Imperfect objects bug
-
Hello Xrok1,
Thank you for your answer but I don't think that you have read my post very well. I do realize how small that is. However, that is not what matters here.
If I change the precision to 3 decimal places (Like I did for a long time) I still have these little numbers at the end when I measure something afterwards with more precision. (It's always there somewhere) And if it wouldn't cause any problems I wouldn't post a topic about it, but the thing is, it does cause problems.
Some examples of problems:
- Suddenly you realize that it is not possible to make an object smaller (lets say a cube) by using the push/pull tool because all sides of the cube are not perfectly equal.
- When you draw another object against a cube with unequal sides, this second object will probably also have an unequal side. As a result, your model gets "infected".
- If you have another object that HAS all equal sides and you want to place it against an object with unequal sides, you can NOT actualy perfectly join these two and this also gives problems afterwards.
(And apart from all of this: It just should not happen. Period.)
-
Not knowing a detailed step sequence on how you get the problem my guess is you are doing some thing allowing these errors to creep in even though it is not intentional. The screen shot indicates your model is complex and if you are not careful you can snap to some thing you are not aware of. I would recommend the following: Turn length snapping off, make sure you use the VCB input to set your lengths vs using any visual ques unless you are sure you are snapping to the intended point. Are all the layers you have shown based on components or groups. If they are not you can snap to geo in them even though it is hidden. Make sure you use components or groups to isolate the geo
Just some thoughts. Good Luck -
Thanks for posting about this problem, I've encountered it too for several times.
I don't know if it's really related thing, but here's a little tip that seems to help to avoid at least part of these inaccuracies: when you're drawing a line with inference, always keep the starting point in the viewport (or lock axis with Shift before changing camera position). For some reason axis snapping can be slightly inaccurate if you've moved your camera away from the vertex. It's almost unnoticeable, but leads to these 0.0000* differences you're probably talking about.
D-oh, since even I can't clearly understand what I wrote about, use the illustration, I draw a bit better.
Again, maybe it's not the reason or just a part of it, but it's all I know.
-
Thanks for the advice. The way I am solving it now is to set the precision value always at the highest value (6 decimal places) so I can always check if the lenght of a line is correct. I also, like you are saying, try to use the VCB input as much as possible and I always use groups and components, never loose geometry.
Turning length snapping off is something I still can try to do. I am not sure if it will work because I cannot reproduce the error. I don't know exactly in which circumstances it happens. Thanks!
That's an interesting idea! The only thing is, I have tried to reproduce it, but it didn't "work" for me. Are you able to reproduce the problem this way?
-
I am speculating that inaccuracies creep in due to the unavoidable use of floating point math. Does setting permissions to 3 instead of 6 places affect how results are displayed, or how results are calculated, or both?
-
Setting it to 3 decimal places doesn't help. The little inaccurancies still happen. The only thing with 3 decimal places is that you can not see them. You will see them when you set the precision to 6 places.
-
I have contacted google about the issue and this was their last reply:
@unknownuser said:
Thank you for the extra details. I just chatted with some of the
Engineering team, and it sounds like the problems you've described (extra
surfaces, problems snapping or using Push/Pull) are related to the width
of the model. Basically the larger the model gets the less precise
SketchUp can be with each individual measurement.My suggestion would be to create many smaller components for different
sections of your model. For example each of the side towers and the middle
tower would all be separate components. Since you can save components as
distinct SketchUp models, you could work on each model independently, thus
avoiding the scale issue. When you go back to your big model you can just
updated that component reference and it will refresh with the changes you
made in the separate file. This is similar to working with XREFs in a CAD
application.Hopefully that makes sense and will allow you to work with a wide model
while still maintaining smaller details.Does anyone know how to work with these "component references" (if that is the right term)? I have searched for it on google, but I wasn't able to find it.
-
Your model units 'accuracy' reporting don't need to be set to anything more than ~1mm.
After all you are modeling a building NOT an electron-microscope - you will be lucky if ANY building in the real world is built to within ~10mm - let alone the thousandths of a mm you have you model's unit reporting set to! There are always rounding errors in ANY 3D modeling application - just like when you measure with a tape in the real world... but worrying about thousandths of a mm is just ridiculous!
Using the built-in typed in lengths from the VCB, and the inferencing and axis locking facilities means than super-accuracy is not important!
So, use 'Model Info > Units' - and change its settings to something more sensible [as explained in the earlier post] - just fix it and stop whining!
It's late on a Friday and we do have real lives to go to as well as helping you... 
Using components to order the way you model is a different issue.............
PS: Sorry to crack at the end of a long and hard week
-
Hi TIG,
The end of a long and hard week or not; I don't like it.
First of all: I am not stupid. The bigest units I use in my model are centimeters, so I don't care about this kind of accuracy. The only thing is: I was thinking (and I am still thinking) that these little differences can cause problems with pushing/pulling, etc. I was looking for an answer and I have learned two things:
-
The little differences and the problems are caused by the big scale of the model.
-
In order to avoid these problems caused by the big scale I can work with linked components outside the main model itself.
-
-
BUT... if you set your reported unit's 'accuracy' in Model Info > Units to be 1mm [or 0.1cm if you work in cm] [and the 'length snapping' to the same value], then you won't get these reported dimensions like 12.3450000001m you'll get 12.345m - so what is the problem?
If you use the built-in inferencing and axis locking to move or pushpull shapes etc they will all appear to be the same size as far as you are concerned in the reporting of them and their dimensioning - even if they are in fact some thousandths of a mm 'out'.
It shouldn't matter to you...
If you are happy to subdivide you model into linked SKPs then fine... -
That was not the case. The little differences were there before I set the preciion to 6 decimal places. I have only set it to 6 decimal places when I was trying to find the source of the problems. Then I discovered the little differences. I normaly do not set the precision to 6 decimal places. (Why would I, when I am working on a model with big buildings etc.?)
-
You just don't (want to:-) get it. Computer math, floating point arithmetic, has a rounding problem that results in inaccurate models.
-
Ok, fine, that's interesting, I beleive you, but that changes nothing to what I've said. On top of that: Why would I be not allowed to post a topic about this? I was just looking for an answer and I have found one.
-
You are clearly allowed to post a topic on this... as here we are
What is the exactproblem? [which perhaps you haven't explained clearly enough?], e.g. "There appear to be small inaccuracies in the size of some elements in my model. These seem to be affecting PushPull."...
I think that if you have your units set to display AND snap to an accuracy of 1mm, AND you use typed-in dimensions in the VCB, or snaps to existing geometry using inferences or axial locking, then these should not normally be a problem in the geometry that's made. Your Google 'expert's advice regarding your model size is a red-herring! OpenGL [and therefore Sketchup] do not like very small or very large dimensions... BUT from what I have seen of your project there are no 'very large' dimensions - 100m is OK, >1000m is probably bad! [at the other end 10mm is OK, <1mm is bad!]... Is the nature of your project to make things kilometres across ? If not, then a model few 'blocks' across should not cause you this kind of problem... if you employ good modeling techniques... -
I have often wondered if these errors are more frequent with people who use metric dimensions, due to the extra math required, as internally SU is based on inches.
Anssi
Advertisement