2010/4/18 Jelle Feringa <jelleferi...@gmail.com> > > The tolerance you talk about is the 'absolute' computation tolerance > > (it's pure mathematics). > > True > > > I think it also has to be connected to the size of the geometry. For > > instance, a tolerance of 10-6m is a non sense when dealing with the > > assembly of aircraft wings. But this order of magnitude is relevant > > to the watch industry. I would then say that the relative tolerance > > (which is engineering) is the criterion leading to the computational > > tolerance that has to be used. Then this one has to be compared to > > the 'minimum tolerance' of the CAD kernel. > > I agree. However, you need to set some sane unit size too; for watch > industry tolerances of micrometers might be more in scope of > fabrication tolerances. Perhaps not a good idea to work with > tolerances that are many orders of magnitude from your unit size. ( in > architecture, a meter is a interesting unit size; a millimeter is an > interesting tolerence ( 1e-3 ), for a watch I can imagine a millimeter > to be a fair unit size, so a micrometer a fair tolerance. Hence I > using micrometers is perhaps not so absurd ). > > OCC has no idea of scale. The label ( meter, mm, micrometer, inch ) > applied to the unit size is what it gives meaning. So, this is why you > cannot set tolerance endlessly low. > Therefore 0.000001 is a bad tolerance, whether applied to any domain. >
Your argument for a reasonable value of 1e-3 is much clear now, and I agree with it. Let's say we can go down to 1e-4. Should I conclude this post with the following recommandation? Recommandation: 1e-4 <= practical tolerance <= 1e-3. Values out of this range should be used only for special use cases and argumented. Thomas
_______________________________________________ Pythonocc-users mailing list Pythonocc-users@gna.org https://mail.gna.org/listinfo/pythonocc-users
