Sketchup is Inacurrate???

lapx · 2 Apr 2012, 17:21

I think sketchup has it backwards. Like I said kinda goofy if you ask me. The "one's and zero's need to be reordered .

jeff hammond · 2 Apr 2012, 17:27

Wait, I think I know this one. It's ......hm.... It's 9 (by inferencing)
Unless you use input "1" when offsetting.

ha Gerrit.. you're killing me here.. (but i suppose this is what TIG was talking about doing in his earlier example.. he had 'desired offset marker' in there so he was using that as an inference point)

so go ahead and inference it in 3D.. with the follow me tool

TIG · 2 Apr 2012, 17:36

Scaling an Arc anchored at its center and through a 'mark' [as in my example] produces a new Arc with equal segments and vertices...
A line drawn through the center and both arc's vertices shows this.
There is no argument that this is a truly offset arc ???

Offsetting an Arc through the equivalent 'mark' extends the new arc's first/last segments so that if you were to draw a line across the two arcs' starts/ends it'd be square.
If you draw a longer line from the center to the original Arc's start/end it will cut the offset arc and if you then erase the extra bits off the start/end [and these temporary lines], then the resulting offset arc is exactly like the scaled version, with equal segments and vertex-spacing...
So it is possible to use Offset on an Arc and get a true arc result - albeit that you need to edit start/end segment to make it 'true'.

As I said before... If you wanted to offset an Arc to have 'square' ends it'd be more tricky.
Some simple editing of an offset=arc gives you both possibilities...

jeff hammond · 2 Apr 2012, 17:38

@lapx said:

I think sketchup has it backwards. Like I said kinda goofy if you ask me. The "one's and zero's need to be reordered .

it's not that sketchup necessarily has it backwards.. i mean, it's doing it perfectly accurate if you're offsetting straight lines (say a square for instance).. the problem is that it should be able to tell the difference between an arc and a collection of straight lines..

instead, it treats both situations the same way..

if i'm offsetting a square, the 4 sides should move the desired distance along the edges' perpendiculars… if i'm offsetting a 4 sidedcircle, the 4 vertices should move the desired distance along the edges' bisectors..

Wo3Dan · 2 Apr 2012, 22:14

@unknownuser said:

...ha Gerrit.. you're killing me here.. (but i suppose this is what TIG was talking about doing in his earlier example.. he had 'desired offset marker' in there so he was using that as an inference point)
so go ahead and inference it in 3D.. with the follow me tool

Sorry about that, Jeff, but I just can't see the inaccuracy. I do understand that you would like the offset tool on an arc to allow you to offset a certain distance to give you a new arc with predictable radius (=RADIUS-distance).
But offsetting with inferencing makes it possible. Even on an arc in 3D. That's all I wanted to say.

The point at center works on any arc.
Drawing a radius from center to arc's vertex also works in any situation. Now overrule the just drawn arc by inputting the new radius value. Offset works in the arc's plane and (apply the same vertex you drew the radius to!) inference the offsetted arc to the short radius's endpoint (opposite from the center on same edge).

As for the "Follow me" tool, it's very consistent, although the twisting can be a pain. But you can be sure that the cross section has the same dimensions on each segment troughout the path. Even with any more or less complicated 3D curve as a path.

This is not what I can say for the first example you presented a few "pages" back. The cross section is not the same.

@jgb,
Not only in engineering 0.5 degrees can be of great importance.
A collumn in a building that isn't plumb, but 0.5 degrees off blue, carrying a heavy load (9mm/m)?.... not that good.
But I only wanted to point out what caused the unwanted 'Follow Me' result.

ArCAD-UK · 3 Apr 2012, 08:32

I can see the argument from both sides for arc offsets because in the past I've wanted both results. Ideally SU will be given true arcs at some point. But from this thread I now understand why things are the way they are and on balance from an architectural perspective I prefer to have any offset from an arc create a consistent (wall) thickness which is what happens now rather than tapering the arc section to maintain a vertex offset.

pilou · 3 Apr 2012, 08:57

And about surface?
A disc of 1 m radius
12 segments is 3.00m2
24 segments is 3.11 m2
36 segments is 3.13 m2
100 segments is 3.14 m2

10 m radius
12 segments is 300 m2
24 segments is 310.58 m2
36 segments is 312.57 m2
100 segments is 313.95 m2

And I suppose that will be more dramatic with volume

So accurate for Sketch and drawing representation not for precises measures in case of Arc, Circle, Curves...

TIG · 3 Apr 2012, 09:34

To stop the issue... here's an Arc_Offet_True tool http://forums.sketchucation.com/viewtopic.php?p=395769#p395769
It has a few issues with Arcs with trimmed first/last segments BUT I'm working on that...
EDIT:
Version v1.1 now correctly mimics the original Arc's start/end segments if they were 'trimmed', it also stops you offsetting with a negative distance to 'zero' or beyond...
Version v1.2 now has a 'Number' option to make arrayed offsets, and a context-menu item too...

noelwarr · 3 Apr 2012, 10:36

Wow! Lots of posts. Think I've read through them all but I don't think the following has been pointed out... Sketchup is based on floating point data structure. These are inherently inaccurate. Try the following out in the ruby console

(Math::PI).to_l
=> 3.14159265358979
(Math::PI * 10**13).to_l
=> 31415926535897.9

This means that the further away from the origin your entities are, the more inaccurate they are.

Floats are a lightweight data structure ideal for a program like Sketchup. There are others that are actually capable of manipulating irrational numbers (pi, square root of two...check out GMP) without ever loosing accuracy but they are very heavy weight and only really necessary if you're, I don't know, sending a rocket to the moon or something.

Sketchup overcomes this floating point inaccuracy by allowing for a little tollerance but as you can see the difference is still there

point1.to_s + point2.to_s
=> (258,878708mm, 172,933835mm, 0mm)(0mm, 0mm, 0mm)
point1.z == point2.z
=> true
point1.to_a
=> [10.1920751212053, 6.80841870273468, 1.77635683940025e-015]
point2.to_a
=> [0.0, 0.0, 0.0]

Nonetheless Sketchup is a great tool and its inaccuracy (that is also present in those other "more professional" packages!) can be overlooked 99% of the time. Hope this was of use to anyone.

Trogluddite · 3 Apr 2012, 12:52

@noelwarr said:

Sketchup is based on floating point data structure. These are inherently inaccurate.

Quite so, I have to deal with this often in my "alter ego" as an audio DSP designer.

The fact that us humans use decimal, and PC's use binary also has some consequences for accuracy.
Just as decimal cannot exactly represent 1/3, binary cannot exactly represent 1/10 - the value to be represented and the number base have unshared prime factors.

A lot of software will display rounded values, so that they look a little more palatable to us humans - but it is a mistake to assume that the display value is exactly the same as the underlying float number. If your display is telling you that your dimension is exactly 0.1units, it is probably lying, as a float number simply cannot have exactly that value, only an approximation.

Never occured to me before, but in this binary age, maybe the traditional "powers of two" divisions of the inch make more sense than decimal notation - those power of two factors can always be represented precisely in binary (up to the limit of the bit-depth) - so 1/16" really is 1/16, whereas 0.1m is only pretending!

sonder · 3 Apr 2012, 17:28

Well, thank god for construction industry standards allowing actual built accuracy to be 1/16" in ten feet or greater for different trades!

genma saotome · 3 Apr 2012, 18:14

Regarding my original assertion about the inaccuracy of the follow me tool -- I now understand I had not set up my work to be "compatible" with the way follow me works. Had I added tangents to each end of my arc and extruded the face along tangent, arc, tangent I'd get a middle section that was exactly what I wanted.

I'm not sure how to judge the implications of that fact WRT accuracy but I do know it is less work for me to do it that way than I had to deal with before and so while I'm not 100% happy about it, it is a better method. To everyone who helped to explain this, my thanks!

mac1 · 3 Apr 2012, 18:26

@noelwarr said:

Wow! Lots of posts. Think I've read through them all but I don't think the following has been pointed out... Sketchup is based on floating point data structure. These are inherently inaccurate. Try the following out in the ruby console

(Math::PI).to_l
=> 3.14159265358979
(Math::PI * 10**13).to_l
=> 31415926535897.9

This means that the further away from the origin your entities are, the more inaccurate they are.

Floats are a lightweight data structure ideal for a program like Sketchup. There are others that are actually capable of manipulating irrational numbers (pi, square root of two...check out GMP) without ever loosing accuracy but they are very heavy weight and only really necessary if you're, I don't know, sending a rocket to the moon or something.

Sketchup overcomes this floating point inaccuracy by allowing for a little tollerance but as you can see the difference is still there

point1.to_s + point2.to_s
=> (258,878708mm, 172,933835mm, 0mm)(0mm, 0mm, 0mm)
point1.z == point2.z
=> true
point1.to_a
=> [10.1920751212053, 6.80841870273468, 1.77635683940025e-015]
point2.to_a
=> [0.0, 0.0, 0.0]

Nonetheless Sketchup is a great tool and its inaccuracy (that is also present in those other "more professional" packages!) can be overlooked 99% of the time. Hope this was of use to anyone.

See my 3-16 post above and has been addressed before. The reason some packages show more accuracy the spec allows for extened or extendable implimentations. The single float 32 is only good to 7.225 digits.

jeff hammond · 4 Apr 2012, 00:43

@unknownuser said:

Well, thank god for construction industry standards allowing actual built accuracy to be 1/16" in ten feet or greater for different trades!

that tolerance should be set aside for the builders though.. with framing, i generally give my crew an 1/8" before i start bching at them or start contemplating a redo..

but, if i do that in the drawing.. say, give myself an eight in the model, then i'm looking at 1/4" errors onsite.. or, my crew has to be 100% perfect according to the drawing (i.e.-basically impossible)

EDIT
well, i already do cut in to the tolerance a bit when handing out a cultist.. everything is rounded to the 16th" even if the drawing calls for x/32" etc.. but yeah, there's definitely some allowable play in there.. i just prefer the drawing or calculations to be as close to perfect as possible (mainly because i think it IS possible.. we are dealing with computers here ).. and as i said early in the thread, sketchup is highly accurate and 100% capable of fulfilling my needs.. the rest of my babbling about in this thread has be regarding one small situation which (i feel) is mistreated in sketchup.. and i can still draw those parts accurately in sketchup-- i just use a different approach than whats offered in the automated toolset (offset tool/followme tool)

thomthom · 4 Apr 2012, 09:28

@unknownuser said:

when handing out a cultist..

Rich O Brien · 4 Apr 2012, 09:32

You dare judge Brother Jeff?

pilou · 4 Apr 2012, 09:43

These 1/4", 1/8", 1/16", 1/32" are for me perfect obscurantist measures from occultism !

thomthom · 4 Apr 2012, 11:38

@unknownuser said:

These 1/4", 1/8", 1/16", 1/32" are for me perfect obscurantist measures from occultism !

TIG · 4 Apr 2012, 12:22

[off:2hxvcf0c]In the UK at least no one uses the arcane 'pounds, shillings and pence' any more!
The UK dragged itself into the 20th century when I was a lad...
12 pence = 1 shilling
20 shillings = 1 pound
There were even coins as ha'penny [½d] and farthing [¼d]
[somewhat illogically 'd'=pence ]
We had other coins worth 1d[penny],3d[joey],6d[tanner], 1s[bob],2s[2-bob/florin],2s/6d[half-crown] and banknotes for 10s[10-bob],£1[a quid],£5[a fiver]...
There were also olde coins for 5s [crown] and guinea [£1/1s]***
So something would be priced as "£1/10s/4½d"
Nows we have decimal pounds/pence and coins up to £2.
So it is now approx. £1.53.........

*** The 'guinea' was an interesting idea.
If you were doing a service like an auctioneer you priced/charged the buyer in guineas and paid the seller in pounds - thereby keeping the standard 5% commission [1s is 1/20=5% of £1].
Having a base-12 shilling [like the feet/inch system!] allows you to divide it up into 1/4,1/3,1/2,2/3,3/4 and 1/6ths [and with ½d you jumpp to base-24 so 1/8ths are possible - 1½d was 1/8th of a shilling; using ¼d even allowed 1/16ths !!], but not 1/10ths ! Of course we still use 12 hours, 60 minutes/seconds etc [and obscure 'degrees'] for that very reason...
[/off:2hxvcf0c]
I can still 'conjure' with the olde fractional feet-and-inches - but 'metric' IS so much easier...

jgb · 4 Apr 2012, 14:07

@thomthom said:

@unknownuser said:

when handing out a cultist..

Yeah, I saw that too. You posted faster than I could.

Handing out cultists.... what a concept.