I already got a perfect visual understanding in my mind, o= bjects part of the net on copper layer is grown one by one and checked for = intersection of objects not part of the net but I should read code also.

2015-12-25 0:0= 0 GMT+01:00 Peter Clifton (petercjclifton AT googlemail DOT com) [via geda-user AT delorie DOT com] <geda-user AT delorie DOT com>:

Btw - in case the Engli= sh wasn't clear,=C2=A0 when we say "bloated", we mean an obje= ct is grown.

The code might also be bloated.... it isn't the pretties= t corner of pcb's internals.

I'd suggest reading through the existing find.c code and= seeing how you get on understanding that.

Peter

On 24 Dec 2015 22:03, "Nicklas Karlsson (nicklas.kar= lsson17 AT gmail DOT com) [via geda-user AT delorie DOT com]" <geda-user AT delorie DOT com> wrote:
Clearanc= e for outer corners: A corner is formed at an intersection of two lines, if= you move them outwards and measure distance between new and old intersecti= on you would know the problem. It would be possible to add a circle segment= with zero radius at the intersection and now if lines are moved outwards r= adius is increased and clearance would be correct.

If lo= oking at a net. Their would probably be one default global clearance value = and if needed more clearance values from this net to other nets with larger= values.

I did not have time to look at existing c= ode yet, just got explanation how it works and a suggestion it is bloated b= ut I actually think a grown copy is a simple good method for clearance. Lif= etime of copy could be created then objected created or created then needed= . It could also cut a holes in polygons and should be simple to visualize f= or debug purposes.

Nicklas Karlsson

2015-12-24 20:16 GMT= +01:00 Britton Kerin (britton DOT kerin AT gmail DOT com) [via geda-user AT delorie DOT com] <= geda-user AT delori= e.com>:

On Thu, D= ec 24, 2015 at 4:51 AM, Nicklas Karlsson (nicklas DOT karlsson17 AT gmail DOT com) [via geda-user AT delorie DOT c= om] <geda-user AT delorie DOT com> wrote:
There have been some discussion about DRC= clearance calculation. As I understand it now clearance i calculated by gr= owing objects on copper layer and check if netlist changed. Someone suggest= ed calculate distance is better but to know there is no copper from other o= bject withing clearance distance is actually good enough. There is however = a need for distance then flooding copper within polygon areas.

Algorithm clearance:
1. A function to check if there a= re other copper objects within clearance distance from a net, currently imp= lemented may be good enough.
1. Assign a net name to all copper o= bjects/"drawing primitives", probably all drawing primitives on c= opper layer.
1. For each net.
2. For each clerance valu= e of this net.
3. For the nets this clearance value should be use= d.
3. Check if this net are within clearance distance of nets thi= s clearance value should be used for.

The first qu= estion is if current algorithm make drawing primitives of net which should = be checked larger and check if netlist changed is good enough? I expect lin= es and arcs would be no problem and I actually also expect text will also w= ork if there is a line width although text is usually

I think text (on copper) is not handled at al= l in DRC at the moment.
=C2=A0
less important but outer corners of square o= bjects? Or outer corners with any angle?

<= /div>
I'm not sure what you mean here.
=C2= =A0
With this alg= orithm net different clearance between different nets should work although = it must also be decided how the clearance values should be attached to the = nets. Even though current method to calculate distance may not be perfect i= t would not stop other parts from being

This is the hardest part.=C2=A0 My understanding is that at= the moment pcb has a few basic types, and a boatload of flags, only a some= of which apply to each type.=C2=A0 There's a lot of talk about adding = attributes but I'm not certain it's a great idea.=C2=A0 Attributes = are a form of duck-typing, and if you've been lucky enough to spend tim= e with JavaScript you know how that can go. =C2=A0gschem has all sorts of p= roblems with the wonky, barely-known interpretations of attribute sets.=C2= =A0 Classes and inheritance feel harder up front but you often end up with = a much more maintainable system.
=C2=A0
implemented while the distance chec= k algorithm may or may not be improved in the mean time.

Do anybody have any comments on the algorithm?

If you haven't already I'd take a loo= k through find.c to get a feeling for how the existing DRC stuff works.=C2= =A0 I think it would work without much change
to do other clearan= ces.=C2=A0 You probably just need to start at the high-clearance part of th= e net first.=C2=A0 Alternately, a filter could be added to only throw
=
a violation if one of the involved parts had a special clearance (I= 9;m a little hazy on the exact order in which DRC finds new connections, bu= t I think the start points happen first).=C2=A0

Br= itton

=C2=A0
<= div dir=3D"ltr">

Nicklas Karlsson

--001a113cf2f8cc4b5c0527b48c33--