the video a few threads before and I did the same thing, the result it’s pretty good!
Finally… moving an object does not mean a super sloooow undo,I hope this can be stabilized because this is the kind of performance needed for a good undo system.
I would not say a hard ‘NO’, but i honestly think this is relatively unlikely to happen. In the end it all depends on how involved and risky will be the fixes for issues found after 2.86 release… LTS aims at stability over all.