I’m being driven mad by “errors” in Sketchup inferencing. It seems with every version since 2017 it gets worse and so with 2024 being even worse yet, I’m now thinking it might be my problem not understanding something in Sketchup. In any case, the problem I have is that my models are often “corrupted” because at some point I grab what appears to be an intersection of two visible edges, but Sketchup inferences off some hidden turd behind a face that has nothing to do with what I see or what I’m expecting. The result is that I move some group and drop it, but since the group was picked up wrong it gets dropped with the error being the distance from the target intersection and the turd that captured the inferencing war. This is often a tiny error that can’t be seen until you’ve worked the error into 100 places in your model when you discover it.
I’m attaching a video and the model that was used in the video. If anyone could tell me how to get inferencing to get back to the 2017 days of picking the corner you actually want, I would appreciate any tips.
This is the simple model used in the video: test.skp (7.1 MB)
This is the video (on Dropbox because it is too big for the forum):
Wow, that was quick. Thanks!! But I’m not sure I’m following your idea. First, I might do 1000 moves in a day, so if the recommendation is to move the camera for every move, that seems the same as using construction lines (which do inference correctly).
But let me ask if you are on the same issue as the main point of the video. My problem is the inference on the intersecting lines in the upper corner of the square group after it has been changed in size (regardless of how I changed the size). The inference point is picking up on the corner behind a face? Is that not happening when you open the model?
That’s not what I’m recommending but you do need to orbit sometimes. Moving objects in SketchUp works better when you aren’t trying to move directly toward or away from the camera. The closer you are to that the more likely you’ll have issues with inferencing.
I made a post earlier today saying 2024 inference system was erratic, showing a comparison between 2022 and 2024 on the same model. I’ve been using 2024 since it came out and regularly have troubles manipulating my models because of that, without knowing exactly why it is acting this way. I always manage to get back on my feet but it’s a great loss of time if they were laid end to end.
Thank you for your video that might give some clues to the Devs and let’s hope it’ll be fixed soon
Don’t forget that you can orbit, pan, zoom, change scenes, etc even in the middle of any action like drawing a geometry, moving, rotating, scaling, etc any object.
We have a meeting tomorrow to talk about tools related things. I can show this problem at the meeting.
It looks like this issue came in with 2023. I made a video, showing the behavior difference between 2022 and 2023. The thing to notice is that when the other component gets highlighted, in 2023 and 2024 the lock onto the intersection gets lost, and the inference switches to whatever is nearby. With 2022 the other component gets highlighted, but it keeps the lock on the intersection.
Yes, but I’m just trying to pick a point for a move! That’s all I want is to pick at a reference point that doesn’t select from behind a surface. I should not have to move the camera for that unless I am too far out of zoom.
Assuming I am at a reasonable zoom to pick an inference point and my target is two edges that intersect, Sketchup should never find an inference point behind a face, period.
Your video is much better than mine. I couldn’t get in to zoom as close as that! Great and thank you for responding.
It is just my opinion, and perhaps there is a use case, but Sketchup should never inference on something that is behind a face. Never! If it does do that, it should turn on that momentary x-ray mode that I see when I’m moving something and the drag point goes behind the group being dragged. That would show what it is actually inferencing on and would be acceptible.
I wonder then if there isn’t some setting that changes the behavior then. Colin’s video is better than mine showing the difference between 2022 and 2023. So others are seeing the same behavior.
Do agree with me on one point: if you have a screen and the only thing you can see on the screen are two edges and those edges intersect, there should be only one possible inference point in that view. Are you in agreement with that? If it doesn’t happen for you then that is great, but if it does happen for me and others, then there is something wrong with a setting or with inferencing in general.
Watch Colin’s video and you’ll see the inference picking up behind a face very clearly.
I just watched your post. It was excellent and reveals what is driving me mad. When it is erratic like you showed, it puzzles me, but occasionally I don’t see that it is being erratic and I think I’ve got an inferenced point as intended. But, I’ve got a nearby unintended point which then can get errors in the model.
I know it’s difficult to grab the right inferencing “point” and things may have changed.
But in your case there seems to be hope.
SketchUp gives priority to the nearby endpoint… unless…
If you change SketchUp’s focus to the intersection it will be easy to move the slanted block object to the desired target. See attached images: before and after moving the slanted block object.
select the ‘Move’ tool and hover over the yellow edge > “On Edge in Group” > now press and hold down [Shift]
hover over the green edge > “Constrained on Line Intersect Line” > click anywhere on this edge, while still holding down [Shift] (you now have the object “attached” to the cursor to be moved)
start moving the slanted block. It will constrain to the yellow direction, grabbed by intersection yellow/green
click anywhere on the blue vertican edge of the other object (except for its edge midpoint), thus dropping the grabbed intersection on the blue line at the intersection yellow/blue
result is measurement 0.00000
Long story but in fact just “a few key strokes” and accurate.
I find inferencing and selecting works much better if you put SketchUp into transparent mode or whatever it’s called. I guess it stands to reason that SketchUp needs to see what its inferencing or selecting the same as we do.
I got that close by playing back my screen recording at a large size, and screen recording that.
One thing to know, if you turn on back edges (tap the K key), which you would do when actually wanting to infer to a hidden edge or corner, it makes the intersect case work a lot better.
I tried that and at least you can see better what it is doing, but how did the meeting go?
I think Paul Millet’s example generalizes the problem more and shows clearly the difference from the previous version. Are we going to see an attempt to fix this?
Some of what Paul showed is a different issue, and we have that fixed already. In a build that has that fix in it, I still see the problem you’re having.
The meeting went well. The most senior developer for the team that would fix this, could see the problem, was intrigued, and is looking into what we had changed with 2023 that introduced the problem.
A product manager also happened to be in the meeting, could try the test for himself, and agreed it was a bug.
The timing is also good. We occasionally spend a two week period dedicated to fixing existing issues, and not only spending time on new features. One of those is coming up, and if the problem isn’t fixed before then, it hopefully will be looked at during those two weeks.
Thanks to all for the work around and to the SketchUp team for getting it fixed.
I’ve been getting this for ages too.
And I can’t possibly make accurate drawings without for example orbiting and zooming in on two end points which are often distant from one another and sometimes on the other side of a plane or body. The use of views and the like cancels the original crowning benefit of using SketchUp in the first place: its ease of navigation and tools designed for use in the 3d environment.
In the end I often have to put up with unflattened faces and smooth them etc just to get the job done.
FYI my work is architectural in the heritage field and nothing is simple from a geometric point of view.
