Trouble with the resulting size of a solid subtraction.

[Bruce Kieffer]

Has anyone else had this problem, and possibly found a solution? When I subtract a 7/8" wide object inserted 7/16" into a 3/4" thick object, the resulting solid subtraction is .750", it should remain 3/4" thick. I have Units set to Exact as Fractions/Non-Exact as Decimals. It appears to be some corruption in my file, and I have better success with the subtraction if I make sure the objects are on the grid, which should have nothing to do with the resulting thickness. I believe I imported a SketchUp file into this file, and again that should have no affect.

[Benson Shaw]

Bruce,

If object dimension displays as a decimal value instead of as a fraction (because of setting for fractions when exact) could that mean that some component was not exact? Or the placement of one of the objects prior to the subtraction was not exact?  A pixel or two off of an exact snap would trigger the fraction.

 

I wonder what happens if the drawing pref decimal precision is increased to 8 or 9 places and the dimension object precision to same 8 or 9 places, or more.  Maybe that will show some placement or object size is just a little bit off.

 

Or, more likely, Vectorworks is taunting us.

 

-B

[Pat Stanford]

Benson is probably correct. I think we had a LOOONG discussion of this about 10 years ago on the tech board on one of the mail lists. I can't find the information right now so here goes from Memory.

 

The problem is that binary (base 2) floating point math is not exact for many simple decimal (base 10) fractions. That means that what really happens for many numbers is that you get a long approximation of the fraction. You basically have to add fractions of powers of 2 (1/2+1/4+1/8....) skipping the ones that don't work well. When you are just a tiny bit off you end up adding the binary version of 1/2048 or 1/4096 or even higher).  

 

So while VW is rounding the value for display to you to 0.750, internally it may be 0.74999997 or 7.500003 

 

Internally somewhere is is a calculation that determines how many significant figures to use to do comparisons. 

 

It appears that something in your drawing is enough off so that it is not comparing exactly to 0.75 at however many decimal places are used internally.  

 

Something in the back of my brain wants to say it is 11 decimals places, so you don't have to be very far off for things to not match.

 

[Bruce Kieffer]

Let me tell you what I'm doing: I'm cutting biscuit grooves on the edges of a 3/4" thick panel. In the real world that operation does not change the width or length of the 3/4" thick panel. In the Vectorworks world in my file it does change the panel dimension, and that is wrong. I never had this problem before Vectorworks 2017, and I cannot create the problem in a new file, which leads me to believe there's corruption in this file and most likely due to the importation of the SketchUp file.

 

I'm doing some experimentation trying to pin down the issue now. I'll report back shortly and possibly attach a sample file.

[Bruce Kieffer]

I copied and pasted the object that had the bad dimensions into a new file. Then I deconstructed it and reconstructed it using the same components, and in the new file the final dimension is correct. I think this tells me there's corruption in my original file.

Edited January 31, 2017 by Bruce Kieffer

[michaelk]

Bruce

 

Just curious.  Was the biscuit cut out a simple extrude or a crazy fillet object.

 

Accurate to 1/64, of course.

 

mk

[Bruce Kieffer]

I made the biscuit as a 2-1/4" x 7/8" ellipse and extruded it 1/8". Nothing crazy about that!

[Bruce Kieffer]

I'm leaning towards this problem being related to mirrored objects. I've been fighting it all day today. I gave up.

[Bruce Kieffer]

I did some trouble shooting this morning trying to figure out why Vectorworks does not do what I think it should do. I can subtract objects from objects without goofy dimensional results, but when I subtract a symbol from an extrude then I get odd dimensional results. I'm convinced this is a previous bug that was reintroduced in Vectorworks 2017. The sad thing is I feel I spend more time trouble shooting then working to make money. I hope the powers that be are listening.

[Pat Stanford]

Bruce,

 

If you have an example that repeatedly give you the incorrect result, then you need to make sure you submit it. VWI does not have your experience in causing the problem. If they can't cause the problem to happen for them, they can't fix it.

 

Please always submit the steps required to create an issue. Makes sure things get fixed faster.

[Bruce Kieffer]

I sent the file to VWI tech, and they could not figure it out. They said .7499999999999" is the same as 3/4". It's not.

[Bruce Kieffer]

I attached an example file I just created. See the solid subtraction says it's .750" thick, that is wrong. Double click the solid subtraction and you'll see the extrusion is 3/4" thick. If I subtract the bisuicts from the panel the panel should remain 3/4" thick, it does not. This is a problem.

Untitled 1.vwx

Edited February 17, 2017 by Bruce Kieffer

[Marissa Farrell]

Hi Bruce,

 

This will end up being a very controversial conversation, however I want to point out that this likely isn't Vectorworks' fault.

Computers, when doing math, have rounding errors that aren't always handled the way that the user would expect.

 

Hopefully this article can help clear it up, at least a little.

https://blog.codinghorror.com/why-do-computers-suck-at-math/

[AlanW]

@Bruce KiefferHi Bruce, I think the problem lies in your units settings see the attached and it shows as 3/4 after I redid it. HTH     0.750 is 3/4 of an inch also.

Untitled 1.vwx

Edited February 17, 2017 by Alan Woodwell

[AlanW]

@Bruce KiefferHiagain, If you set your Units to fractions only your file will show 3/4 "

Edited February 17, 2017 by Alan Woodwell

[Bruce Kieffer]

Alan,

This appears to be a Mac problem. It still shows .750" when I have the settings as you do on your PC. I want my settings exact as fraction, non-exact as decimals for the very reason the setting exists. I want to know when the number is not an exact fraction. Like I said before, this was a problem a few VW versions ago, and now it's back.

[Bruce Kieffer]

This shows the error where I increase the number of decimals. 3/4" - 0" is 3/4", not .7500000004

 

Edited February 17, 2017 by Bruce Kieffer

[AlanW]

@Bruce KiefferHere's the file try it the one on right should show 3/4 " Also you need to show fractions only . in the Units, you dont have that checked.

Untitled 1_AW.vwx

Edited February 17, 2017 by Alan Woodwell

[Bruce Kieffer]

I see the one on the right is 3/4", but when I move the biscuit and have it cut a groove somewhere else in the panel, then it's back to not 3/4". VW Tech needs to look at this file and find a fix.

[AlanW]

@Bruce KiefferBruce I see the issue, each panel appears to hold the Units settings that you used originally. Double click on the other units and go units and you will see it is selecting the bottom one. select the fraction one and it will be ok.Generaly you need to have nothing selected then set units. Do each of the 2 on the left and they should stay 3/4 "

[PVA - Admin]

On 1/30/2017 at 10:56 PM, Pat Stanford said:

Benson is probably correct. I think we had a LOOONG discussion of this about 10 years ago on the tech board on one of the mail lists. I can't find the information right now so here goes from Memory.

 

The problem is that binary (base 2) floating point math is not exact for many simple decimal (base 10) fractions. That means that what really happens for many numbers is that you get a long approximation of the fraction. You basically have to add fractions of powers of 2 (1/2+1/4+1/8....) skipping the ones that don't work well. When you are just a tiny bit off you end up adding the binary version of 1/2048 or 1/4096 or even higher).  

 

So while VW is rounding the value for display to you to 0.750, internally it may be 0.74999997 or 7.500003 

 

Internally somewhere is is a calculation that determines how many significant figures to use to do comparisons. 

 

It appears that something in your drawing is enough off so that it is not comparing exactly to 0.75 at however many decimal places are used internally.  

 

Something in the back of my brain wants to say it is 11 decimals places, so you don't have to be very far off for things to not match.

 

 

51 minutes ago, MarissaF said:

Hi Bruce,

 

This will end up being a very controversial conversation, however I want to point out that this likely isn't Vectorworks' fault.

Computers, when doing math, have rounding errors that aren't always handled the way that the user would expect.

 

Hopefully this article can help clear it up, at least a little.

https://blog.codinghorror.com/why-do-computers-suck-at-math/

 

The two above posts are where this issue ends unfortunately. This can not be corrected in any meaningful way in Vectorworks. If engineering was forced to act on it, most likely they would simply take away the dimensional fields in the OIP for objects that had questionable results at that incredibly small scale and ensure that the dimensions always read correctly instead. I don't like it, but it isn't going to change anytime soon, keeping geometry near or appropriate for the the full unit size in dimensioning is the only way to avoid it.

[AlanW]

Just think of all the money transactions handled by computers?????? mm wonder who is making money and who is losing money as these little bits add up to Million over time.

[Pat Stanford]

I don't agree with this. A better option for engineering would be a user settable option for absolute. If a user could set the accuracy they are willing to accept for an absolute value, things would be much better.

 

With Bruce doing wood working, an absolute value of 1000th would be fine. So if the returned value from a calculation wast 0.74999, it would round up and display as an absolute of 3/4 inch. He does not need an accuracy of 11 decimal places. 3 2ould be enough. 

 

VW should be able to use a user settable decimal to fraction conversion and handle this all internally.

 

JM2¢W

 

Pat

 

[Kevin McAllister]

I think this is why programs like Rhino have different templates for different situations (image below). Even C4D has this type of adjustment.

It will become more and more of an issue as more users use VW for digital manufacturing. When I work on events, the tolerance can be 1" or more. When I work on jewellery, I'm working to tolerances of 0.1 of a mm or even less. 3D printing requires these kinds of tolerances.

 

Kevin

 

[Bruce Kieffer]

52 minutes ago, Pat Stanford said:

I don't agree with this. A better option for engineering would be a user settable option for absolute. If a user could set the accuracy they are willing to accept for an absolute value, things would be much better.

 

With Bruce doing wood working, an absolute value of 1000th would be fine. So if the returned value from a calculation wast 0.74999, it would round up and display as an absolute of 3/4 inch. He does not need an accuracy of 11 decimal places. 3 2ould be enough. 

 

VW should be able to use a user settable decimal to fraction conversion and handle this all internally.

 

JM2¢W

 

Pat

 

Pat is absolutely correct with this answer. I need to know when I'm off by 1/32, not by .0000000001.

Edited February 17, 2017 by Bruce Kieffer

[PVA - Admin]

Correct, if the document units were set so that a whole number in those units was within the realm of the accuracy needed, then it would most likely not be an issue. if you need to measure down to the accuracy of a percentage of the thickness of a strand of hair (effectively the size of the inaccuracy being discussed here) then inches and 16ths of an inch are too large a unit to do so.

We don't force it like other applications do, they just automatically round the value the user is shown to make them feel better (or automatically change to a smaller unit of precision), the error still exists underneath. I would say letting the user set an absolute rounding value would also make sense, but it's still a "dishonest" value underneath. I'll broach the topic with the directors and see what they think, this is likely not something that's come up for them in awhile according to a quick internal tracking and forum search.

EDIT: However, to clarify, wouldn't asking for manual control over the rounding be giving the user the same control as they currently have when setting the Fraction precision to 1/64 and not showing decimals for inexact fractions?